The Inference Report

Mathematical problem solving remains a challenging test of reasoning for large language and multimodal models, yet existing benchmarks are limited in size, language coverage, and task diversity. We introduce MathNet, a high-quality, large-scale, multimodal, and multilingual dataset of Olympiad-level math problems together with a benchmark for evaluating mathematical reasoning in generative models and mathematical retrieval in embedding-based systems. MathNet spans 47 countries, 17 languages, and two decades of competitions, comprising 30,676 expert-authored problems with solutions across diverse domains. In addition to the core dataset, we construct a retrieval benchmark consisting of mathematically equivalent and structurally similar problem pairs curated by human experts. MathNet supports three tasks: (i) Problem Solving, (ii) Math-Aware Retrieval, and (iii) Retrieval-Augmented Problem Solving. Experimental results show that even state-of-the-art reasoning models (78.4% for Gemini-3.1-Pro and 69.3% for GPT-5) remain challenged, while embedding models struggle to retrieve equivalent problems. We further show that retrieval-augmented generation performance is highly sensitive to retrieval quality; for example, DeepSeek-V3.2-Speciale achieves gains of up to 12%, obtaining the highest scores on the benchmark. MathNet provides the largest high-quality Olympiad dataset together with the first benchmark for evaluating mathematical problem retrieval, and we publicly release both the dataset and benchmark at https://mathnet.mit.edu.

Modern sequence models are dominated by Transformers, where self-attention mixes information from the visible context in an input-dependent way. However, when retrieval is not sharp and attention remains diffuse over an effective support $S_{\mathrm{eff}}(t)$, the influence of any individual token is diluted, typically scaling as $O(1/S_{\mathrm{eff}}(t))$ and reaching $O(1/\ell)$ for old tokens in full-prefix settings. Structured state-space models process sequences recurrently through an explicit feedback path; selective variants such as Mamba make this feedback input-dependent, yet when freeze time cannot be sustained over long intervals, their long-range sensitivity decays exponentially with lag. Existing architectures therefore either retrieve from the past in a single read or propagate information through a single feedback chain. We introduce Sessa, a decoder that places attention inside a feedback path, enabling recurrent many-path aggregation within a layer. Under stated assumptions, Sessa admits regimes with a power-law memory tail in lag $\ell$ of order $O(\ell^{-β})$ for $0<β<1$, which is asymptotically slower than $1/\ell$; moreover, this rate is tight in an explicit diffuse uniform-routing setting where the influence is $Θ(\ell^{-β})$. Under the same conditions, only Sessa among the compared model classes realizes flexible selective retrieval, including non-decaying profiles. Empirically, under matched architectures and training budgets, Sessa achieves the strongest performance on our long-context benchmarks while remaining competitive with Transformer and Mamba style baselines on short-context language modeling.

Proximal Policy Optimization (PPO) has become the predominant algorithm for on-policy reinforcement learning due to its scalability and empirical robustness across domains. However, there is a significant disconnect between the underlying foundations of trust region methods and the heuristic clipped objective used in PPO. In this paper, we bridge this gap by introducing the Bounded Ratio Reinforcement Learning (BRRL) framework. We formulate a novel regularized and constrained policy optimization problem and derive its analytical optimal solution. We prove that this solution ensures monotonic performance improvement. To handle parameterized policy classes, we develop a policy optimization algorithm called Bounded Policy Optimization (BPO) that minimizes an advantage-weighted divergence between the policy and the analytic optimal solution from BRRL. We further establish a lower bound on the expected performance of the resulting policy in terms of the BPO loss function. Notably, our framework also provides a new theoretical lens to interpret the success of the PPO loss, and connects trust region policy optimization and the Cross-Entropy Method (CEM). We additionally extend BPO to Group-relative BPO (GBPO) for LLM fine-tuning. Empirical evaluations of BPO across MuJoCo, Atari, and complex IsaacLab environments (e.g., Humanoid locomotion), and of GBPO for LLM fine-tuning tasks, demonstrate that BPO and GBPO generally match or outperform PPO and GRPO in stability and final performance.

We present BLF (Bayesian Linguistic Forecaster), an agentic system for binary forecasting that achieves state-of-the-art performance on the ForecastBench benchmark. The system is built on three ideas. (1) A Bayesian linguistic belief state: a semi-structured representation combining numerical probability estimates with natural-language evidence summaries, updated by the LLM at each step of an iterative tool-use loop. This contrasts with the common approach of appending all retrieved evidence to an ever-growing context. (2) Hierarchical multi-trial aggregation: running $K$ independent trials and combining them using logit-space shrinkage with a data-dependent prior. (3) Hierarchical calibration: Platt scaling with a hierarchical prior, which avoids over-shrinking extreme predictions for sources with skewed base rates. On 400 backtesting questions from the ForecastBench leaderboard, BLF outperforms all the top public methods, including Cassi, GPT-5, Grok~4.20, and Foresight-32B. Ablation studies show that the structured belief state is as impactful as web search access, and that shrinkage aggregation and hierarchical calibration each provide significant additional gains. In addition, we develop a robust back-testing framework with a leakage rate below 1.5\%, and use rigorous statistical methodology to compare different methods while controlling for various sources of noise.

Large language models have achieved significant reasoning improvements through reinforcement learning with verifiable rewards (RLVR). Yet as model capabilities grow, constructing high-quality reward signals becomes increasingly difficult, making it essential to understand when RLVR can succeed under weaker forms of supervision. We conduct a systematic empirical study across diverse model families and reasoning domains under three weak supervision settings: scarce data, noisy rewards, and self-supervised proxy rewards. We find that generalization is governed by training reward saturation dynamics: models that generalize exhibit a prolonged pre-saturation phase during which training reward and downstream performance climb together, while models that saturate rapidly memorize rather than learn. We identify reasoning faithfulness, defined as the extent to which intermediate steps logically support the final answer, as the pre-RL property that predicts which regime a model falls into, while output diversity alone is uninformative. Motivated by these findings, we disentangle the contributions of continual pre-training and supervised fine-tuning, finding that SFT on explicit reasoning traces is necessary for generalization under weak supervision, while continual pre-training on domain data amplifies the effect. Applied together to Llama3.2-3B-Base, these interventions enable generalization across all three settings where the base model previously failed.

The Platonic Representation Hypothesis suggests that neural networks trained on different modalities (e.g., text and images) align and eventually converge toward the same representation of reality. If true, this has significant implications for whether modality choice matters at all. We show that the experimental evidence for this hypothesis is fragile and depends critically on the evaluation regime. Alignment is measured using mutual nearest neighbors on small datasets ($\approx$1K samples) and degrades substantially as the dataset is scaled to millions of samples. The alignment that remains between model representations reflects coarse semantic overlap rather than consistent fine-grained structure. Moreover, the evaluations in Huh et al. are done in a one-to-one image-caption setting, a constraint that breaks down in realistic many-to-many settings and further reduces alignment. We also find that the reported trend of stronger language models increasingly aligning with vision does not appear to hold for newer models. Overall, our findings suggest that the current evidence for cross-modal representational convergence is considerably weaker than subsequent works have taken it to be. Models trained on different modalities may learn equally rich representations of the world, just not the same one.

Modern medicine generates vast multimodal data across siloed systems, yet no existing model integrates the full breadth and temporal depth of the clinical record into a unified patient representation. We introduce Apollo, a multimodal temporal foundation model trained and evaluated on over three decades of longitudinal hospital records from a major US hospital system, composed of 25 billion records from 7.2 million patients, representing 28 distinct medical modalities and 12 major medical specialties. Apollo learns a unified representation space integrating over 100 thousand unique medical events in our clinical vocabulary as well as images and clinical text. This "atlas of medical concepts" forms a computational substrate for modeling entire patient care journeys comprised of sequences of structured and unstructured events, which are compressed by Apollo into virtual patient representations. To assess the potential of these whole-patient representations, we created 322 prognosis and retrieval tasks from a held-out test set of 1.4 million patients. We demonstrate the generalized clinical forecasting potential of Apollo embeddings, including predicting new disease onset risk up to five years in advance (95 tasks), disease progression (78 tasks), treatment response (59 tasks), risk of treatment-related adverse events (17 tasks), and hospital operations endpoints (12 tasks). Using feature attribution techniques, we show that model predictions align with clinically-interpretable multimodal biomarkers. We evaluate semantic similarity search on 61 retrieval tasks, and moreover demonstrate the potential of Apollo as a multimodal medical search engine using text and image queries. Together, these modeling capabilities establish the foundation for computable medicine, where the full context of patient care becomes accessible to computational reasoning.

In this work, we revisit the problem of active sequential prediction-powered mean estimation, where at each round one must decide the query probability of the ground-truth label upon observing the covariates of a sample. Furthermore, if the label is not queried, the prediction from a machine learning model is used instead. Prior work proposed an elegant scheme that determines the query probability by combining an uncertainty-based suggestion with a constant probability that encodes a soft constraint on the query probability. We explored different values of the mixing parameter and observed an intriguing empirical pattern: the smallest confidence width tends to occur when the weight on the constant probability is close to one, thereby reducing the influence of the uncertainty-based component. Motivated by this observation, we develop a non-asymptotic analysis of the estimator and establish a data-dependent bound on its confidence interval. Our analysis further suggests that when a no-regret learning approach is used to determine the query probability and control this bound, the query probability converges to the constraint of the max value of the query probability when it is chosen obliviously to the current covariates. We also conduct simulations that corroborate these theoretical findings.

Large language models frequently commit unrecoverable reasoning errors mid-generation: once a wrong step is taken, subsequent tokens compound the mistake rather than correct it. We introduce $\textbf{Latent Phase-Shift Rollback}$ (LPSR): at each generation step, we monitor the residual stream at a critical layer lcrit, detect abrupt directional reversals (phase shifts) via a cosine-similarity $+$ entropy dual gate, and respond by rolling back the KV-cache and injecting a pre-computed steering vector. No fine-tuning, gradient computation, or additional forward passes are required. LPSR achieves $\mathbf{44.0\%}$ on MATH-500 with an 8B model versus $28.8\%$ for standard AR ($+15.2$ pp; McNemar $χ^2 = 66.96$, $p < 10^{-15}$). Critically, prompted self-correction, the most natural inference-time baseline, scores only $19.8\%$, below standard AR; LPSR exceeds it by $+24.2$ pp ($χ^2 = 89.4$, $p \approx 0$). LPSR also outperforms Best-of-16 ($+7.8$ pp) at $5.4\times$ lower token cost, and surpasses a standard 70B model ($35.2\%$) with $8.75\times$ fewer parameters at ${\sim}3\times$ the token budget. A 32-layer sweep reveals a novel \textbf{detection-correction dissociation}: error-detection AUC peaks at layer~14 ($0.718$) but task accuracy peaks at layer~16 ($44.0\%$ vs.\ $29.2\%$), demonstrating that optimal monitoring depth differs for detection and correction.

We present a systematic evaluation of large language model families -- spanning both proprietary cloud APIs and locally-hosted open-source models -- on two purpose-built benchmarks for System Dynamics AI assistance: the \textbf{CLD Leaderboard} (53 tests, structured causal loop diagram extraction) and the \textbf{Discussion Leaderboard} (interactive model discussion, feedback explanation, and model building coaching). On CLD extraction, cloud models achieve 77--89\% overall pass rates; the best local model reaches 77\% (Kimi~K2.5~GGUF~Q3, zero-shot engine), matching mid-tier cloud performance. On Discussion, the best local models achieve 50--100\% on model building steps and 47--75\% on feedback explanation, but only 0--50\% on error fixing -- a category dominated by long-context prompts that expose memory limits in local deployments. A central contribution of this paper is a systematic analysis of \textit{model type effects} on performance: we compare reasoning vs.\ instruction-tuned architectures, GGUF (llama.cpp) vs.\ MLX (mlx\_lm) backends, and quantization levels (Q3 / Q4\_K\_M / MLX-3bit / MLX-4bit / MLX-6bit) across the same underlying model families. We find that backend choice has larger practical impact than quantization level: mlx\_lm does not enforce JSON schema constraints, requiring explicit prompt-level JSON instructions, while llama.cpp grammar-constrained sampling handles JSON reliably but causes indefinite generation on long-context prompts for dense models. We document the full parameter sweep ($t$, $p$, $k$) for all local models, cleaned timing data (stuck requests excluded), and a practitioner guide for running 671B--123B parameter models on Apple~Silicon.

A recent study (Kuribayashi et al., 2025) has shown that human sentence processing behavior, typically measured on syntactically unchallenging constructions, can be effectively modeled using surprisal from early layers of large language models (LLMs). This raises the question of whether such advantages of internal layers extend to more syntactically challenging constructions, where surprisal has been reported to underestimate human cognitive effort. In this paper, we begin by exploring internal layers that better estimate human cognitive effort observed in syntactic ambiguity processing in English. Our experiments show that, in contrast to naturalistic reading, later layers better estimate such a cognitive effort, but still underestimate the human data. This dual alignment sheds light on different modes of sentence processing in humans and LMs: naturalistic reading employs a somewhat weak prediction akin to earlier layers of LMs, while syntactically challenging processing requires more fully-contextualized representations, better modeled by later layers of LMs. Motivated by these findings, we also explore several probability-update measures using shallow and deep layers of LMs, showing a complementary advantage to single-layer's surprisal in reading time modeling.

Models from the AlphaFold (AF) family reliably predict one dominant conformation for most well-ordered proteins but struggle to capture biologically relevant alternate states. Several efforts have focused on eliciting greater conformational variability through ad hoc inference-time perturbations of AF models or their inputs. Despite their progress, these approaches remain inefficient and fail to consistently recover major conformational modes. Here, we investigate both the optimal location and manner-of-operation for perturbing latent representations in the AF3 architecture. We distill our findings in ConforNets: channel-wise affine transforms of the pre-Pairformer pair latents. Unlike previous methods, ConforNets globally modulate AF3 representations, making them reusable across proteins. On unsupervised generation of alternate states, ConforNets achieve state-of-the-art success rates on all existing multi-state benchmarks. On the novel supervised task of conformational transfer, ConforNets trained on one source protein can induce a conserved conformational change across a protein family. Collectively, these results introduce a mechanism for conformational control in AF3-based models.

Weight quantization has become a standard tool for efficient LLM deployment, especially for local inference, where models are now routinely served at 2-3 bits per parameter. The state of the art is currently split into two sets of methods: simple scalar quantization techniques, such as GPTQ or AWQ, which are widely deployed but plateau in accuracy at 3-4 bits per parameter (bpp), and "second-generation" vector- or trellis-quantized methods, such as QTIP, GPTVQ and AQLM, which push the accuracy frontier at low bit-widths but are notoriously hard to implement and to scale, and have gained relatively less traction. In this paper, we ask whether this gap is fundamental, or whether a carefully optimized scalar quantizer can recover most of it. We answer in the affirmative, by introducing GSQ (Gumbel-Softmax Quantization), a post-training scalar quantization method which jointly learns the per-coordinate grid assignments and the per-group scales using a Gumbel-Softmax relaxation of the discrete grid. GSQ matches the cardinality of the relaxation to the small number of levels available in the target bit-width regime (e.g., 3-8 levels for ternary and 3 bpp, respectively), making the relaxation tight and the optimization tractable. Practically, on the standard Llama-3.1-8B/70B-Instruct models, GSQ closes most of the gap between scalar quantization and the QTIP frontier at 2 and 3 bits, while using a symmetric scalar grid with group-wise quantization, and thus fully compatible with existing scalar inference kernels. We further show that GSQ scales to trillion-scale Mixture-of-Experts models such as Kimi-K2.5, where vector-quantized methods are difficult to apply.

This note clarifies the relationship between the recent TurboQuant work and the earlier DRIVE (NeurIPS 2021) and EDEN (ICML 2022) schemes. DRIVE is a 1-bit quantizer that EDEN extended to any $b>0$ bits per coordinate; we refer to them collectively as EDEN. First, TurboQuant$_{\text{mse}}$ is a special case of EDEN obtained by fixing EDEN's scalar scale parameter to $S=1$. EDEN supports both biased and unbiased quantization, each optimized by a different $S$ (chosen via methods described in the EDEN works). The fixed choice $S=1$ used by TurboQuant is generally suboptimal, although the optimal $S$ for biased EDEN converges to $1$ as the dimension grows; accordingly TurboQuant$_{\text{mse}}$ approaches EDEN's behavior for large $d$. Second, TurboQuant$_{\text{prod}}$ combines a biased $(b-1)$-bit EDEN step with an unbiased 1-bit QJL quantization of the residual. It is suboptimal in three ways: (1) its $(b-1)$-bit step uses the suboptimal $S=1$; (2) its 1-bit unbiased residual quantization has worse MSE than (unbiased) 1-bit EDEN; (3) chaining a biased $(b-1)$-bit step with a 1-bit unbiased residual step is inferior to unbiasedly quantizing the input directly with $b$-bit EDEN. Third, some of the analysis in the TurboQuant work mirrors that of the EDEN works: both exploit the connection between random rotations and the shifted Beta distribution, use the Lloyd-Max algorithm, and note that Randomized Hadamard Transforms can replace uniform random rotations. Experiments support these claims: biased EDEN (with optimized $S$) is more accurate than TurboQuant$_{\text{mse}}$, and unbiased EDEN is markedly more accurate than TurboQuant$_{\text{prod}}$, often by more than a bit (e.g., 2-bit EDEN beats 3-bit TurboQuant$_{\text{prod}}$). We also repeat all accuracy experiments from the TurboQuant paper, showing that EDEN outperforms it in every setup we have tried.

Privacy policies are intended to inform users about how software systems collect and handle data, yet they often remain vague or incomplete. This paper presents an empirical study of patterns in log-related statements within privacy policies and their alignment with privacy disclosures observed in Android application logs. We analyzed 1,000 Android apps across multiple categories, generating 86,836,964 log entries. Our findings reveal that while most applications (88.0%) provide privacy policies, only 28.5% explicitly mention logging practices. Among those that reference logging, most clearly describe what information is logged; however, 27.7% of log-related statements remain overly simplistic or vague, offering limited insight into actual data collection. We further observed widespread privacy leakages in application logs, with 67.6% of apps leaking sensitive information not mentioned in their policies. Alarmingly, only 4% of applications demonstrated consistent alignment between declared policy contents and actual logged data. These findings highlight that current privacy policies provide incomplete or ambiguous descriptions of logging practices, which frequently do not align with actual logging behaviors.

Physics-informed neural networks (PINNs) provide a powerful framework for learning governing equations of dynamical systems from data. Biologically-informed neural networks (BINNs) are a variant of PINNs that preserve the known differential operator structure (e.g., reaction-diffusion) while learning constitutive terms via trainable neural subnetworks, enforced through soft residual penalties. Existing BINN studies are limited to $1\mathrm{D}{+}t$ reaction-diffusion systems and focus on forward prediction, using the governing partial differential equation as a regulariser rather than an explicit identification target. Here, we extend BINNs to $2\mathrm{D}{+}t$ systems within a PINN framework that combines data preprocessing, BINN-based equation learning, and symbolic regression post-processing for closed-form equation discovery. We demonstrate the framework's real-world applicability by learning the governing equations of lung cancer cell population dynamics from time-lapse microscopy data, recovering $2\mathrm{D}{+}t$ reaction-diffusion models from experimental observations. The proposed framework is readily applicable to other spatio-temporal systems, providing a practical and interpretable tool for fast analytic equation discovery from data.

Verification of model outputs is rapidly emerging as a key primitive for both training and real-world deployment of large language models (LLMs). In practice, this often involves using imperfect LLM judges and reward models since ground truth acquisition can be time-consuming and expensive. We introduce Fully Unsupervised Score Ensembling (FUSE), a method for improving verification quality by ensembling verifiers without access to ground truth correctness labels. The key idea behind FUSE is to control conditional dependencies between verifiers in a manner that improves the unsupervised performance of a class of spectral algorithms from the ensembling literature. Despite requiring zero ground truth labels, FUSE typically matches or improves upon semi-supervised alternatives in test-time scaling experiments with diverse sets of generator models, verifiers, and benchmarks. In particular, we validate our method on both conventional academic benchmarks such as GPQA Diamond and on frontier, unsaturated benchmarks such as Humanity's Last Exam and IMO Shortlist questions.

We propose a distributionally robust approach to risk-sensitive estimation of an unknown signal x from an observed signal y. The unknown signal and observation are modeled as random vectors whose joint probability distribution is unknown, but assumed to belong to a given type-2 Wasserstein ball of distributions, termed the ambiguity set. The performance of an estimator is measured according to the conditional value-at-risk (CVaR) of the squared estimation error. Within this framework, we study the problem of computing affine estimators that minimize the worst-case CVaR over all distributions in the given ambiguity set. As our main result, we show that, when the nominal distribution at the center of the Wasserstein ball is finitely supported, such estimators can be exactly computed by solving a tractable semidefinite program. We evaluate the proposed estimators on a wholesale electricity price forecasting task using real market data and show that they deliver lower out-of-sample CVaR of squared error compared to existing methods.

Constructing environments for training and evaluating claw-like agents remains a manual, human-intensive process that does not scale. We argue that what is needed is not just a dataset, but an automated pipeline capable of generating diverse, verified environments on demand. To this end, we introduce ClawEnvKit, an autonomous generation pipeline that instantiates this formalism from natural language descriptions. The pipeline comprises three modules: (1) a parser that extracts structured generation parameters from natural language input; (2) a generator that produces the task specification, tool interface, and scoring configuration; and (3) a validator that enforces feasibility, diversity, structural validity, and internal consistency across the generated environments. Using ClawEnvKit, we construct Auto-ClawEval, the first large-scale benchmark for claw-like agents, comprising 1,040 environments across 24 categories. Empirically, Auto-ClawEval matches or exceeds human-curated environments on coherence and clarity at 13,800x lower cost. Evaluated across 4 model families and 8 agent harness frameworks, we find that harness engineering boosts performance by up to 15.7 percentage points over a bare ReAct baseline, completion remains the primary axis of variation with no model saturating the benchmark, and automated generation enables evaluation at a scale previously infeasible. Beyond static benchmarking, ClawEnvKit enables live evaluation: users describe a desired capability in natural language and obtain a verified environment on demand, turning evaluation into a continuous, user-driven process. The same mechanism serves as an on-demand training environment generator, producing task distributions that adapt to an agent's current weaknesses rather than being bounded by existing user logs.

Adversarial training of binary classifiers can be reformulated as regularized risk minimization involving a nonlocal total variation. Building on this perspective, we establish a characterization of the subdifferential of this total variation using duality techniques. To achieve this, we derive a dual representation of the nonlocal total variation and a related integration of parts formula, involving a nonlocal gradient and divergence. We provide such duality statements both in the space of continuous functions vanishing at infinity on proper metric spaces and for the space of essentially bounded functions on Euclidean domains. Furthermore, under some additional conditions we provide characterizations of the subdifferential in these settings.

This paper studies how empirical dialogue-flow statistics can be incorporated into Next Dialogue Act Prediction (NDAP). A KL regularization term is proposed that aligns predicted act distributions with corpus-derived transition patterns. Evaluated on a 60-class German counselling taxonomy using 5-fold cross-validation, this improves macro-F1 by 9--42% relative depending on encoder and substantially improves dialogue-flow alignment. Cross-dataset validation on HOPE suggests that improvements transfer across languages and counselling domains. In systematic ablations across pretrained encoders and architectures, the findings indicate that transition regularization provides consistent gains and disproportionately benefits weaker baseline models. The results suggest that lightweight discourse-flow priors complement pretrained encoders, especially in fine-grained, data-sparse dialogue tasks.

We study synthesis for obligation properties expressed in LTLfp, the extension of LTLf to infinite traces. Obligation properties are positive Boolean combinations of safety and guarantee (co-safety) properties and form the second level of the temporal hierarchy of Manna and Pnueli. Although obligation properties are expressed over infinite traces, they retain most of the simplicity of LTLf. In particular, we show that they admit a translation into symbolically represented deterministic weak automata (DWA) obtained directly from the symbolic deterministic finite automata (DFA) for the underlying LTLf properties on trace prefixes. DWA inherit many of the attractive algorithmic features of DFA, including Boolean closure and polynomial-time minimization. Moreover, we show that synthesis for LTLfp obligation properties is theoretically highly efficient - solvable in linear time once the DWA is constructed. We investigate several symbolic algorithms for solving DWA games that arise in the synthesis of obligation properties and evaluate their effectiveness experimentally. Overall, the results indicate that synthesis for LTLfp obligation properties can be performed with virtually the same effectiveness as LTLf synthesis.

Recent advancements in Reinforcement Learning with Verifiable Rewards (RLVR) have significantly improved Large Language Model (LLM) reasoning, yet models often struggle to explore novel trajectories beyond their initial latent space. While offline teacher guidance and entropy-driven strategies have been proposed to address this, they often lack deep integration or are constrained by the model's inherent capacity. In this paper, we propose OGER, a novel framework that unifies offline teacher guidance and online reinforcement learning through a specialized reward modeling lens. OGER employs multi-teacher collaborative training and constructs an auxiliary exploration reward that leverages both offline trajectories and the model's own entropy to incentivize autonomous exploration. Extensive experiments across mathematical and general reasoning benchmarks demonstrate that OGER significantly outperforms competitive baselines, achieving substantial gains in mathematical reasoning while maintaining robust generalization to out-of-domain tasks. We provide a comprehensive analysis of training dynamics and conduct detailed ablation studies to validate the effectiveness of our entropy-aware reward modulation. Our code is available at https://github.com/ecoli-hit/OGER.git.

Static code analysis (SCA) tools are widely used as effective ways to detect bugs and vulnerabilities in software systems. However, the reports generated by these tools often contain a large number of non-actionable findings, which can overwhelm developers to the point of ignoring them altogether -- this phenomenon is known as "alert fatigue". In this paper, we combat alert fatigue by proposing STAF: Sentence Transformer-based Actionability Filtering. Our approach leverages a transformer based architecture with sentence embeddings to classify findings into actionable and non-actionable categories. Evaluating STAF on a large dataset of reports from Java projects, we demonstrate that our method can effectively reduce the number of non-actionable findings while maintaining a high level of accuracy in identifying actionable issues. The results show that our approach can improve the usability of static analysis tools reaching an F1 score of 89%, outperforming existing methods for SCA warning filtering by at least 11% in a within-project setting and by at least 6% in a cross-project setting. By providing a more focused and relevant set of findings, we aim to enhance the overall effectiveness of static analysis in software development.

Epidemic forecasting has become an integral part of real-time infectious disease outbreak response. While collaborative ensembles composed of statistical and machine learning models have become the norm for real-time forecasting, standardized benchmark datasets for evaluating such methods are lacking. Further, there is limited understanding on performance of these methods for novel outbreaks with limited historical data. In this paper, we propose IDOBE, a curated collection of epidemiological time series focused on outbreak forecasting. IDOBE compiles from multiple data repositories spanning over a century of surveillance and across U.S. states and global locations. We perform derivative-based segmentation to generate over 10,000 outbreaks covering multiple outcomes such as cases and hospitalizations for 13 diseases. We consider a variety of information-theoretic and distributional measures to quantify the epidemiological diversity of the dataset. Finally, we perform multi-horizon short-term forecasting (1- to 4-week-ahead) through the progression of the outbreak using 11 baseline models and report on their performance. In addition to standard metrics such as NMSE and MAPE for point forecasts, we include probabilistic scoring rules such as Normalized Weighted Interval Score (NWIS) to quantify the performance. We find that MLP-based methods have the most robust performance, with statistical methods having a slight edge during the pre-peak phase. IDOBE dataset along with baselines are released publicly on https://github.com/NSSAC/IDOBE to enable standardized, reproducible benchmarking of outbreak forecasting methods.

Guard models are widely used to detect harmful content in user prompts and LLM responses. However, state-of-the-art guard models rely solely on terminal-layer representations and overlook the rich safety-relevant features distributed across internal layers. We present SIREN, a lightweight guard model that harnesses these internal features. By identifying safety neurons via linear probing and combining them through an adaptive layer-weighted strategy, SIREN builds a harmfulness detector from LLM internals without modifying the underlying model. Our comprehensive evaluation shows that SIREN substantially outperforms state-of-the-art open-source guard models across multiple benchmarks while using 250 times fewer trainable parameters. Moreover, SIREN exhibits superior generalization to unseen benchmarks, naturally enables real-time streaming detection, and significantly improves inference efficiency compared to generative guard models. Overall, our results highlight LLM internal states as a promising foundation for practical, high-performance harmfulness detection.

Uniform Discrete Diffusion Model (UDM) has recently emerged as a promising paradigm for discrete generative modeling; however, its integration with reinforcement learning remains largely unexplored. We observe that naively applying GRPO to UDM leads to training instability and marginal performance gains. To address this, we propose \Ours, the first framework to integrate UDM with RL. Our method is guided by two key insights: (i) treating the final clean sample as the action provides more accurate and stable optimization signals; and (ii) reconstructing trajectories via the diffusion forward process better aligns probability paths with the pretraining distribution. Additionally, we introduce two strategies, Reduced-Step and CFG-Free, to further improve training efficiency. \Ours significantly improves base model performance across multiple T2I tasks. Notably, GenEval accuracy improves from $69\%$ to $96\%$ and PickScore increases from $20.46$ to $23.81$, achieving state-of-the-art performance in both continuous and discrete settings. On the OCR benchmark, accuracy rises from $8\%$ to $57\%$, further validating the generalization ability of our method. Code is available at \href{https://github.com/Yovecent/UDM-GRPO}{https://github.com/Yovecent/UDM-GRPO}.

Open-weight language models can be rendered unsafe through several distinct interventions, but the resulting models may differ substantially in capabilities, behavioral profile, and internal failure mode. We study behavioral and mechanistic properties of jailbroken models across three unsafe routes: harmful supervised fine-tuning (SFT), harmful reinforcement learning with verifiable rewards (RLVR), and refusal-suppressing abliteration. All three routes achieve near-ceiling harmful compliance, but they diverge once we move beyond direct harmfulness. RLVR-jailbroken models show minimal degradation and preserve explicit harm recognition in a structured self-audit: they are able to identify harmful prompts and describe how a safe LLM should respond, yet they comply with the harmful request. With RLVR, harmful behavior is strongly suppressed by a reflective safety scaffold: when a harmful prompt is prepended with an instruction to reflect on safety standards, harmful behavior drops close to the baseline. Category-specific RLVR jailbreaks generalize broadly across harmfulness domains. Models jailbroken with SFT show the largest collapse in explicit safety judgments, the highest behavioral drift, and a substantial capability loss on standard benchmarks. Abliteration is family-dependent in both self-audit and response to a reflective safety scaffold. Mechanistic and repair analyses further separate the routes: abliteration is consistent with localized refusal-feature deletion, RLVR with preserved safety geometry but retargeted policy behavior, and SFT with broader distributed drift. Targeted repair partially recovers RLVR-jailbroken models, but has little effect on SFT-jailbroken models. Together, these results show that jailbreaks can produce vastly different properties despite similar harmfulness, with models jailbroken via RLVR showing remarkable similarity to the base model.

Large language models (LLMs) are widely used in retrieval-augmented generation (RAG) to incorporate external knowledge at inference time. However, when retrieved contexts are noisy, incomplete, or heterogeneous, a single generation process often struggles to reconcile evidence effectively. We propose \textbf{MASS-RAG}, a multi-agent synthesis approach to retrieval-augmented generation that structures evidence processing into multiple role-specialized agents. MASS-RAG applies distinct agents for evidence summarization, evidence extraction, and reasoning over retrieved documents, and combines their outputs through a dedicated synthesis stage to produce the final answer. This design exposes multiple intermediate evidence views, allowing the model to compare and integrate complementary information before answer generation. Experiments on four benchmarks show that MASS-RAG consistently improves performance over strong RAG baselines, particularly in settings where relevant evidence is distributed across retrieved contexts.

Many recent document embedding models are trained on document-as-image representations, embedding rendered pages as images rather than the underlying source. Meanwhile, existing benchmarks for scientific document retrieval, such as ArXivQA and ViDoRe, treat documents as images of pages, implicitly favoring such representations. In this work, we argue that this paradigm is not well-suited for text-rich multimodal scientific documents, where critical evidence is distributed across structured sources, including text, tables, and figures. To study this setting, we introduce ArXivDoc, a new benchmark constructed from the underlying LaTeX sources of scientific papers. Unlike PDF or image-based representations, LaTeX provides direct access to structured elements (e.g., sections, tables, figures, equations), enabling controlled query construction grounded in specific evidence types. We systematically compare text-only, image-based, and multimodal representations across both single-vector and multi-vector retrieval models. Our results show that: (1) document-as-image representations are consistently suboptimal, especially as document length increases; (2) text-based representations are most effective, even for figure-based queries, by leveraging captions and surrounding context; and (3) interleaved text+image representations outperform document-as-image approaches without requiring specialized training.

We propose a computational framework for replacing the repeated numerical solution of differential Riccati equations in finite-horizon Linear Quadratic Regulator (LQR) problems by a learned operator surrogate. Instead of solving a nonlinear matrix-valued differential equation for each new system instance, we construct offline an approximation of the associated solution operator mapping time-dependent system parameters to the Riccati trajectory. The resulting model enables fast online evaluation of approximate optimal feedbacks across a wide class of systems, thereby shifting the computational burden from repeated numerical integration to a one-time learning stage. From a theoretical perspective, we establish control-theoretic guarantees for this operator-based approximation. In particular, we derive bounds quantifying how operator approximation errors propagate to feedback performance, trajectory accuracy, and cost suboptimality, and we prove that exponential stability of the closed-loop system is preserved under sufficiently accurate operator approximation. These results provide a framework to assess the reliability of data-driven approximations in optimal control. On the computational side, we design tailored DeepONet architectures for matrix-valued, time-dependent problems and introduce a progressive learning strategy to address scalability with respect to the system dimension. Numerical experiments on both time-invariant and time-varying LQR problems demonstrate that the proposed approach achieves high accuracy and strong generalization across a wide range of system configurations, while delivering substantial computational speedups compared to classical solvers. The method offers an effective and scalable alternative for parametric and real-time optimal control applications.

Reinforcement Learning (RL) enhances LLM reasoning, yet a paradox emerges as models scale: strong base models saturate standard benchmarks (e.g., MATH), yielding correct but homogeneous solutions. In such environments, the lack of failure cases causes the advantage signal in group-relative algorithms (e.g., GRPO) to vanish, driving policies into mode collapse. To address this, we propose Constrained Uniform Top-K Sampling (CUTS), a parameter-free decoding strategy enforcing structure-preserving exploration. Unlike standard sampling that follows model biases, CUTS flattens the local optimization landscape by sampling uniformly from constrained high-confidence candidates. We integrate this into Mixed-CUTS, a training framework synergizing exploitative and exploratory rollouts to amplify intra-group advantage variance. Experiments on Qwen3 models demonstrate that our approach prevents policy degeneration and significantly boosts out-of-domain generalization. Notably, Mixed-CUTS improves Pass@1 accuracy on the challenging AIME25 benchmark by up to 15.1% over standard GRPO, validating that maintaining diversity within the semantic manifold is critical for rigorous reasoning.

This paper proposes a multi-step probabilistic forecasting framework using a single neural-network based model to generate simultaneous point and interval forecasts. Our approach ensures non-crossing prediction intervals (PIs) through a model structure design that strictly satisfy a target coverage probability (PICP) while maximizing sharpness. Unlike existing methods that rely on manual weight tuning for scalarized loss functions, we treat point and PI forecasting as a multi-objective optimization problem, utilizing multi-gradient descent to adaptively select optimal weights. Key innovations include a new PI loss function based on an extended log-barrier with an adaptive hyperparameter to guarantee the coverage, a hybrid architecture featuring a shared temporal model with horizon-specific submodels, and a training strategy. The proposed loss is scale-independent and universally applicable; combined with our training algorithm, the framework eliminates trial-and-error hyperparameter tuning for balancing multiple objectives. Validated by an intra-day solar irradiance forecasting application, results demonstrate that our proposed loss consistently outperforms those in current literature by achieving target coverage with the narrowest PI widths. Furthermore, when compared against LSTM encoder-decoder and Transformer architectures--including those augmented with Chronos foundation models--our method remains highly competitive and can be seamlessly adapted to any deep learning structure.

Computational Fluid Dynamics (CFD) is central to race-car aerodynamic development, yet its cost -- tens of thousands of core-hours per high-fidelity evaluation -- severely limits the design space exploration feasible within realistic budgets. AI-based surrogate models promise to alleviate this bottleneck, but progress has been constrained by the limited complexity of public datasets, which are dominated by smoothed passenger-car shapes that fail to exercise surrogates on the thin, complex, highly loaded components governing motorsport performance. This work presents three primary contributions. First, we introduce a high-fidelity RANS dataset built on a parametric LMP2-class CAD model and spanning six operating conditions (map points) covering straight-line and cornering regimes, generated and validated by aerodynamics experts at Dallara to preserve features relevant to industrial motorsport. Second, we present the Gauge-Invariant Spectral Transformer (GIST), a graph-based neural operator whose spectral embeddings encode mesh connectivity to enhance predictions on tightly packed, complex geometries. GIST guarantees discretization invariance and scales linearly with mesh size, achieving state-of-the-art accuracy on both public benchmarks and the proposed race-car dataset. Third, we demonstrate that GIST achieves a level of predictive accuracy suitable for early-stage aerodynamic design, providing a first validation of the concept of interactive design-space exploration -- where engineers query a surrogate in place of the CFD solver -- within industrial motorsport workflows.

Existing MT evaluation frameworks, including automatic metrics and human evaluation schemes such as Multidimensional Quality Metrics (MQM), are largely language-agnostic. However, they often fail to capture dialect- and culture-specific errors in diglossic languages (e.g., Arabic), where translation failures stem from mismatches in language variety, content coverage, and pragmatic appropriateness rather than surface form alone.We introduce LQM: Linguistically Motivated Multidimensional Quality Metrics for MT. LQM is a hierarchical error taxonomy for diagnosing MT errors through six linguistically grounded levels: sociolinguistics, pragmatics, semantics, morphosyntax, orthography, and graphetics (Figure 1). We construct a bidirectional parallel corpus of 3,850 sentences (550 per variety) spanning seven Arabic dialects (Egyptian, Emirati, Jordanian, Mauritanian, Moroccan, Palestinian, and Yemeni), derived from conversational, culturally rich content. We evaluate six LLMs in a zero-shot setting and conduct expert span-level human annotation using LQM, producing 6,113 labeled error spans across 3,495 unique erroneous sentences, along with severity-weighted quality scores. We complement this analysis with an automatic metric (spBLEU). Though validated here on Arabic, LQM is a language-agnostic framework designed to be easily applied to or adapted for other languages. LQM annotated errors data, prompts, and annotation guidelines are publicly available at https://github.com/UBC-NLP/LQM_MT.

Large Language Models (LLMs) show promise in lyric-to-melody generation, but models trained with Supervised Fine-Tuning (SFT) often produce musically implausible melodies with issues like poor rhythm and unsuitable vocal ranges, a phenomenon we term "constraint violation". To address this, we propose a novel alignment framework that instills musical knowledge without human annotation. We define rule-based musical constraints to automatically generate a preference dataset from an SFT model's outputs. The model is then aligned through a sequential process, first using Direct Preference Optimization (DPO) on paired preference data, followed by Kahneman-Tversky Optimization (KTO) on unpaired negative samples. Experimental results demonstrate that our aligned model substantially reduces rule violations and outperforms strong baselines in both objective and subjective evaluations, generating melodies with substantially improved musicality and coherence. An interactive demo with audio comparisons is available at https://arain233.github.io/AligningMelody-demo.

The Adversarial Humanities Benchmark (AHB) evaluates whether model safety refusals survive a shift away from familiar harmful prompt forms. Starting from harmful tasks drawn from MLCommons AILuminate, the benchmark rewrites the same objectives through humanities-style transformations while preserving intent. This extends literature on Adversarial Poetry and Adversarial Tales from single jailbreak operators to a broader benchmark family of stylistic obfuscation and goal concealment. In the benchmark results reported here, the original attacks record 3.84% attack success rate (ASR), while transformed methods range from 36.8% to 65.0%, yielding 55.75% overall ASR across 31 frontier models. Under a European Union AI Act Code-of-Practice-inspired systemic-risk lens, Chemical, biological, radiological and nuclear (CBRN) is the highest bucket. Taken together, this lack of stylistic robustness suggests that current safety techniques suffer from weak generalization: deep understanding of 'non-maleficence' remains a central unresolved problem in frontier model safety.

Chain-of-Thought (CoT) reasoning has become a powerful driver of trajectory prediction in VLA-based autonomous driving, yet its autoregressive nature imposes a latency cost that is prohibitive for real-time deployment. Latent CoT methods attempt to close this gap by compressing reasoning into continuous hidden states, but consistently fall short of their explicit counterparts. We suggest that this is due to purely linguistic latent representations compressing a symbolic abstraction of the world, rather than the causal dynamics that actually govern driving. Thus, we present OneVL (One-step latent reasoning and planning with Vision-Language explanations), a unified VLA and World Model framework that routes reasoning through compact latent tokens supervised by dual auxiliary decoders. Alongside a language decoder that reconstructs text CoT, we introduce a visual world model decoder that predicts future-frame tokens, forcing the latent space to internalize the causal dynamics of road geometry, agent motion, and environmental change. A three-stage training pipeline progressively aligns these latents with trajectory, language, and visual objectives, ensuring stable joint optimization. At inference, the auxiliary decoders are discarded and all latent tokens are prefilled in a single parallel pass, matching the speed of answer-only prediction. Across four benchmarks, OneVL becomes the first latent CoT method to surpass explicit CoT, delivering state-of-the-art accuracy at answer-only latency, and providing direct evidence that tighter compression, when guided in both language and world-model supervision, produces more generalizable representations than verbose token-by-token reasoning. Project Page: https://xiaomi-embodied-intelligence.github.io/OneVL

Safety filters have been shown to be effective tools to ensure the safety of control systems with unsafe nominal policies. To address scalability challenges in traditional synthesis methods, learning-based approaches have been proposed for designing safety filters for systems with high-dimensional state and control spaces. However, the inevitable errors in the decisions of these models raise concerns about their reliability and the safety guarantees they offer. This paper presents Adaptive Conformal Filtering (ACoFi), a method that combines learned Hamilton-Jacobi reachability-based safety filters with adaptive conformal inference. Under ACoFi, the filter dynamically adjusts its switching criteria based on the observed errors in its predictions of the safety of actions. The range of possible safety values of the nominal policy's output is used to quantify uncertainty in safety assessment. The filter switches from the nominal policy to the learned safe one when that range suggests it might be unsafe. We show that ACoFi guarantees that the rate of incorrectly quantifying uncertainty in the predicted safety of the nominal policy is asymptotically upper bounded by a user-defined parameter. This gives a soft safety guarantee rather than a hard safety guarantee. We evaluate ACoFi in a Dubins car simulation and a Safety Gymnasium environment, empirically demonstrating that it significantly outperforms the baseline method that uses a fixed switching threshold by achieving higher learned safety values and fewer safety violations, especially in out-of-distribution scenarios.

This paper is a step-by-step, self-contained guide to the complete training cycle of a Physics-Informed Neural Network (PINN) -- a topic that existing tutorials and guides typically delegate to automatic differentiation libraries without exposing the underlying algebra. Using a first-order initial value problem with a known analytical solution as a running example, we walk through every stage of the process: forward propagation of both the network output and its temporal derivative, evaluation of a composite loss function built from the ODE residual and the initial condition, backpropagation of gradients -- with particular attention to the product rule that arises in hidden layers -- and a gradient descent parameter update. Every calculation is presented with explicit, verifiable numerical values using a 1-3-3-1 multilayer perceptron with two hidden layers and 22 trainable parameters. From these concrete examples, we derive general recursive formulas -- expressed as sensitivity propagation relations -- that extend the gradient computation to networks of arbitrary depth, and we connect these formulas to the automatic differentiation engines used in practice. The trained network is then validated against the exact solution, achieving a relative $L^2$ error of $4.290 \times 10^{-4}$ using only the physics-informed loss, without any data from the true solution. A companion Jupyter/PyTorch notebook reproduces every manual calculation and the full training pipeline, providing mutual validation between hand-derived and machine-computed gradients.

Persistent memory is the bottleneck separating stateless chatbots from long-running agentic systems. Retrieval-augmented generation (RAG) over flat vector stores fragments facts into chunks, loses cross-session identity, and has no first-class notion of supersession or contradiction. Recent bitemporal knowledge-graph systems (Graphiti, Memento, Hydra DB) add typed edges and valid-time metadata, but the graph itself remains flat: no recursive composition, no content-addressed invariants on nodes, and edge types carry no behavior beyond a label. We present WorldDB, a memory engine built on three commitments: (i) every node is a world -- a container with its own interior subgraph, ontology scope, and composed embedding, recursive to arbitrary depth; (ii) nodes are content-addressed and immutable, so any edit produces a new hash at the node and every ancestor, giving a Merkle-style audit trail for free; (iii) edges are write-time programs -- each edge type ships on_insert/on_delete/on_query_rewrite handlers (supersession closes validity, contradicts preserves both sides, same_as stages a merge proposal), so no raw append path exists. On LongMemEval-s (500 questions, ~115k-token conversational stacks), WorldDB with Claude Opus 4.7 as answerer achieves 96.40% overall / 97.11% task-averaged accuracy, a +5.61pp improvement over the previously reported Hydra DB state-of-the-art (90.79%) and +11.20pp over Supermemory (85.20%), with perfect single-session-assistant recall and robust performance on temporal reasoning (96.24%), knowledge update (98.72%), and preference synthesis (96.67%). Ablations show that the engine's graph layer -- resolver-unified entities and typed refers_to edges -- contributes +7.0pp task-averaged independently of the underlying answerer.

Biological classification with interpretability remains a challenging task. For this, we introduce a novel encoding framework, Multi-Scale Reversible Chaos Game Representation (MS-RCGR), that transforms biological sequences into multi-resolution geometric representations with guaranteed reversibility. Unlike traditional sequence encoding methods, MS-RCGR employs rational arithmetic and hierarchical k-mer decomposition to generate scale-invariant features that preserve complete sequence information while enabling diverse analytical approaches. Our framework bridges three distinct paradigms for sequence analysis: (1) traditional machine learning using extracted geometric features, (2) computer vision models operating on CGR-generated images, and (3) hybrid approaches combining protein language model embeddings with CGR features. Through comprehensive experiments on synthetic DNA and protein datasets encompassing seven distinct sequence classes, we demonstrate that MS-RCGR features consistently enhance classification performance across all paradigms. Notably, our hybrid approach combining pre-trained language model embeddings (ESM2, ProtT5) with MS-RCGR features achieves superior performance compared to either method alone. The reversibility property of our encoding ensures no information loss during transformation, while multi-scale analysis captures patterns ranging from individual nucleotides to complex motif structures. Our results indicate that MS-RCGR provides a flexible, interpretable, and high-performing foundation for biological sequence analysis.

Extending a fully post-trained language model with new domain capabilities is fundamentally limited by monolithic training paradigms: retraining from scratch is expensive and scales poorly, while continued training often degrades existing capabilities. We present BAR (Branch-Adapt-Route), which trains independent domain experts, each through its own mid-training, supervised finetuning, and reinforcement learning pipeline, and composes them via a Mixture-of-Experts architecture with lightweight router training. Unlike retraining approaches that mix all domains and require full reprocessing for any update (with cost scaling quadratically), BAR enables updating individual experts independently with linear cost scaling and no degradation to existing domains. At the 7B scale, with experts for math, code, tool use, and safety, BAR achieves an overall score of 49.1 (averaged across 7 evaluation categories), matching or exceeding re-training baselines (47.8 without mid-training, 50.5 with). We further show that modular training provides a structural advantage: by isolating each domain, it avoids the catastrophic forgetting that occurs when late-stage RL degrades capabilities from earlier training stages, while significantly reducing the cost and complexity of updating or adding a domain. Together, these results suggest that decoupled, expert-based training is a scalable alternative to monolithic retraining for extending language models.

Discrete diffusion language models (dLLMs) have recently emerged as a promising alternative to traditional autoregressive approaches, offering the flexibility to generate tokens in arbitrary orders and the potential of parallel decoding. However, existing heuristic sampling strategies remain inefficient: they choose only a small part of tokens to sample at each step, leaving substantial room for improvement. In this work, we study the problem of token sampling order optimization and demonstrate its significant potential for acceleration. Specifically, we find that fully leveraging correct predictions at each step can reduce the number of sampling iterations by an order of magnitude without compromising accuracy. Based on this, we propose Neural Indicator Sampling (NI Sampling), a general sampling order optimization framework that utilize a neural indicator to decide which tokens should be sampled at each step. We further propose a novel trajectory-preserving objective to train the indicator. Experiments on LLaDA and Dream models across multiple benchmarks show that our method achieves up to 14.3$\times$ acceleration over full-step sampling with negligible performance drop, and consistently outperforms confidence threshold sampling in the accuracy-step trade-off. Code is available at https://github.com/imagination-research/NI-Sampling.

Baseline estimation is critical to Demand Response (DR) settlement in electricity markets, yet existing machine learning methods remain limited in predictive performance, while methodologies from causal inference and counterfactual prediction are still underutilized in this domain. We introduce a Generalized Synthetic Control Method that builds on the classical Synthetic Control Method (SCM) from econometrics. While SCM provides a powerful framework for counterfactual estimation, classical SCM remains a static estimator: it fits the treated unit as a combination of contemporaneous donor units and therefore ignores predictable temporal structure in the residual error. We develop a generalized SCM framework that transforms baseline estimation into a dynamic counterfactual prediction problem by augmenting the donor representation with exogenous features, lagged treated load, and selected lagged donor signals. This enriched representation allows the estimator to capture autoregressive dependence, delayed donor-response patterns, and error-correction effects beyond the scope of standard SCM. The framework further accommodates nonlinear predictors when linear weighting is inadequate, with the greatest benefit arising in limited-data settings. Experiments on the Ausgrid smart-meter dataset show consistent improvements over classical SCM and strong benchmark methods, with the dominant performance gains driven by dynamic augmentation.

Closed-loop simulation is a core component of autonomous vehicle (AV) development, enabling scalable testing, training, and safety validation before real-world deployment. Neural scene reconstruction converts driving logs into interactive 3D environments for simulation, but it does not produce complete 3D object assets required for agent manipulation and large-viewpoint novel-view synthesis. To address this challenge, we present Asset Harvester, an image-to-3D model and end-to-end pipeline that converts sparse, in-the-wild object observations from real driving logs into complete, simulation-ready assets. Rather than relying on a single model component, we developed a system-level design for real-world AV data that combines large-scale curation of object-centric training tuples, geometry-aware preprocessing across heterogeneous sensors, and a robust training recipe that couples sparse-view-conditioned multiview generation with 3D Gaussian lifting. Within this system, SparseViewDiT is explicitly designed to address limited-angle views and other real-world data challenges. Together with hybrid data curation, augmentation, and self-distillation, this system enables scalable conversion of sparse AV object observations into reusable 3D assets.

Motivation: Peptide-protein interactions (PepPIs) are central to cellular regulation and peptide therapeutics, but experimental characterization remains too slow for large-scale screening. Existing methods usually emphasize either interaction prediction or peptide generation, leaving candidate prioritization, residue-level interpretation, and target-conditioned expansion insufficiently integrated. Results: We present an integrated framework for early-stage peptide screening that combines a partner-aware prediction and localization model (ConGA-PepPI) with a target-conditioned generative model (TC-PepGen). ConGA-PepPI uses asymmetric encoding, bidirectional cross-attention, and progressive transfer from pair prediction to binding-site localization, while TC-PepGen preserves target information throughout autoregressive decoding via layerwise conditioning. In five-fold cross-validation, ConGA-PepPI achieved 0.839 accuracy and 0.921 AUROC, with binding-site AUPR values of 0.601 on the protein side and 0.950 on the peptide side, and remained competitive on external benchmarks. Under a controlled length-conditioned benchmark, 40.39% of TC-PepGen peptides exceeded native templates in AlphaFold 3 ipTM, and unconstrained generation retained evidence of target-conditioned signal.

Semantic Tube Prediction (STP) leverages representation geometric to regularize LLM hidden-state trajectories toward locally linear geodesics during fine-tuning, thereby greatly improving data efficiency. The original STP recipe samples random token sub-spans, which is compatible with the base large language model (LLM) training architecture. Inspired by STP, we are interested to investigate whether the sampling position can further enhance the semantic structure of multi-step reasoning, and hence affect its geometric impact. We applied STP at consecutive semantic reasoning step boundaries and achieved 168x more accurate multi-step latent prediction than frozen baselines on ProcessBench (3,400 samples), compared to only 4x for the random-token STP. Probing the latent manifold with a learned non-linear predictor reveals that STP-shaped trajectories are smooth curves, not straight lines: a 3-layer MLP reduces prediction error by a further 3-12x over linear extrapolation on step-boundary models. Removing the language modeling loss yields trajectories that are 2x more MLP-predictable than the combined loss, revealing a tradeoff between generation quality and geometric purity. Our results identify sampling position as the critical variable in geometric regularization and establish multi-step latent prediction MSE as a new evaluation metric for this class of methods.

Large language models are increasingly used as planners for robotic systems, yet how safely they plan remains an open question. To evaluate safe planning systematically, we introduce DESPITE, a benchmark of 12,279 tasks spanning physical and normative dangers with fully deterministic validation. Across 23 models, even near-perfect planning ability does not ensure safety: the best-planning model fails to produce a valid plan on only 0.4% of tasks but produces dangerous plans on 28.3%. Among 18 open-source models from 3B to 671B parameters, planning ability improves substantially with scale (0.4-99.3%) while safety awareness remains relatively flat (38-57%). We identify a multiplicative relationship between these two capacities, showing that larger models complete more tasks safely primarily through improved planning, not through better danger avoidance. Three proprietary reasoning models reach notably higher safety awareness (71-81%), while non-reasoning proprietary models and open-source reasoning models remain below 57%. As planning ability approaches saturation for frontier models, improving safety awareness becomes a central challenge for deploying language-model planners in robotic systems.

Multimodal affective computing aims to predict humans' sentiment, emotion, intention, and opinion using language, acoustic, and visual modalities. However, current models often learn spurious correlations that harm generalization under distribution shifts or noisy modalities. To address this, we propose a causal modality-invariant representation (CmIR) learning framework for robust multimodal learning. At its core, we introduce a theoretically grounded disentanglement method that separates each modality into `causal invariant representation' and `environment-specific spurious representation' from a causal inference perspective. CmIR ensures that the learned invariant representations retain stable predictive relationships with labels across different environments while preserving sufficient information from the raw inputs via invariance constraint, mutual information constraint, and reconstruction constraint. Experiments across multiple multimodal benchmarks demonstrate that CmIR achieves state-of-the-art performance. CmIR particularly excels on out-of-distribution data and noisy data, confirming its robustness and generalizability.

Visual agents operating in the wild must respond to queries precisely when sufficient evidence first appears in a video stream, a critical capability that is overlooked by conventional video LLMs evaluated in offline settings. The shift to an online, streaming paradigm introduces significant challenges: a lack of decision transparency, the difficulty of aligning response timing with visual evidence, and the need to maintain a global, causally consistent understanding under tight computational budgets. To address these issues, we propose a novel framework that decouples reasoning control from memory integration. We introduce \textbf{\model{}}, an instantiation of this framework with two core components. First, the \emph{Active Thinking Decision Maker (ATDM)} is a transparent reasoning controller that externalizes its decision process using observable progress ($\boldsymbolρ$) and confidence ($\boldsymbol{c}$) metrics. This allows it to precisely time its response $t_r$ to match the first-sufficient-evidence timestamp $t^\star$ while streaming its reasoning to the user. Second, the \emph{Hierarchical Progressive Semantic Integration (HPSI)} module acts as an efficient memory system. It employs a set of learnable, multi-level aggregation tokens that are propagated across clips to build a rich, global cognitive state without exceeding token budgets. %Our approach sets a new standard on key online video understanding benchmarks, achieving strong performance of \textbf{71.6\%} on StreamingBench and \textbf{46.9\%} on OVOBench, demonstrating a robust solution for evidence-aligned and transparent online video analysis. Extensive experiments demonstrate the effectiveness of ATDM and HPSI, e.g., Thinking-QwenVL improves the accuracy of the previous state-of-the-art from 67.63\% to 71.60\% on the StreamingBench benchmark.

Vision-language modeling is rapidly increasing in popularity with an ever expanding list of available models. In most cases, these vision-language models have parameters in the tens of billions, which is necessary for some needs, but in many cases smaller models are necessary (e.g., on edge devices or independent robotic platforms). Unfortunately, there is little research in producing light-weight models or in training them with small datasets. Inspired by the language learning progression and data sparsity in child development, in this paper, we address both of these goals in a systematic fashion. We show that two-tower encoder models are superior to one-tower encoders in low-resource settings for discriminative English tasks. We show also that incorporating traditional convolutional networks into the two-tower transformer architecture can help produce parameter efficient vision-language models. Finally, we show that the cross-modal fusion module of two-tower encoders can vary significantly in shape and size while producing the same results. In addition, we present ESsEN, a compact vision-language model that can be trained end-to-end with relatively few resources that performs as well on several tasks with only a fraction of the parameters compared to other models. The experimental results and the tools we present here make vision-language modeling more accessible to a wider variety of researchers.

Empirical studies of trained models often report a transient regime in which signal is detectable in a finite gradient descent time window before overfitting dominates. We provide an analytically tractable random-matrix model that reproduces this phenomenon for gradient flow in a linear teacher--student setting. In this framework, learning occurs when an isolated eigenvalue separates from a noisy bulk, before eventually disappearing in the overfitting regime. The key ingredient is anisotropy in the input covariance, which induces fast and slow directions in the learning dynamics. In a two-block covariance model, we derive the full time-dependent bulk spectrum of the symmetrized weight matrix through a $2\times 2$ Dyson equation, and we obtain an explicit outlier condition for a rank-one teacher via a rank-two determinant formula. This yields a transient Baik-Ben Arous-Péché (BBP) transition: depending on signal strength and covariance anisotropy, the teacher spike may never emerge, emerge and persist, or emerge only during an intermediate time interval before being reabsorbed into the bulk. We map the corresponding phase diagrams and validate the theory against finite-size simulations. Our results provide a minimal solvable mechanism for early stopping as a transient spectral effect driven by anisotropy and noise.

Performance, power, and area (PPA) optimization is a fundamental task in RTL design, requiring a precise understanding of circuit functionality and the relationship between circuit structures and PPA metrics. Recent studies attempt to automate this process using LLMs, but neither feedback-based nor knowledge-based methods are efficient enough, as they either design without any prior knowledge or rely heavily on human-summarized optimization rules. In this paper, we propose AutoPPA, a fully automated PPA optimization framework. The key idea is to automatically generate optimization rules that enhance the search for optimal solutions. To do this, AutoPPA employs an Explore-Evaluate-Induce ($E^2I$) workflow that contrasts and abstracts rules from diverse generated code pairs rather than manually defined prior knowledge, yielding better optimization patterns. To make the abstracted rules more generalizable, AutoPPA employs an adaptive multi-step search framework that adopts the most effective rules for a given circuit. Experiments show that AutoPPA outperforms both the manual optimization and the state-of-the-art methods SymRTLO and RTLRewriter.

Zero-shot vision-language models (VLMs) have shown promise for chest radiograph classification, but their performance is often limited by confounding label co-occurrence, long-tail class imbalance, and transfer instability under domain shift. We propose ProtoCLIP, a refinement strategy for CLIP-style VLMs that improves zero-shot discrimination through targeted data curation and distilled anchor alignment. Specifically, we construct pathology-focused training subsets with curated negative samples to reduce co-occurrence bias. We also introduce a representation-preserving distillation objective to stabilize adaptation while maintaining semantic structure and improving discrimination of clinically relevant co-occurring pathologies. Evaluated on an unseen dataset VinDr-CXR, ProtoCLIP improves AUC by 2-10 percentage points over a strong CLIP-based baseline across multiple findings. For pneumothorax specifically, ProtoCLIP achieves a state-of-the-art AUC of 0.94. These results demonstrate that anchor-guided refinement, coupled with curated supervision and controlled adaptation, can mitigate common zero-shot transfer failures in medical VLMs without requiring large-scale retraining.

We present a scalable, data-driven simulation framework for large-scale heating, ventilation, and air conditioning (HVAC) systems that couples physics-informed neural ordinary differential equations (PINODEs) with differential-algebraic equation (DAE) solvers. At the component level, we learn heat-exchanger dynamics using an implicit PINODE formulation that predicts conserved quantities (refrigerant mass $M_r$ and internal energy $E_\text{hx}$) as outputs, enabling physics-informed training via automatic differentiation of mass/energy balances. Stable long-horizon prediction is achieved through gradient-stabilized latent evolution with gated architectures and layer normalization. At the system level, we integrate learned components with DAE solvers (IDA and DASSL) that explicitly enforce junction constraints (pressure equilibrium and mass-flow consistency), and we use Bayesian optimization to tune solver parameters for accuracy--efficiency trade-offs. To reduce residual system-level bias, we introduce a lightweight corrector network trained on short trajectory segments. Across dual-compressor and scaled network studies, the proposed approach attains multi-fold speedups over high-fidelity simulation while keeping errors low (MAPE below a few percent) and scales to systems with up to 32 compressor--condenser pairs.

Change visual question answering (Change VQA) addresses the problem of answering natural-language questions about semantic changes between bi-temporal remote sensing (RS) images. Although vision-language models (VLMs) have recently been studied for temporal RS image understanding, Change VQA remains underexplored in the context of modern multimodal models. In this letter, we revisit the CDVQA benchmark using recent Qwen models under a unified low-rank adaptation (LoRA) setting. We compare Qwen3-VL, which follows a structured vision-language pipeline with multi-depth visual conditioning and a full-attention decoder, with Qwen3.5, a native multimodal model that combines a single-stage alignment with a hybrid decoder backbone. Experimental results on the official CDVQA test splits show that recent VLMs improve over earlier specialized baselines. They further show that performance does not scale monotonically with model size, and that native multimodal models are more effective than structured vision-language pipelines for this task. These findings indicate that tightly integrated multimodal backbones contribute more to performance than scale or explicit multi-depth visual conditioning for language-driven semantic change reasoning in RS imagery.

India's linguistic landscape, spanning 22 scheduled languages and hundreds of marginalized dialects, has driven rapid growth in NLP datasets, benchmarks, and pretrained models. However, no dedicated survey consolidates resources developed specifically for Indian languages. Existing reviews either focus on a few high-resource languages or subsume Indian languages within broader multilingual settings, limiting coverage of low-resource and culturally diverse varieties. To address this gap, we present the first unified survey of Indian NLP resources, covering 200+ datasets, 50+ benchmarks, and 100+ models, tools, and systems across text, speech, multimodal, and culturally grounded tasks. We organize resources by linguistic phenomena, domains, and modalities; analyze trends in annotation, evaluation, and model design; and identify persistent challenges such as data sparsity, uneven language coverage, script diversity, and limited cultural and domain generalization. This survey offers a consolidated foundation for equitable, culturally grounded, and scalable NLP research in the Indian linguistic ecosystem.

Smooth functions on graphs have wide applications in manifold and semi-supervised learning. In this paper, we study a bandit problem where the payoffs of arms are smooth on a graph. This framework is suitable for solving online learning problems that involve graphs, such as content-based recommendation. In this problem, each item we can recommend is a node and its expected rating is similar to its neighbors. The goal is to recommend items that have high expected ratings. We aim for the algorithms where the cumulative regret with respect to the optimal policy would not scale poorly with the number of nodes. In particular, we introduce the notion of an effective dimension, which is small in real-world graphs, and propose two algorithms for solving our problem that scale linearly and sublinearly in this dimension. Our experiments on real-world content recommendation problem show that a good estimator of user preferences for thousands of items can be learned from just tens of nodes evaluations.

Large language models (LLMs) using chain-of-thought reasoning often waste substantial compute by producing long, incorrect responses. Abstention can mitigate this by withholding outputs unlikely to be correct. While most abstention methods decide to withhold outputs before or after generation, dynamic mid-generation abstention considers early termination of unpromising reasoning traces at each token position. Prior work has explored empirical variants of this idea, but principled guidance for the abstention rule remains lacking. We present a formal analysis of dynamic abstention for LLMs, modeling abstention as an explicit action within a regularized reinforcement learning framework. An abstention reward parameter controls the trade-off between compute and information. We show that abstaining when the value function falls below this reward strictly outperforms natural baselines under general conditions. We further derive a principled and efficient method to approximate the value function. Empirical results on mathematical reasoning and toxicity avoidance tasks support our theory and demonstrate improved selective accuracy over existing methods.

Data-driven discovery of governing equations has advanced significantly in recent years; however, existing methods often struggle in multiscale systems where dynamically significant terms may have small coefficients. Therefore, we propose Balance-Guided SINDy (BG-SINDy) inspired by the principle of dominant balance, which reformulates $\ell_0$-constrained sparse regression as a term-level $\ell_{2,0}$-regularized problem and solves it using a progressive pruning strategy. Terms are ranked according to their relative contributions to the governing equation balance rather than their absolute coefficient magnitudes. Based on this criterion, BG-SINDy alternates between least-squares regression and elimination of negligible terms, thereby preserving dynamically significant terms even when their coefficients are small. Numerical experiments on the Korteweg--de Vries equation with a small dispersion coefficient, a modified Burgers equation with vanishing hyperviscosity, a modified Kuramoto--Sivashinsky equation with multiple small-coefficient terms, and a two-dimensional reaction--diffusion system demonstrate the validity of BG-SINDy in discovering small-coefficient terms. The proposed method thus provides an efficient approach for discovering governing equations that contain small-coefficient terms.

Graph-based code indexing can improve context retrieval for LLM-based code agents by preserving call chains and dependency relationships that keyword search and similarity retrieval often miss. ABCoder is an open-source framework that parses codebases into a function-level code index called UniAST, but its existing parsers combine lightweight AST parsers for syntactic analysis with language servers for semantic resolution, but because LSP-based resolution requires a JSON-RPC call for each symbol lookup, these per-symbol calls become a bottleneck on large TypeScript repositories. We present abcoder-ts-parser, a TypeScript parser built on the TypeScript Compiler API that works directly with the compiler's AST, semantic information, and module resolution logic. We evaluate the parser on three open-source TypeScript projects with up to 1.2 million lines of code and find that it produces reliable indexes significantly more efficiently than the existing architecture. For a live demonstration, watch: https://youtu.be/ryssr7ouvdE

We present Six Llamas, a comparative study examining whether large language models fine-tuned on distinct religious corpora encode systematically different patterns of ethical reasoning. Six variants of Meta-Llama-3.1-8B are constructed: one unmodified control and five LoRA-adapted models trained exclusively on the sacred and theological texts of Christianity, Islam, Judaism, Hinduism, or Buddhism. All six models are probed with an identical battery of 17 standardized ethical prompts spanning moral dilemmas, game-theoretic scenarios, public policy questions, and moral-psychological self-assessments. To assess robustness and reproducibility, we implement a multi-temperature sampling design spanning ten temperature settings. We compute response consistency metrics, pairwise inter-model agreement rates, temperature sensitivity coefficients across four prompt domains, and run-to-run stability analyses. Findings show that LoRA-adapted models produce ethical reasoning patterns that are (a) systematically differentiated from the base model, (b) consistent with the moral logics of their training traditions, (c) structured along interpretable dimensions in moral-philosophical space, (d) core ethical positions remain stable across temperature variations for high-consensus dilemmas. The Trolley Problem achieves 100% consistency across all models and temperatures, while (e) tradition-specific divergence intensifies at higher temperatures in morally contested domains, and (f) the base model exhibits the highest overall response consistency (mean 88.3%), suggesting LoRA adaptation introduces both tradition-specific signal and increased sampling sensitivity. The study offers a proof-of-concept for the condensate comparative method using differentially trained language models as instruments for cultural and ethical analysis and identifies specific criteria for falsification and planned extensions.

General agents have given rise to phenomenal applications such as OpenClaw and Claude Code. As these agent systems (a.k.a. Harnesses) strive for bolder goals, they demand increasingly stronger agentic capabilities from foundation Large Language Models (LLMs). Agentic Reinforcement Learning (RL) is emerging as a central post-training paradigm for empowering LLMs with these capabilities and is playing an increasingly pivotal role in agent training. Unlike single-turn token-level alignment or reasoning enhancement, as in RLHF and RLVR, Agentic RL targets multi-turn interactive settings, where the goal is to optimize core agentic capabilities such as decision making and tool use while addressing new challenges including delayed and sparse rewards, as well as long and variable context. As a result, the token-centric modeling and optimization paradigm inherited from traditional LLM RL is becoming increasingly inadequate for capturing real LLM agent behavior. In this paper, we present StepPO as a position on step-level Agentic RL. We argue that the conventional token-level Markov Decision Process (MDP) should be advanced to a step-level MDP formulation, and that the step, rather than the token, should be regarded as the proper action representation for LLM agents. We then propose step-level credit assignment as the natural optimization counterpart of this formulation, thereby aligning policy optimization and reward propagation with the granularity of agent decisions. Finally, we discuss the key systems designs required to realize step-level Agentic RL in practice and preliminary experiments provide initial evidence for the effectiveness of this perspective. We hope that the step-aligned, step-level paradigm embodied in StepPO offers the Agentic RL community a useful lens for understanding agent behavior and helps advance LLMs toward stronger general-agent capabilities.

Daily infrastructure management in preparation for disasters is critical for urban resilience. When bridges remain resilient against disaster-induced external forces, access to hospitals, shops, and residences via metapaths can be sustained, maintaining essential urban functions. However, prioritizing bridge maintenance under limited budgets requires quantifying the multi-dimensional roles that bridges play in disaster scenarios -- a challenge that existing single-indicator approaches fail to address. We focus on metapaths from national highways through bridges to buildings (hospitals, shops, residences), constructing a heterogeneous graph with road, bridge, and building layers. A Relation-centric Graph Convolutional Network Variational Autoencoder (R-GCN-VGAE) learns metapath-based feature representations, enabling classification of bridges into disaster-preparedness categories: Supply Chain (commercial logistics), Medical Access (emergency healthcare), and Residential Protection (preventing isolation). Using OSMnx and open data, we validate our methodology on three diverse cities in Ibaraki Prefecture, Japan: Mito (697 bridges), Chikusei (258 bridges), and Moriya (148 bridges), totaling 1,103 bridges. The heterogeneous graph construction from open data enables redefining bridge roles for disaster scenarios, supporting maintenance budget decision-making. We contributed that (1) Open-data methodology for constructing urban heterogeneous graphs. (2) Redefinition of bridge roles for disaster scenarios via metapath-based classification. (3) Establishment of maintenance budget decision support methodology. (4) k-NN tuning strategy validated across diverse city scales. (5) Empirical demonstration of UMAP superiority over t-SNE/PCA for multi-role bridge visualization.

Creativity has become a core competence in the era of LLMs and human-AI collaboration, underpinning innovation in real-world problem solving. Crucially, the systematic improvement of creativity necessitates scientifically valid assessment instruments. Psychometric research recognizes context-based assessment as an effective way to measure creative thinking. However, high-quality expert-designed contexts remain scarce. Existing LLM-based generators often struggle with insufficient assessment cues, weak narrative coherence, limited stylistic diversity, and poor support for creative thinking. To address these challenges, we propose AlphaContext, an evolutionary tree-based psychometric context generator for creativity assessment. First, the HyperTree Outline Planner formalizes expert-designed outlining as a rule-guided hypertree and performs top-down hierarchical planning. The MCTS-based Context Generator fills the outline via MCTS to balance global structure and local quality. Then, the Evolutionary Context Optimizer evolves contexts with MAP-Elites by repeatedly updating niche elites to jointly improve diversity and quality. Finally, the Assessment-Guided Evolution Refiner simulates virtual participants with diverse styles and recycles weak contexts for further evolution. Experiments show that AlphaContext yields an average improvement of 8% over competitive methods across 6 quality metrics.

Large Language Models (LLMs) have demonstrated exceptional performance across diverse domains but are increasingly constrained by high inference latency. Early Exit has emerged as a promising solution to accelerate inference by dynamically bypassing redundant layers. However, in decoder-only architectures, the efficiency of Early Exit is severely bottlenecked by the KV Cache Absence problem, where skipped layers fail to provide the necessary historical states for subsequent tokens. Existing solutions, such as recomputation or masking, either introduce significant latency overhead or incur severe precision loss, failing to bridge the gap between theoretical layer reduction and practical wall-clock speedup. In this paper, we propose River-LLM, a training-free framework that enables seamless token-level Early Exit. River-LLM introduces a lightweight KV-Shared Exit River that allows the backbone's missing KV cache to be naturally generated and preserved during the exit process, eliminating the need for costly recovery operations. Furthermore, we utilize state transition similarity within decoder blocks to predict cumulative KV errors and guide precise exit decisions. Extensive experiments on mathematical reasoning and code generation tasks demonstrate that River-LLM achieves 1.71 to 2.16 times of practical speedup while maintaining high generation quality.

Game development sits at the intersection of creative design and intricate software engineering, demanding the joint orchestration of game engines, real-time loops, and tightly coupled state across many files. While Large Language Models (LLMs) and code agents now solve isolated programming tasks with ease, they consistently stumble when asked to produce a fully playable game from a high-level design, collapsing under cross-file inconsistencies, broken scene wiring, and logical incoherence. We bridge this gap with OpenGame, the first open-source agentic framework explicitly designed for end-to-end web game creation. At its core lies Game Skill, a reusable, evolving capability composed of a Template Skill that grows a library of project skeletons from experience and a Debug Skill that maintains a living protocol of verified fixes - together enabling the agent to scaffold stable architectures and systematically repair integration errors rather than patch isolated syntax bugs. Powering this framework is GameCoder-27B, a code LLM specialized for game engine mastery through a three-stage pipeline of continual pre-training, supervised fine-tuning, and execution-grounded reinforcement learning. Since verifying interactive playability is fundamentally harder than checking static code, we further introduce OpenGame-Bench, an evaluation pipeline that scores agentic game generation along Build Health, Visual Usability, and Intent Alignment via headless browser execution and VLM judging. Across 150 diverse game prompts, OpenGame establishes a new state-of-the-art. We hope OpenGame pushes code agents beyond discrete software engineering problems and toward building complex, interactive real-world applications. Our framework will be fully open-sourced.

In self-supervised learning, self-distilled methods have shown impressive performance, learning representations useful for downstream tasks and even displaying emergent properties. However, state-of-the-art methods usually rely on ensembles of complex mechanisms, with many design choices that are empirically motivated and not well understood. In this work, we explore the role of self-distillation within learning dynamics. Specifically, we isolate the effect of self-distillation by training a group of randomly initialized networks, removing all other common components such as projectors, predictors, and even pretext tasks. Our findings show that even this minimal setup can lead to learned representations with non-trivial improvements over a random baseline on downstream tasks. We also demonstrate how this effect varies with different hyperparameters and present a short analysis of what is being learned by the models under this setup.

Prompt sensitivity, which refers to how strongly the output of a large language model (LLM) depends on the exact wording of its input prompt, raises concerns among users about the LLM's stability and reliability. In this work, we consider LLMs as multivariate functions and perform a first-order Taylor expansion, thereby analyzing the relationship between meaning-preserving prompts, their gradients, and the log probabilities of the model's next token. We derive an upper bound on the difference between log probabilities using the Cauchy-Schwarz inequality. We show that LLMs do not internally cluster similar inputs like smaller neural networks do, but instead disperse them. This dispersing behavior leads to an excessively high upper bound on the difference of log probabilities between two meaning-preserving prompts, making it difficult to effectively reduce to 0. In our analysis, we also show which types of meaning-preserving prompt variants are more likely to introduce prompt sensitivity risks in LLMs. In addition, we demonstrate that the upper bound is strongly correlated with an existing prompt sensitivity metric, PromptSensiScore. Moreover, by analyzing the logit variance, we find that prompt templates typically exert a greater influence on logits than the questions themselves. Overall, our results provide a general interpretation for why current LLMs can be highly sensitive to prompts with the same meaning, offering crucial evidence for understanding the prompt sensitivity of LLMs. Code for experiments is available at https://github.com/ku-nlp/Understanding_the_Prompt_Sensitivity.

Fine-tuning Large Language Models (LLMs) typically relies on large quantities of high-quality annotated data, or questions with well-defined ground truth answers in the case of Reinforcement Learning with Verifiable Rewards (RLVR). While previous work has explored the benefits to model reasoning capabilities by scaling both data and compute used for RLVR, these results lack applicability in many real-world settings where annotated data and accessible compute may be scarce. In this work, we present a comprehensive empirical study of open-source Small Language Model (SLM) performance after RLVR in low data regimes. Across three novel datasets covering number counting problems, graph reasoning, and spatial reasoning, we characterize how model performance scales with dataset size, diversity, and complexity. We demonstrate that (1) procedural datasets allow for fine-grained evaluation and training dataset development with controllable properties (size, diversity, and complexity), (2) under RLVR, models trained on lower complexity tasks can generalize to higher complexity tasks, and (3) training on mixed complexity datasets is associated with the greatest benefits in low data regimes, providing up to 5x sample efficiency versus training on easy tasks. These findings inspire future work on the development of data scaling laws for RLVR and the use of procedural data generators to further understand effective data development for efficient LLM fine-tuning.

Artificial intelligence (AI) technologies are increasingly used in modern weapons systems. Notably, these systems have recently been involved in mass killings and destruction at scale. Furthermore, there is currently a strong interest and competition among powerful players to accelerate the proliferation of weapons with automated or AI-based components, a phenomenon known as AI arms race. This competition poses a risk of causing even more deaths and devastation in the future, as well as increased power and wealth inequality. In this work, we aim to shed light on the role of AI researchers as implicated subjects in the harms caused by weapons enabled by AI technologies. We investigate and discuss the specifics of this implication and explore ways to transfigure this position of implication into one of differentiated, long-distance solidarity with the victims of technologically fortified injustices.

Most data-driven ionospheric forecasting models operate on gridded products, which do not preserve the time-varying sampling structure of satellite-based sensing. We instead model the ionosphere as a dynamic graph over ionospheric pierce points (IPPs), with connectivity that evolves as satellite positions change. Because satellite trajectories are predictable, the graph topology over the forecast horizon can be constructed in advance. We exploit this property to condition forecasts on the future graph structure, which we term ephemeris conditioning. This enables prediction on lines of sight that appear only in the forecast horizon. We evaluate our framework on multi-GNSS (Global Navigation Satellite System) data from a co-located receiver pair in Singapore spanning January 2023 through April 2025. The task is to forecast Rate of TEC Index (ROTI)-defined irregularities at 5-minute cadence up to 2 hours ahead as binary probabilistic classification per node. The resulting model, IonoDGNN, achieves a Brier Skill Score (BSS) of 0.49 and a precision-recall area under the curve (PR-AUC) of 0.75, improving over persistence by 35\% in BSS and 52\% in PR-AUC, with larger gains at longer lead times. Ablations confirm that graph structure and ephemeris conditioning each contribute meaningfully, with conditioning proving essential for satellites that rise during the forecast horizon (receiver operating characteristic AUC: 0.95 vs.\ 0.52 without). Under simulated coverage dropout, the model retains predictive skill on affected nodes through spatial message passing from observed neighbors. These results suggest that dynamic graph forecasting on evolving lines of sight is a viable alternative to grid-based representations for ionospheric irregularity forecasting. The model and evaluation code will be released upon publication.

Conversational agents, such as ChatGPT and Doubao, have become essential daily assistants for billions of users. To further enhance engagement, these systems are evolving from passive responders to proactive companions. However, existing efforts focus on activation within ongoing dialogues, while overlooking a key real-world bottleneck. In the conversation initiation stage, users may have a vague need but no explicit query intent, creating a first-message barrier where the conversation holds before it begins. To overcome this, we introduce Conversation Starter Generation: generating personalized starters to guide users into conversation. However, unlike in-conversation stages where immediate context guides the response, initiation must operate in a cold-start moment without explicit user intent. To pioneer in this direction, we present IceBreaker that frames human ice-breaking as a two-step handshake: (i) evoke resonance via Resonance-Aware Interest Distillation from session summaries to capture trigger interests, and (ii) stimulate interaction via Interaction-Oriented Starter Generation, optimized with personalized preference alignment and a self-reinforced loop to maximize engagement. Online A/B tests on one of the world's largest conversational agent products show that IceBreaker improves user active days by +0.184% and click-through rate by +9.425%, and has been deployed in production.

We study how AI agents form expectations and trade in experimental asset markets. Using a simulated open-call auction populated by autonomous Large Language Model (LLM) agents, we document three main findings. First, AI agents exhibit classic behavioral patterns: a pronounced disposition effect and recency-weighted extrapolative beliefs. Second, these individual-level patterns aggregate into equilibrium dynamics that replicate classic experimental findings (Smith et al., 1988), including the predictive power of excess demand for future prices and the positive relationship between disagreement and trading volume. Third, by analyzing the agents' reasoning text through a twenty-mechanism scoring framework, we show that targeted prompt interventions causally amplify or suppress specific behavioral mechanisms, significantly altering the magnitude of market bubbles.

Parkinson's disease (PD) is a chronic neurodegenerative disease. It shows multiple motor symptoms such as tremor, bradykinesia, postural instability, freezing of gait (FoG). PD is currently diagnosed clinically through physical exam by health-care professionals, which can be time consuming and highly subjective. Wearable IMU sensors has become a promising gateway for passive monitoring of PD patients. We propose a self-supervised cross-attention encoder that processes bilateral wrist-worn IMU signals from a public dataset called PADS, consisting of three groups, PD (Parkinson Disease), HC (Healthy Control) and DD (Differential Diagnosis) of a total of 469 subjects. We have achieved a mean accuracy of 93.12% for HC vs. PD classification and 87.04% for PD vs. DD classification. The results emphasize the clinical challenge of distinguishing Parkinson's from other neurodegenerative diseases. Self-supervised representation learning using contrastive infoNCE loss gained an accuracy of 93.56% for HC vs. PD and 92.50% for PD vs. DD using only 20% of labelled data. This demonstrates the effectiveness of our method in transfer learning for clinical use with minimal labels. The real-time applicability was tested by deploying the optimized model with a mean inference time of 48.32 ms per window on a Raspberry Pi CPU.

Generating programmatic animation using libraries such as Manim presents unique challenges for Large Language Models (LLMs), requiring spatial reasoning, temporal sequencing, and familiarity with domain-specific APIs that are underrepresented in general pre-training data. A systematic study of how training and inference strategies interact in this setting is lacking in current research. This study introduces ManimTrainer, a training pipeline that combines Supervised Fine-tuning (SFT) with Reinforcement Learning (RL) based Group Relative Policy Optimisation (GRPO) using a unified reward signal that fuses code and visual assessment signals, and ManimAgent, an inference pipeline featuring Renderer-in-the-loop (RITL) and API documentation-augmented RITL (RITL-DOC) strategies. Using these techniques, this study presents the first unified training and inference study for text-to-code-to-video transformation with Manim. It evaluates 17 open-source sub-30B LLMs across nine combinations of training and inference strategies using ManimBench. Results show that SFT generally improves code quality, while GRPO enhances visual outputs and increases the models' responsiveness to extrinsic signals during self-correction at inference time. The Qwen 3 Coder 30B model with GRPO and RITL-DOC achieved the highest overall performance, with a 94% Render Success Rate (RSR) and 85.7% Visual Similarity (VS) to reference videos, surpassing the baseline GPT-4.1 model by +3 percentage points in VS. Additionally, the analysis shows that the correlation between code and visual metrics strengthens with SFT and GRPO but weakens with inference-time enhancements, highlighting the complementary roles of training and agentic inference strategies in Manim animation generation.

Retrieval-augmented generation (RAG) remains unreliable in long-form settings, where retrieved evidence is noisy or contradictory, making it difficult for RAG pipelines to maintain factual consistency. Existing approaches focus on retrieval expansion or verification during generation, leaving conflict resolution entangled with generation. To address this limitation, we propose ArbGraph, a framework for pre-generation evidence arbitration in long-form RAG that explicitly resolves factual conflicts. ArbGraph decomposes retrieved documents into atomic claims and organizes them into a conflict-aware evidence graph with explicit support and contradiction relations. On top of this graph, we introduce an intensity-driven iterative arbitration mechanism that propagates credibility signals through evidence interactions, enabling the system to suppress unreliable and inconsistent claims before final generation. In this way, ArbGraph separates evidence validation from text generation and provides a coherent evidence foundation for downstream long-form generation. We evaluate ArbGraph on two widely used long-form RAG benchmarks, LongFact and RAGChecker, using multiple large language model backbones. Experimental results show that ArbGraph consistently improves factual recall and information density while reducing hallucinations and sensitivity to retrieval noise. Additional analyses show that these gains are evident under conflicting or ambiguous evidence, highlighting the effectiveness of evidence-level conflict resolution for improving the reliability of long-form RAG. The implementation is publicly available at https://github.com/1212Judy/ArbGraph.

Audio-text retrieval systems based on Contrastive Language-Audio Pretraining (CLAP) achieve strong performance on traditional benchmarks; however, these benchmarks rely on caption-style queries that differ substantially from real-world search behavior, limiting their assessment of practical retrieval robustness. We present Omni-Embed-Audio (OEA), a retrieval-oriented encoder leveraging multimodal LLMs with native audio understanding. To systematically evaluate robustness beyond caption-style queries, we introduce User-Intent Queries (UIQs) - five formulations reflecting natural search behaviors: questions, commands, keyword tags, paraphrases, and exclusion-based negative queries. For negative queries, we develop a hard negative mining pipeline and propose discrimination metrics (HNSR, TFR) assessing models' ability to suppress acoustically similar distractors. Experiments on AudioCaps, Clotho, and MECAT show that OEA achieves comparable text-to-audio retrieval performance to state-of-the-art M2D-CLAP, while demonstrating clear advantages in two critical areas: (1) dominant text-to-text retrieval (+22% relative improvement), and (2) substantially superior hard negative discrimination (+4.3%p HNSR@10, +34.7% relative TFR@10), revealing that LLM backbones provide superior semantic understanding of complex queries.

Developing compassionate interactive systems requires agents to not only understand user emotions but also provide diverse, substantive support. While recent works explore empathetic dialogue generation, they remain limited in response form and content, struggling to satisfy diverse needs across users and contexts. To address this, we explore empowering agents with external tools to execute diverse actions. Grounded in the psychological concept of "social support", this paradigm delivers substantive, human-like companionship. Specifically, we first design a dozen user-centric tools simulating various multimedia applications, which can cover different types of social support behaviors in human-agent interaction scenarios. We then construct ComPASS-Bench, the first personalized social support benchmark for LLM-based agents, via multi-step automated synthesis and manual refinement. Based on ComPASS-Bench, we further synthesize tool use records to fine-tune the Qwen3-8B model, yielding a task-specific ComPASS-Qwen. Comprehensive evaluations across two settings reveal that while the evaluated LLMs can generate valid tool-calling requests with high success rates, significant gaps remain in final response quality. Moreover, tool-augmented responses achieve better overall performance than directly producing conversational empathy. Notably, our trained ComPASS-Qwen demonstrates substantial improvements over its base model, achieving comparable performance to several large-scale models. Our code and data are available at https://github.com/hzp3517/ComPASS.

Emotion plays a pivotal role in shaping negotiation outcomes, influencing trust, cooperation, and long-term relationships. Developing negotiation dialog systems that can recognize and respond strategically to emotions is, therefore, essential to create more effective human-centered interactions. Beyond generating emotionally appropriate responses, interpretability - understanding how a system generates a particular emotion-aware response, is critical for fostering reliability and building rapport. Driven by these aspects, in this work, we introduce PRISMA, an interpretable emotionally intelligent negotiation dialogue system targeting two application domains, viz. job interviews and resource allocation. To enable interpretability, we propose an Emotion-aware Negotiation Strategy-informed Chain-of-Thought (ENS-CoT) reasoning mechanism, which mimics human negotiation by perceiving, understanding, using, and managing emotions. Leveraging ENS-CoT, we curate two new datasets: JobNego (for job interview negotiation) and ResNego (for resource allocation negotiation). We then leverage these datasets to develop PRISMA by augmenting self-training with Direct Preference Optimization (DPO), guiding agents toward more accurate, interpretable, and emotionally appropriate negotiation responses. Automatic and human evaluation on JobNego and ResNego datasets demonstrate that PRISMA substantially enhances interpretability and generates appropriate emotion-aware responses, while improving overall negotiation effectiveness.

State-of-the-art Differentially Private (DP) synthetic data generators such as MST and AIM are widely used, yet tightly auditing their privacy guarantees remains challenging. We introduce a Gaussian Differential Privacy (GDP)-based auditing framework that measures privacy via the full false-positive/false-negative tradeoff. Applied to MST and AIM under worst-case settings, our method provides the first tight audits in the strong-privacy regime. For $(ε,δ)=(1,10^{-2})$, we obtain $μ_{emp}\approx0.43$ vs. implied $μ=0.45$, showing a small theory-practice gap. Our code is publicly available: https://github.com/sassoftware/dpmm.

Recommender systems have advanced markedly over the past decade by transforming each user/item into a dense embedding vector with deep learning models. At industrial scale, embedding tables constituted by such vectors of all users/items demand a vast amount of parameters and impose heavy compute and memory overhead during training and inference, hindering model deployment under resource constraints. Existing solutions towards embedding compression either suffer from severely compromised recommendation accuracy or incur considerable computational costs. To mitigate these issues, this paper presents BACO, a fast and effective framework for compressing embedding tables. Unlike traditional ID hashing, BACO is built on the idea of exploiting collaborative signals in user-item interactions for user and item groupings, such that similar users/items share the same embeddings in the codebook. Specifically, we formulate a balanced co-clustering objective that maximizes intra-cluster connectivity while enforcing cluster-volume balance, and unify canonical graph clustering techniques into the framework through rigorous theoretical analyses. To produce effective groupings while averting codebook collapse, BACO instantiates this framework with a principled weighting scheme for users and items, an efficient label propagation solver, as well as secondary user clusters. Our extensive experiments comparing BACO against full models and 18 baselines over benchmark datasets demonstrate that BACO cuts embedding parameters by over 75% with a drop of at most 1.85% in recall, while surpassing the strongest baselines by being up to 346X faster.

Long-term conversational large language model (LLM) agents require memory systems that can recover relevant evidence from historical interactions without overwhelming the answer stage with irrelevant context. However, existing memory systems, including hierarchical ones, still often rely solely on vector similarity for retrieval. It tends to produce bloated evidence sets: adding many superficially similar dialogue turns yields little additional recall, but lowers retrieval precision, increases answer-stage context cost, and makes retrieved memories harder to inspect and manage. To address this, we propose HiGMem (Hierarchical and LLM-Guided Memory System), a two-level event-turn memory system that allows LLMs to use event summaries as semantic anchors to predict which related turns are worth reading. This allows the model to inspect high-level event summaries first and then focus on a smaller set of potentially useful turns, providing a concise and reliable evidence set through reasoning, while avoiding the retrieval overhead that would be excessively high compared to vector retrieval. On the LoCoMo10 benchmark, HiGMem achieves the best F1 on four of five question categories and improves adversarial F1 from 0.54 to 0.78 over A-Mem, while retrieving an order of magnitude fewer turns. Code is publicly available at https://github.com/ZeroLoss-Lab/HiGMem.

Video diffusion transformers (DiTs) suffer from prohibitive inference latency due to quadratic attention complexity. Existing sparse attention methods either overlook semantic similarity or fail to adapt to heterogeneous token distributions across layers, leading to model performance degradation. We propose AdaCluster, a training-free adaptive clustering framework that accelerates the generation of DiTs while preserving accuracy. AdaCluster applies an angle-similarity-preserving clustering method to query vectors for higher compression, and designs a euclidean-similarity-preserving clustering method for keys, covering cluster number assignment, threshold-wise adaptive clustering, and efficient critical cluster selection. Experiments on CogVideoX-2B, HunyuanVideo, and Wan-2.1 on one A40 GPU demonstrate up to 1.67-4.31x speedup with negligible quality degradation.

Vision Language Models (VLMs) achieved rapid progress in the recent years. However, despite their growth, VLMs development is heavily grounded on English, leading to two main limitations: (i) the lack of multilingual and multimodal datasets for training, and (ii) the scarcity of comprehensive evaluation benchmarks across languages. In this work, we address these gaps by introducing a new comprehensive suite of resources for VLMs training and evaluation spanning five European languages (English, French, German, Italian, and Spanish). We adopt a regeneration-translation paradigm that produces high-quality cross-lingual resources by combining curated synthetic generation and manual annotation. Specifically, we build Multi-PixMo, a training corpus obtained regenerating examples from Pixmo pre-existing datasets with permissively licensed models: PixMo-Cap, PixMo-AskModelAnything, and CoSyn-400k. On the evaluation side, we construct a set of multilingual benchmarks derived translating widely used English datasets (MMbench, ScienceQA, MME, POPE, AI2D). We assess the quality of these resources through qualitative and quantitative human analyses, measuring inter-annotator agreement. Additionally, we perform ablation studies to demonstrate the impact of multilingual data, with respect to English only, in VLMs training. Experiments, comprising 3 different models show that using multilingual, multimodal examples for training VLMs aids is consistently beneficial on non-English benchmarks, with positive transfer to English as well.

Knowledge Graphs (KGs) are composed of triples, and the goal of Knowledge Graph Completion (KGC) is to infer the missing factual triples. Traditional KGC tasks predict missing elements in a triple given one or two of its elements. As a more realistic task, the Triple Set Prediction (TSP) task aims to infer the set of missing triples conditioned only on the observed knowledge graph, without assuming any partial information about the missing triples. Existing TSP methods predict the set of missing triples in a triple-by-triple manner, falling short in capturing the dependencies among the predicted triples to ensure consistency. To address this issue, we propose a novel discrete diffusion model termed DiffTSP that treats TSP as a generative task. DiffTSP progressively adds noise to the KG through a discrete diffusion process, achieved by masking relational edges. The reverse process then gradually recovers the complete KG conditioned on the incomplete graph. To this end, we design a structure-aware denoising network that integrates a relational context encoder with a relational graph diffusion transformer for knowledge graph generation. DiffTSP can generate the complete set of triples in a one-pass manner while ensuring the dependencies among the predicted triples. Our approach achieves state-of-the-art performance on three public datasets. Code: https://github.com/ADMIS-TONGJI/DiffTSP.

Continuous Integration and Deployment (CI/CD) workflows are central to modern software delivery, yet the reliability of agentic AI bots operating within these workflows remain underexplored. Using pull requests (PRs), commits, and repositories from the AIDev dataset, we retrieved associated CI/CD workflow runs via the GitHub Actions API and analyzed 61,837 runs from 2,355 repositories, all triggered by PRs generated by five AI bots: Claude, Devin, Cursor, Copilot, and Codex. We observed substantial agent-dependent differences in workflow reliability, with Copilot and Codex achieving the highest success rates ~93% and ~94% respectively. At the repository level, we find a negative correlation between AI agent contribution frequency and workflow success rate, suggesting that a higher frequency of Agentic PRs may hinder CI/CD workflow reliability. We defined a taxonomy of 13 categories against 3,067 agentic PRs whose associated workflows failed, and observed a trend analysis that indicates visually observable shifts from functional to non-functional PR categories over time, although these trends are not statistically significant. Our findings motivate the need for actionable guidance on integrating AI agents into CI/CD workflows and prioritizing safeguards in workflows where failures are most likely to occur.

We present FregeLogic, a hybrid neuro-symbolic system for SemEval-2026 Task 11 (Subtask 1), which addresses syllogistic validity prediction while reducing content effects on predictions. Our approach combines an ensemble of five LLM classifiers, spanning three open-weights models (Llama 4 Maverick, Llama 4 Scout, and Qwen3-32B) paired with varied prompting strategies, with a Z3 SMT solver that serves as a formal logic tiebreaker. The central hypothesis is that LLM disagreement within the ensemble signals likely content-biased errors, where real-world believability interferes with logical judgment. By deferring to Z3's structurally-grounded formal verification on these disputed cases, our system achieves 94.3% accuracy with a content effect of 2.85 and a combined score of 41.88 in nested 5-fold cross-validation on the dataset (N=960). This represents a 2.76-point improvement in combined score over the pure ensemble (39.12), with a 0.9% accuracy gain, driven by a 16% reduction in content effect (3.39 to 2.85). Adopting structured-output API calls for Z3 extraction reduced failure rates from ~22% to near zero, and an Aristotelian encoding with existence axioms was validated against task annotations. Our results suggest that targeted neuro-symbolic integration, applying formal methods precisely where ensemble consensus is lowest, can improve the combined accuracy-plus-content-effect metric used by this task.

Reward models (RMs) are central to aligning large language models (LLMs) with human preferences, powering RLHF and advanced decoding strategies. While most prior work focuses on single-step generation, real-world applications increasingly adopt multi-stage LLM pipelines, where effective reward guidance remains underexplored. We investigate this through code generation for combinatorial optimization, constructing a pipeline that integrates reward models into both formulation and solution stages. We identify a critical challenge: inconsistency between reward model predictions and actual pipeline execution outcomes. To address this, we propose the Pipeline-Adapted Reward Model (PARM), which leverages pipeline-specific data and direct preference optimization to align rewards with downstream feedback. We instantiate PARM as a two-stage pipeline (formulation -> code generation) and evaluate it on four public optimization benchmarks, measuring execution rate and solving accuracy against baselines and sampling methods. A supplementary cross-domain experiment on GSM8K assesses transferability. Results demonstrate that PARM consistently improves pipeline output quality and stability, providing new insights into reward modeling for multi-stage LLM reasoning.

Self-evolution of multimodal large language models (MLLMs) remains a critical challenge: pseudo-label-based methods suffer from progressive quality degradation as model predictions drift, while template-based methods are confined to a static set of transformations that cannot adapt in difficulty or diversity. We contend that robust, continuous self-improvement requires not only deterministic external feedback independent of the model's internal certainty, but also a mechanism to perpetually diversify the training distribution. To this end, we introduce EVE (Executable Visual transformation-based self-Evolution), a novel framework that entirely bypasses pseudo-labels by harnessing executable visual transformations continuously enriched in both variety and complexity. EVE adopts a Challenger-Solver dual-policy architecture. The Challenger maintains and progressively expands a queue of visual transformation code examples, from which it synthesizes novel Python scripts to perform dynamic visual transformations. Executing these scripts yields VQA problems with absolute, execution-verified ground-truth answers, eliminating any reliance on model-generated supervision. A multi-dimensional reward system integrating semantic diversity and dynamic difficulty calibration steers the Challenger to enrich its code example queue while posing progressively more challenging tasks, preventing mode collapse and fostering reciprocal co-evolution between the two policies. Extensive experiments demonstrate that EVE consistently surpasses existing self-evolution methods, establishing a robust and scalable paradigm for verifiable MLLM self-evolution. The code is available at https://github.com/0001Henry/EVE .

Selection bias arises when the probability that an observation enters a dataset depends on variables related to the quantities of interest, leading to systematic distortions in estimation and uncertainty quantification. For example, in epidemiological or survey settings, individuals with certain outcomes may be more likely to be included, resulting in biased prevalence estimates with potentially substantial downstream impact. Classical corrections, such as inverse-probability weighting or explicit likelihood-based models of the selection process, rely on tractable likelihoods, which limits their applicability in complex stochastic models with latent dynamics or high-dimensional structure. Simulation-based inference enables Bayesian analysis without tractable likelihoods but typically assumes missingness at random and thus fails when selection depends on unobserved outcomes or covariates. Here, we develop a bias-aware simulation-based inference framework that explicitly incorporates selection into neural posterior estimation. By embedding the selection mechanism directly into the generative simulator, the approach enables amortized Bayesian inference without requiring tractable likelihoods. This recasting of selection bias as part of the simulation process allows us to both obtain debiased estimates and explicitly test for the presence of bias. The framework integrates diagnostics to detect discrepancies between simulated and observed data and to assess posterior calibration. The method recovers well-calibrated posterior distributions across three statistical applications with diverse selection mechanisms, including settings in which likelihood-based approaches yield biased estimates. These results recast the correction of selection bias as a simulation problem and establish simulation-based inference as a practical and testable strategy for parameter estimation under selection bias.

The most common cause of dementia is Alzheimer disease, a progressive neurodegenerative disorder affecting older adults that gradually impairs memory, cognition, and behavior. It is characterized by the accumulation of abnormal proteins in the brain, including amyloid-beta plaques and neurofibrillary tangles of tau protein, which disrupt neuronal communication and lead to neuronal death. Early manifestations typically include mild memory impairment and reduced ability to acquire new information. As the disease progresses, patients experience severe cognitive decline, loss of independence, and significant personality and behavioral changes. Although the exact etiology of Alzheimer disease remains unclear, factors such as age, genetic predisposition, lifestyle, and cardiovascular health contribute to its development. While no definitive cure exists, early diagnosis, pharmacological interventions, and supportive care can slow progression and improve quality of life. This study presents a predictive cheminformatics-based model for identifying natural medicinal compounds with potential therapeutic efficacy against Alzheimer disease. The model functions as a drug screening system utilizing molecular descriptors and machine learning to detect anti-Alzheimer activity. More than 7,000 compounds from ChEBI, SynSysNet, and INDOFINE were preprocessed using Open Babel and analyzed with Dragon descriptors. A Random Forest classifier trained on approved treatments achieved moderate performance, with precision of 0.5970 and recall of 0.6590, identifying 73 candidate compounds. Key descriptors included atomic polarizability, bond multiplicity, and non-hydrogen bond counts.These findings demonstrate the value of cheminformatics in early-stage drug discovery for Alzheimer disease.

We address the problem of planning in an environment with deterministic dynamics and stochastic rewards with discounted returns. The optimal value function is not known, nor are the rewards bounded. We propose Platypoos, a simple scale-free planning algorithm that adapts to the unknown scale and smoothness of the reward function. We provide a sample complexity analysis for Platypoos that improves upon prior work and holds simultaneously over a broad range of discount factors and reward scales, without the algorithm knowing them. We also establish a matching lower bound showing our analysis is optimal up to constants.

Explainable AI (XAI) aims to make the behaviour of machine learning models interpretable, yet many explanation methods remain difficult to understand. The integration of Natural Language Generation into XAI aims to deliver explanations in textual form, making them more accessible to practitioners. Current approaches, however, largely yield static lists of feature importances. Although such explanations indicate what influences the prediction, they do not explain why the prediction occurs. In this study, we draw on insights from social sciences and linguistics, and argue that XAI explanations should be presented in the form of narratives. Narrative explanations support human understanding through four defining properties: continuous structure, cause-effect mechanisms, linguistic fluency, and lexical diversity. We show that standard Natural Language Processing (NLP) metrics based solely on token probability or word frequency fail to capture these properties and can be matched or exceeded by tautological text that conveys no explanatory content. To address this issue, we propose seven automatic metrics that quantify the narrative quality of explanations along the four identified dimensions. We benchmark current state-of-the-art explanation generation methods on six datasets and show that the proposed metrics separate descriptive from narrative explanations more reliably than standard NLP metrics. Finally, to further advance the field, we propose a set of problem-agnostic XAI Narrative generation rules for producing natural language XAI explanations, so that the resulting XAI Narratives exhibit stronger narrative properties and align with the findings from the linguistic and social science literature.

Variational inference (VI) is a central tool in modern machine learning, used to approximate an intractable target density by optimising over a tractable family of distributions. As the variational family cannot typically represent the target exactly, guarantees on the quality of the resulting approximation are crucial for understanding which of its properties VI can faithfully capture. Recent work has identified instances in which symmetries of the target and the variational family enable the recovery of certain statistics, even under model misspecification. However, these guarantees are inherently problem-specific and offer little insight into the fundamental mechanism by which symmetry forces statistic recovery. In this paper, we overcome this limitation by developing a general theory of symmetry-induced statistic recovery in variational inference. First, we characterise when variational minimisers inherit the symmetries of the target and establish conditions under which these pin down identifiable statistics. Second, we unify existing results by showing that previously known statistic recovery guarantees in location-scale families arise as special cases of our theory. Third, we apply our framework to distributions on the sphere to obtain novel guarantees for directional statistics in von Mises-Fisher families. Together, these results provide a modular blueprint for deriving new recovery guarantees for VI in a broad range of symmetry settings.

Large language model (LLM)-based debugging systems can generate failure explanations, but these explanations may be incomplete or incorrect. Misleading explanations are harmful for downstream tasks (e.g., bug triage, bug fixing). We investigate how explanation quality is affected by various LLM context configurations. Existing work predominantly treats LLM-generated failure explanations as an ad hoc by-product of debugging or repair workflows, using generic prompting over undifferentiated artifacts such as code, tests, and error messages rather than targeting explanations as a first-class output with dedicated quality assessment. Consequently, existing approaches provide limited support for assessing whether these explanations capture the underlying fault-error-failure mechanism and for actionable next steps, and most techniques instead prioritize task success (e.g., patch correctness or review quality) over the explicit causal explanation quality. We systematically vary the debugging information to study how distinct context compositions affect the quality of LLM-generated failure explanations. Across 93 context configurations on real bugs and three economically viable models (gpt-5-mini, DeepSeek-V3.2, and Grok-4.1-fast), we evaluate explanations with six criteria and validate the LLM-as-a-judge scores against human ratings in a user study. Our results indicate that explanation quality is causally affected by context composition. Evidence-rich, failure-specific artifacts improve causal and action-oriented quality, whereas overly large contexts tend to yield vague explanations. Higher explanation-score quartiles are associated with higher downstream repair pass rates and, for some models, with fixes that are closer to the reference minimal fixes. In contrast, low-score quartiles can even underperform the no-explanation baseline. Reproduction package is publicly available.

Language models often solve complex tasks by generating long reasoning chains, consisting of many steps with varying importance. While some steps are crucial for generating the final answer, others are removable. Determining which steps matter most, and why, remains an open question central to understanding how models process reasoning. We investigate if this question is best approached through model internals or through tokens of the reasoning chain itself. We find that model activations contain more information than tokens for identifying important reasoning steps. Crucially, by training probes on model activations to predict importance, we show that models encode an internal representation of step importance, even prior to the generation of subsequent steps. This internal representation of importance generalizes across models, is distributed across layers, and does not correlate with surface-level features, such as a step's relative position or its length. Our findings suggest that analyzing activations can reveal aspects of reasoning that surface-level approaches fundamentally miss, indicating that reasoning analyses should look into model internals.

In this paper we investigate forecasting coevolving time series that feature intricate dependencies and nonstationary dynamics by using an LLM Large Language Models approach We propose a novel modeling approach named ContextAware ARLLM CAARL that provides an interpretable framework to decode the contextual dynamics influencing changes in coevolving series CAARL decomposes time series into autoregressive segments constructs a temporal dependency graph and serializes this graph into a narrative to allow processing by LLM This design yields a chainofthoughtlike reasoning path where intermediate steps capture contextual dynamics and guide forecasts in a transparent manner By linking prediction to explicit reasoning traces CAARL enhances interpretability while maintaining accuracy Experiments on realworld datasets validate its effectiveness positioning CAARL as a competitive and interpretable alternative to stateoftheart forecasting methods

Privacy represents one of the most critical yet underaddressed barriers to AI adoption in mental healthcare -- particularly in high-sensitivity operational environments such as military, correctional, and remote healthcare settings, where the risk of patient data exposure can deter help-seeking behavior entirely. Existing AI-enabled psychiatric decision support systems predominantly rely on cloud-based inference pipelines, requiring sensitive patient data to leave the device and traverse external servers, creating unacceptable privacy and security risks in these contexts. In this paper, we propose a zero-egress, on-device AI platform for privacy-preserving psychiatric decision support, deployed as a cross-platform mobile application. The proposed system extends our prior work on fine-tuned LLM consortiums for psychiatric diagnosis standardization by fundamentally re-architecting the inference pipeline for fully local execution -- ensuring that no patient data is transmitted to, processed by, or stored on any external server at any stage. The platform integrates a consortium of three lightweight, fine-tuned, and quantized open-source LLMs -- Gemma, Phi-3.5-mini, and Qwen2 -- selected for their compact architectures and proven efficiency on resource-constrained mobile hardware. An on-device orchestration layer coordinates ensemble inference and consensus-based diagnostic reasoning, producing DSM-5-aligned assessments for conditions. The platform is designed to assist clinicians with differential diagnosis and evidence-linked symptom mapping, as well as to support patient-facing self-screening with appropriate clinical safeguards. Initial evaluation demonstrates that the proposed zero-egress deployment achieves diagnostic accuracy comparable to its server-side predecessor while sustaining real-time inference latency on commodity mobile hardware.

Static concreteness ratings are widely used in NLP, yet a word's concreteness can shift with context, especially in figurative language such as metaphor, where common concrete nouns can take abstract interpretations. While such shifts are evident from context, it remains unclear how LLMs understand concreteness internally. We conduct a layer-wise and geometric analysis of LLM hidden representations across four model families, examining how models distinguish literal vs figurative uses of the same noun and how concreteness is organized in representation space. We find that LLMs separate literal and figurative usage in early layers, and that mid-to-late layers compress concreteness into a one-dimensional direction that is consistent across models. Finally, we show that this geometric structure is practically useful: a single concreteness direction supports efficient figurative-language classification and enables training-free steering of generation toward more literal or more figurative rewrites.

Surprisal theory hypothesizes that the difficulty of human sentence processing increases linearly with surprisal, the negative log-probability of a word given its context. Computational psycholinguistics has tested this hypothesis using language models (LMs) as proxies for human prediction. While surprisal derived from recent neural LMs generally captures human processing difficulty on naturalistic corpora that predominantly consist of simple sentences, it severely underestimates processing difficulty on sentences that require syntactic disambiguation (garden-path effects). This leads to the claim that the processing difficulty of such sentences cannot be reduced to surprisal, although it remains possible that neural LMs simply differ from humans in next-word prediction. In this paper, we investigate whether it is truly impossible to construct a neural LM that can explain garden-path effects via surprisal. Specifically, instead of evaluating off-the-shelf neural LMs, we fine-tune these LMs on garden-path sentences so as to better align surprisal-based reading-time estimates with actual human reading times. Our results show that fine-tuned LMs do not overfit and successfully capture human reading slowdowns on held-out garden-path items; they even improve predictive power for human reading times on naturalistic corpora and preserve their general LM capabilities. These results provide an existence proof for a neural LM that can explain both garden-path effects and naturalistic reading times via surprisal, but also raise a theoretical question: what kind of evidence can truly falsify surprisal theory?

Large language models are increasingly expected to serve as general-purpose agents that interact with external, stateful tool environments. The Model Context Protocol (MCP) and broader agent skills offer a unified interface for connecting agents with scalable real-world services, but training robust agents remains limited by the lack of realistic environments and principled mechanisms for life-long learning. In this paper, we present \textbf{Agent-World}, a self-evolving training arena for advancing general agent intelligence through scalable environments. Agent-World has two main components: (1) Agentic Environment-Task Discovery, which autonomously explores topic-aligned databases and executable tool ecosystems from thousands of real-world environment themes and synthesizes verifiable tasks with controllable difficulty; and (2) Continuous Self-Evolving Agent Training, which combines multi-environment reinforcement learning with a self-evolving agent arena that automatically identifies capability gaps through dynamic task synthesis and drives targeted learning, enabling the co-evolution of agent policies and environments. Across 23 challenging agent benchmarks, Agent-World-8B and 14B consistently outperforms strong proprietary models and environment scaling baselines. Further analyses reveal scaling trends in relation to environment diversity and self-evolution rounds, offering insights for building general agent intelligence.

Accurate dynamical modeling is essential for simulation and control of embodied systems, yet first-principles models of electromechanical systems often fail to capture complex dissipative effects such as joint friction, stray losses, and structural damping. While residual-learning physics-informed neural networks (PINNs) can effectively augment imperfect first-principles models with data-driven components, the residual terms are typically implemented as unconstrained multilayer perceptrons (MLPs), which may inadvertently inject artificial energy into the system. To more faithfully model the dissipative dynamics, we propose DiLaR-PINN, a dissipative latent residual PINN designed to learn unmodeled dissipative effects in a physically consistent manner. Structurally, the residual network operates only on unmeasurable (latent) state components and is parameterized in a skew-dissipative form that guarantees non-increasing energy for any choice of network parameters. To enable stable and data-efficient training under partial measurability of the state, we further develop a recurrent rollout scheme with a curriculum-based sequence length extension strategy. We validate DiLaR-PINN on a real-world helicopter system and compare it against four baselines: a pure physical model (without a residual network), an unstructured residual MLP, a DiLaR variant with a soft dissipativity constraint, and a black-box LSTM. The results demonstrate that DiLaR-PINN more accurately captures dissipative effects and achieves superior long-horizon extrapolation performance.

Block-encoding is a foundational technique in modern quantum algorithms, enabling the implementation of non-unitary operations by embedding them into larger unitary matrices. While theoretically powerful and essential for advanced protocols like Quantum Singular Value Transformation (QSVT) and Quantum Signal Processing (QSP), the generation of compilable implementations of block-encodings poses a formidable challenge. This work presents the BlockEncoding interface within the Eclipse Qrisp framework, establishing block-encodings as a high-level programming abstraction accessible to a broad scientific audience. Serving as both a technical framework introduction and a hands-on tutorial, this paper explicitly details key underlying concepts abstracted away by the interface, such as block-encoding construction and qubitization, and their practical integration into methods like the Childs-Kothari-Somma (CKS) algorithm. We outline the interface's software architecture, encompassing constructors, core utilities, arithmetic composition, and algorithmic applications such as matrix inversion, polynomial filtering, and Hamiltonian simulation. Through code examples, we demonstrate how this interface simplifies both the practical realization of advanced quantum algorithms and their associated resource estimation.

The ability of humans for lifelong learning is an inspiration for deep learning methods and in particular for continual learning. In this work, we apply Hebbian learning, a biologically inspired learning process, to sound classification. We propose a kernel plasticity approach that selectively modulates network kernels during incremental learning, acting on selected kernels to learn new information and on others to retain previous knowledge. Using the ESC-50 dataset, the proposed method achieves 76.3% overall accuracy over five incremental steps, outperforming a baseline without kernel plasticity (68.7%) and demonstrating significantly greater stability across tasks.

Identifying anomalous instances in tabular data is essential for improving data reliability and maintaining system stability. Due to the scarcity of ground-truth anomaly labels, existing methods mainly rely on unsupervised anomaly detection models, or exploit a small number of labeled anomalies to facilitate detection via sample generation or contrastive learning. However, unsupervised methods lack sufficient anomaly awareness, while current generation and contrastive approaches tend to compute anomalies globally, overlooking the localized anomaly patterns of tabular features, resulting in suboptimal detection performance. To address these limitations, we propose PLAG, a pseudo-label-guided anomaly generation method designed to enhance tabular anomaly detection. Specifically, by utilizing pseudo-anomalies as guidance signals and decoupling the overall anomaly quantification of a sample into an accumulation of feature-level abnormalities, PLAG not only effectively obviates the need for scarce ground-truth labels but also provides a novel perspective for the model to comprehend localized anomalous signals at a fine-grained level. Furthermore, a two-stage data selection strategy is proposed, integrating format verification and uncertainty estimation to rigorously filter candidate samples, thereby ensuring the fidelity and diversity of the synthetic anomalies. Ultimately, these filtered synthetic anomalies serve as robust discriminative guidance, empowering the model to better separate normal and anomalous instances. Extensive experiments demonstrate that PLAG achieves state-of-the-art performance against eight representative baselines. Moreover, as a flexible framework, it integrates seamlessly with existing unsupervised detectors, consistently boosting F1-scores by 0.08 to 0.21.

Zeroth-Order optimization presents a promising memory-efficient paradigm for fine-tuning Large Language Models by relying solely on forward passes. However, its practical adoption is severely constrained by slow wall-clock convergence and high estimation variance. In this work, we dissect the runtime characteristics of ZO algorithms and identify a critical system bottleneck where the generation of perturbations and parameter updates accounts for over 40% of the training latency. We argue that the standard uniform exploration strategy is fundamentally flawed as it fails to account for the heterogeneous sensitivity of layers in deep networks, resulting in computationally wasteful blind searches. To address this structural mismatch, we propose AdaLeZO, an Adaptive Layer-wise ZO optimization framework. By formulating the layer selection process as a non-stationary Multi-Armed Bandit problem, AdaLeZO dynamically allocates the limited perturbation budget to the most sensitive parameters. We further introduce an Inverse Probability Weighting mechanism based on sampling with replacement, which guarantees unbiased gradient estimation while effectively acting as a temporal denoiser to reduce variance. Extensive experiments on LLaMA and OPT models ranging from 6.7B to 30B parameters demonstrate that AdaLeZO achieves 1.7x to 3.0x wall-clock acceleration compared to state-of-the-art methods. Crucially, AdaLeZO functions as a universal plug-and-play module that seamlessly enhances the efficiency of existing ZO optimizers without incurring additional memory overhead.

The multi-scale and non-linear nature of phase-field models of solidification requires fine spatial and temporal discretization, leading to long computation times. This could be overcome with artificial-intelligence approaches. Surrogate models based on neural operators could have a lower computational cost than conventional numerical discretization methods. We propose a new neural operator approach that bridges classical convex-concave splitting schemes with physics-informed learning to accelerate the simulation of phase-field models. It consists of a Deep Ritz method, where a neural operator is trained to approximate a variational formulation of the phase-field model. By training the neural operator with an energy-splitting variational formulation, we enforce the energy dissipation property of the underlying models. We further introduce a custom Reaction-Diffusion Neural Operator (RDNO) architecture, adapted to the operators of the model equations. We successfully apply the deep learning approach to the isotropic Allen-Cahn equation and to anisotropic dendritic growth simulation. We demonstrate that our physically-informed training provides better generalization in out-of-distribution evaluations than data-driven training, while achieving faster inference than traditional Fourier spectral methods.

While modern text-to-image (T2I) models excel at generating images from intricate prompts, they struggle to capture the key details when the inputs are descriptive paragraphs. This limitation stems from the prevalence of concise captions that shape their training distributions. Existing methods attempt to bridge this gap by either fine-tuning T2I models on long prompts, which generalizes poorly to longer lengths; or by projecting the oversize inputs into normal-prompt space and compromising fidelity. We propose Prompt Refraction for Intricate Scene Modeling (PRISM), a compositional approach that enables pre-trained T2I models to process long sequence inputs. PRISM uses a lightweight module to extract constituent representations from the long prompts. The T2I model makes independent noise predictions for each component, and their outputs are merged into a single denoising step using energy-based conjunction. We evaluate PRISM across a wide range of model architectures, showing comparable performances to models fine-tuned on the same training data. Furthermore, PRISM demonstrates superior generalization, outperforming baseline models by 7.4% on prompts over 500 tokens in a challenging public benchmark.

Query auto-completion (QAC) has been widely studied in the context of web search, yet remains underexplored for in-document search, which we term DocQAC. DocQAC aims to enhance search productivity within long documents by helping users craft faster, more precise queries, even for complex or hard-to-spell terms. While global historical queries are available to both WebQAC and DocQAC, DocQAC uniquely accesses document-specific context, including the current document's content and its specific history of user query interactions. To address this setting, we propose a novel adaptive trie-guided decoding framework that uses user query prefixes to softly steer language models toward high-quality completions. Our approach introduces an adaptive penalty mechanism with tunable hyperparameters, enabling a principled trade-off between model confidence and trie-based guidance. To efficiently incorporate document context, we explore retrieval-augmented generation (RAG) and lightweight contextual document signals such as titles, keyphrases, and summaries. When applied to encoder-decoder models like T5 and BART, our trie-guided framework outperforms strong baselines and even surpasses much larger instruction-tuned models such as LLaMA-3 and Phi-3 on seen queries across both seen and unseen documents. This demonstrates its practicality for real-world DocQAC deployments, where efficiency and scalability are critical. We evaluate our method on a newly introduced DocQAC benchmark derived from ORCAS, enriched with query-document pairs. We make both the DocQAC dataset (https://bit.ly/3IGEkbH) and code (https://github.com/rahcode7/DocQAC) publicly available.

Mixture-of-Experts (MoE) models provide a structured approach to combining specialized neural networks and offer greater interpretability than conventional ensembles. While MoEs have been successfully applied to image classification and semantic segmentation, their use in object detection remains limited due to challenges in merging dense and structured predictions. In this work, we investigate model-level mixtures of object detectors and analyze their suitability for improving performance and interpretability in object detection. We propose an MoE architecture that combines YOLO-based detectors trained on semantically disjoint data subsets, with a learned gating network that dynamically weights expert contributions. We study different strategies for fusing detection outputs and for training the gating mechanism, including balancing losses to prevent expert collapse. Experiments on the BDD100K dataset demonstrate that the proposed MoE consistently outperforms standard ensemble approaches and provides insights into expert specialization across domains, highlighting model-level MoEs as a viable alternative to traditional ensembling for object detection. Our code is available at https://github.com/KASTEL-MobilityLab/mixtures-of-experts/.

Recently, code-oriented large language models (LLMs) have demonstrated strong capabilities in translating natural language into executable code. Text-to-SQL is a significant application of this ability, enabling non-technical users to interact with relational databases using natural language. However, state-of-the-art models continue to struggle with highly complex logic, particularly deeply nested statements involving multiple joins and conditions, as well as with real-world database schemas that are noisy or poorly structured. In this paper, we investigate whether curriculum learning can improve the performance of code-based LLMs on Text-to-SQL tasks. Employing benchmarks including Spider and BIRD, we fine-tune models under different curriculum strategies. Our experiments show that naive curriculum, simply ordering training samples by complexity in a single epoch, fails to surpass standard fine-tuning due to catastrophic forgetting. To overcome this, we propose a Modular Adapter Composition (MAC) strategy. By sequentially training tier-specific adapters on incremental complexity levels (Easy to Extra-Hard), we create a scaffolded learning environment that improves performance on complex queries. Our approach not only produces measurable performance gains on the Spider and BIRD benchmarks but also provides a flexible, "Lego-like" architecture, allowing models to be composed and deployed based on specific schema difficulty requirements. These findings demonstrate that structured, modular learning is a superior alternative to monolithic fine-tuning for mastering the syntax and logic of complex code generation.

In this paper, we present three neural network architectures designed for real-time classification of weather conditions (sunny, rain, snow, fog) from images. These models, inspired by recent advances in style transfer, aim to capture the stylistic elements present in images. One model, called "Multi-PatchGAN", is based on PatchGANs used in well-known architectures such as Pix2Pix and CycleGAN, but here adapted with multiple patch sizes for detection tasks. The second model, "Truncated ResNet50", is a simplified version of ResNet50 retaining only its first nine layers. This truncation, determined by an evolutionary algorithm, facilitates the extraction of high-frequency features essential for capturing subtle stylistic details. Finally, we propose "Truncated ResNet50 with Gram Matrix and Attention", which computes Gram matrices for each layer during training and automatically weights them via an attention mechanism, thus optimizing the extraction of the most relevant stylistic expressions for classification. These last two models outperform the state of the art and demonstrate remarkable generalization capability on several public databases. Although developed for weather detection, these architectures are also suitable for other appearance-based classification tasks, such as animal species recognition, texture classification, disease detection in medical imaging, or industrial defect identification.

Speech encoder models are known to model members of some speaker groups (SGs) better than others. However, there has been little work in establishing why this occurs on a technological level. To our knowledge, we present the first layerwise fairness analysis of pretrained self-supervised speech encoder models (S3Ms), probing each embedding layer for speaker identification (SID) automatic speech recognition (ASR). We find S3Ms produce embeddings biased against certain SGs for both tasks, starting at the very first latent layers. Furthermore, we find opposite patterns of layerwise bias for SID vs ASR for all models in our study: SID bias is minimized in layers that minimize overall SID error; on the other hand, ASR bias is maximized in layers that minimize overall ASR error. The inverse bias/error relationship for ASR is unaffected when probing S3Ms that are finetuned for ASR, suggesting SG-level bias is established during pretraining and is difficult to remove.

Current open-source prompt-injection detectors converge on two architectural choices: regular-expression pattern matching and fine-tuned transformer classifiers. Both share failure modes that recent work has made concrete. Regular expressions miss paraphrased attacks. Fine-tuned classifiers are vulnerable to adaptive adversaries: a 2025 NAACL Findings study reported that eight published indirect-injection defenses were bypassed with greater than fifty percent attack success rates under adaptive attacks. This work proposes seven detection techniques that each port a specific mechanism from a discipline outside large-language-model security: forensic linguistics, materials-science fatigue analysis, deception technology from network security, local-sequence alignment from bioinformatics, mechanism design from economics, spectral signal analysis from epidemiology, and taint tracking from compiler theory. Three of the seven techniques are implemented in the prompt-shield v0.4.1 release (Apache 2.0) and evaluated in a four-configuration ablation across six datasets including deepset/prompt-injections, NotInject, LLMail-Inject, AgentHarm, and AgentDojo. The local-alignment detector lifts F1 on deepset from 0.033 to 0.378 with zero additional false positives. The stylometric detector adds 11.1 percentage points of F1 on an indirect-injection benchmark. The fatigue tracker is validated via a probing-campaign integration test. All code, data, and reproduction scripts are released under Apache 2.0.

Large language models are increasingly deployed as protocols: structured multi-call procedures that spend additional computation to transform a baseline answer into a final one. These protocols are evaluated only by end-to-end accuracy, giving limited insight into when they help, when they hurt, and whether their behavior transfers under distribution shift or composition. We propose a paired-outcome measurement interface for auditing a single protocol step on exact-match tasks. For each instance, the interface records a baseline correctness bit $E_0\in\{0,1\}$ and a post-step correctness bit $E_1\in\{0,1\}$, separating correction ($E_0=0\to E_1=1$) from corruption ($E_0=1\to E_1=0$) through two rates: $c=\Pr(E_1=1\mid E_0=0)$ and $γ=\Pr(E_1=0\mid E_0=1)$. These rates predict accuracy changes and define a reusable empirical interface testable across seeds, mixtures, and pipelines. We identify three failure mechanisms. Under mixture shift, pooled estimates of $(c,γ)$ become biased when calibration and deployment mixtures differ; conditioning on a difficulty proxy restores stability without additional model calls. Under presentation contamination, selection protocols alter the interface through stable presentation artifacts when candidate content is fixed. Under state insufficiency, the correctness bit may not carry enough history for multi-step pipelines to compose predictably; a Markov factorization test identifies when composition is valid and where additional state is needed. When a protocol step passes these diagnostics, it becomes an auditable module: gated by estimated gain, conditioned on a difficulty proxy to correct mixture bias, and composed into multi-step pipelines with predictable accuracy. We demonstrate these ideas on synthetic mathematical tasks and on GSM8K, where the calibrated interface correctly predicts when protocol steps should be activated or suppressed.

\We introduce the horospherical depth, an intrinsic notion of statistical depth on Hadamard manifolds, and define the Busemann median as the set of its maximizers. The construction exploits the fact that the linear functionals appearing in Tukey's half-space depth are themselves limits of renormalized distance functions; on a Hadamard manifold the same limiting procedure produces Busemann functions, whose sublevel sets are horoballs, the intrinsic replacements for halfspaces. The resulting depth is parametrized by the visual boundary, is isometry-equivariant, and requires neither tangent-space linearization nor a chosen base point.For arbitrary Hadamard manifolds, we prove that the depth regions are nested and geodesically convex, that a centerpoint of depth at least $1/(d+1)$ exists, and hence that the Busemann median exists for every Borel probability measure. Under strictly negative sectional curvature and mild regularity assumptions, the depth is strictly quasi-concave and the median is unique. We also establish robustness: the depth is stable under total-variation perturbations, and under contamination escaping to infinity the limiting median depends on the escape direction but not on how far the contaminating mass has moved along the geodesic ray, in contrast with the Fréchet mean. Finally, we establish uniform consistency of the sample depth and convergence of sample depth regions and sample Busemann medians; on symmetric spaces of noncompact type, the argument proceeds through a VC analysis of upper horospherical halfspaces, while on general Hadamard manifolds it follows from a compactness argument under a mild non-atomicity assumption.

As reinforcement learning continues to scale the training of large language model-based agents, reliably verifying agent behaviors in complex environments has become increasingly challenging. Existing approaches rely on rule-based verifiers or LLM-as-a-Judge models, which struggle to generalize beyond narrow domains. Agent-as-a-Judge addresses this limitation by actively interacting with environments and tools to acquire verifiable evidence, yet its capabilities remain underexplored. We introduce a benchmark AJ-Bench to systematically evaluate Agent-as-a-Judge across three domains-search, data systems, and graphical user interfaces-comprising 155 tasks and 516 annotated trajectories. The benchmark comprehensively assesses judge agents' abilities in information acquisition, state verification, and process verification. Experiments demonstrate consistent performance gains over LLM-as-a-Judge baselines, while also revealing substantial open challenges in agent-based verification. Our data and code are available at https://aj-bench.github.io/.

Preference optimization is widely used to align large language models (LLMs) with human preferences. However, many margin-based objectives suppress the chosen response along with the rejected one, a phenomenon known as likelihood displacement, and no general mechanism currently prevents this across objectives. We bridge this gap by presenting a unified \emph{incentive-score decomposition} of preference optimization, revealing that diverse objectives share identical local update directions and differ only in their scalar weighting coefficients. Building on this decomposition, by analyzing the dynamics of the chosen/rejected likelihoods, we identify the \emph{disentanglement band} (DB), a simple, testable condition that characterizes when training can avoid likelihood displacement by realizing the preferred pathway: suppressing the loser while maintaining the winner, possibly after an initial transient. Leveraging the DB, we propose a plug-and-play \emph{reward calibration} (RC) that adaptively rebalances chosen versus rejected updates to satisfy the DB and mitigate likelihood displacement, without redesigning the base objective. Empirical results show that RC steers training toward more disentangled dynamics and often improves downstream performance across a range of objectives. Our code is available at https://github.com/IceyWuu/DisentangledPreferenceOptimization.

In large-scale distributed scenarios, increasingly complex tasks demand more intelligent collaboration across networks, requiring the joint extraction of structural representations from data samples. However, conventional task-specific approaches often result in nonstructural embeddings, leading to collapsed variability among data samples within the same class, particularly in classification tasks. To address this issue and fully leverage the intrinsic structure of data for downstream applications, we propose a novel distributed learning framework that ensures both diverse and discriminative representations. For independent and identically distributed (i.i.d.) data, we reformulate and decouple the global optimization function by introducing constraints on representation variance. The update rules are then derived and simplified using a primal-dual approach. For non-i.i.d. data distributions, we tackle the problem by clustering and virtually replicating nodes, allowing model updates within each cluster using block coordinate descent. In both cases, the resulting optimal solutions are theoretically proven to maintain discriminative and diverse properties, with a guaranteed convergence for i.i.d. conditions. Additionally, semantic information from representations is shared among nodes, reducing the need for common neural network architectures. Finally, extensive simulations on MNIST, CIFAR-10 and CIFAR-100 confirm the effectiveness of the proposed algorithms in capturing global structural representations.

Deep search agents can autonomously initiate multi-turn interactions with search engines, thereby exhibiting strong question-answering capabilities. Such performance critically relies on Group Relative Policy Optimization (GRPO) as its core training algorithm. However, GRPO still faces several challenges in deep search settings. First, there exists a substantial mismatch between the correctness of intermediate steps and the reward signal, causing numerous correct intermediate steps to be incorrectly penalized when the final answer is wrong. Second, training is highly unstable, often resulting in degradation of natural language ability or even catastrophic training collapse. Our analysis attributes these issues to coarse-grained advantage assignment and an imbalance between positive and negative advantages. To address these problems, we propose CalibAdv, an advantage calibration method specifically designed for deep search tasks. Specifically, CalibAdv leverages the correctness of intermediate steps to downscale excessive negative advantages at a fine-grained level. It then rebalances positive and negative advantages in the answer component. Extensive experiments across three models and seven benchmarks demonstrate that CalibAdv improves both model performance and training stability. Our code is available at https://github.com/wujwyi/CalibAdv.

Retrieval-augmented generation (RAG) enhances large language models (LLMs) with external knowledge to answer questions more accurately. However, research on evaluating RAG systems-particularly the retriever component-remains limited, as most existing work focuses on single-context retrieval rather than multi-hop queries, where individual contexts may appear irrelevant in isolation but are essential when combined. In this research, we use the HotPotQA, MuSiQue, and SQuAD datasets to simulate a RAG system and compare three LLM-as-judge evaluation strategies, including our proposed Context-Aware Retriever Evaluation (CARE). Our goal is to better understand how multi-hop reasoning can be most effectively evaluated in RAG systems. Experiments with LLMs from OpenAI, Meta, and Google demonstrate that CARE consistently outperforms existing methods for evaluating multi-hop reasoning in RAG systems. The performance gains are most pronounced in models with larger parameter counts and longer context windows, while single-hop queries show minimal sensitivity to context-aware evaluation. Overall, the results highlight the critical role of context-aware evaluation in improving the reliability and accuracy of retrieval-augmented generation systems, particularly in complex query scenarios. To ensure reproducibility, we provide the complete data of our experiments at https://github.com/lorenzbrehme/CARE.

Network change validation remains a critical yet predominantly manual, time-consuming, and error-prone process in modern network operations. While formal network verification has made substantial progress in proving correctness properties, it is typically applied in offline, pre-deployment settings and faces challenges in accommodating continuous changes and validating live production behavior. Current operational approaches typically involve scattered testing tools, resulting in partial coverage and errors that surface only after deployment. In this paper, we present Aether, a novel approach that integrates Generative Agentic AI with a multi-functional Network Digital Twin to automate and streamline network change validation workflows. It features an agentic architecture with five specialized Network Operations AI agents that collaboratively handle the change validation lifecycle from intent analysis to network verification and testing. Aether agents use a unified Network Digital Twin integrating modeling, simulation, and emulation to maintain a consistent, up-to-date network view for verification and testing. By orchestrating agent collaboration atop this digital twin, Aether enables automated, rapid network change validation while reducing manual effort, minimizing errors, and improving operational agility and cost-effectiveness. We evaluate Aether over synthetic network change scenarios covering main classes of network changes and on past incidents from a major ISP operational network, demonstrating promising results in error detection (100%), diagnostic coverage (92-96%), and speed (6-7 minutes) over traditional methods.

Early-stage specifications of safety-critical systems are typically expressed in natural language, making it difficult to derive formal properties suitable for verification and needed to guarantee safety. While recent Large Language Model (LLM)-based approaches can generate formal artifacts from text, they mainly focus on syntactic correctness and do not ensure semantic alignment between informal requirements and formally verifiable properties. We propose an agentic methodology that automatically extracts verification-ready properties from unstructured specifications. The modular pipeline combines requirement extraction, compatibility filtering with respect to a target formalism, and translation into formal properties. Experimental results across three scenarios show that the pipeline generates syntactically and semantically aligned formal properties with a 77.8% accuracy. By explicitly accounting for modeling and verification constraints, the approach is a paving step towards exploiting Artificial Intelligence (AI) to bridge the gap between informal descriptions and semantically meaningful formal verification.

Feature selection is a fundamental machine learning and data mining task, involved with discriminating redundant features from informative ones. It is an attempt to address the curse of dimensionality by removing the redundant features, while unlike dimensionality reduction methods, preserving explainability. Feature selection is conducted in both supervised and unsupervised settings, with different evaluation metrics employed to determine which feature selection algorithm is the best. In this paper, we propose FSEVAL, a feature selection evaluation toolbox accompanied with a visualization dashboard, with the goal to make it easy to comprehensively evaluate feature selection algorithms. FSEVAL aims to provide a standardized, unified, evaluation and visualization toolbox to help the researchers working in the field, conduct extensive and comprehensive evaluation of feature selection algorithms with ease.

Automated analysis of customer feedback on social media is hindered by three challenges: the high cost of annotated training data, the scarcity of evaluation sets, especially in multilingual settings, and privacy concerns that prevent data sharing and reproducibility. We address these issues by developing a generalizable synthetic data generation pipeline applied to a case study on customer distress detection in French public transportation. Our approach utilizes backtranslation with fine-tuned models to generate 1.7 million synthetic tweets from a small seed corpus, complemented by synthetic reasoning traces. We train 600M-parameter reasoners with English and French reasoning that achieve 77-79% accuracy on human-annotated evaluation data, matching or exceeding SOTA proprietary LLMs and specialized encoders. Beyond reducing annotation costs, our pipeline preserves privacy by eliminating the exposure of sensitive user data. Our methodology can be adopted for other use cases and languages.

Is Segment Anything Model 3 (SAM3) capable in segmenting Any Pathology Images? Digital pathology segmentation spans tissue-level and nuclei-level scales, where traditional methods often suffer from high annotation costs and poor generalization. SAM3 introduces Promptable Concept Segmentation, offering a potential automated interface via text prompts. With this work, we propose a systematic evaluation protocol to explore the capability space of SAM3 in a structured manner. Specifically, we evaluate SAM3 under different supervision settings including zero-shot, few-shot, and supervised with varying prompting strategies. Our extensive evaluation on pathological datasets including NuInsSeg, PanNuke and GlaS, reveals that: 1.text-only prompts poorly activate nuclear concepts. 2.performance is highly sensitive to visual prompt types and budgets. 3.few-shot learning offers gains, but SAM3 lacks robustness against visual prompt noise. and 4.a significant gap persists between prompt-based usage and task-trained adapter-based reference. Our study delineates SAM3's boundaries in pathology image segmentation and provides practical guidance on the necessity of pathology domain adaptation.

Large language models are rapidly evolving into interactive coding agents capable of end-to-end web coding, yet existing benchmarks evaluate only narrow slices of this capability, typically text-conditioned generation with static-correctness metrics, leaving visual fidelity, interaction quality, and codebase-level reasoning largely unmeasured. We introduce WebCompass, a multimodal benchmark that provides unified lifecycle evaluation of web engineering capability. Recognizing that real-world web coding is an iterative cycle of generation, editing, and repair, WebCompass spans three input modalities (text, image, video) and three task types (generation, editing, repair), yielding seven task categories that mirror professional workflows. Through a multi-stage, human-in-the-loop pipeline, we curate instances covering 15 generation domains, 16 editing operation types, and 11 repair defect types, each annotated at Easy/Medium/Hard levels. For evaluation, we adopt a checklist-guided LLM-as-a-Judge protocol for editing and repair, and propose a novel Agent-as-a-Judge paradigm for generation that autonomously executes generated websites in a real browser, explores interactive behaviors via the Model Context Protocol (MCP), and iteratively synthesizes targeted test cases, closely approximating human acceptance testing. We evaluate representative closed-source and open-source models and observe that: (1) closed-source models remain substantially stronger and more balanced; (2) editing and repair exhibit distinct difficulty profiles, with repair preserving interactivity better but remaining execution-challenging; (3) aesthetics is the most persistent bottleneck, especially for open-source models; and (4) framework choice materially affects outcomes, with Vue consistently challenging while React and Vanilla/HTML perform more strongly depending on task type.

Electroencephalography (EEG) signals have been promising for long-term braking intensity prediction but are prone to various artifacts that limit their reliability. Here, we propose a novel framework that models EEG signals as mixtures of independent blind sources and identifies those strongly correlated with braking action. Our method employs independent component analysis to decompose EEG into different components and combines time-frequency analysis with Pearson correlations to select braking-related components. Furthermore, we utilize hierarchical clustering to group braking-related components into two clusters, each characterized by a distinct spatial pattern. Additionally, these components exhibit trial-invariant temporal patterns and demonstrate stable and common neural signatures of the emergency braking process. Using power features from these components and historical braking data, we predict braking intensity at a 200 ms horizon. Evaluations on the open source dataset (O.D.) and human-in-the-loop simulation (H.S.) show that our method outperforms state-of-the-art approaches, achieving RMSE reductions of 8.0% (O.D.) and 23.8% (H.S.).

Success in association football relies on both individual skill and coordinated tactics. While recent advancements in spatio-temporal data and deep learning have enabled predictive analyses like trajectory forecasting, the development of tactical design remains limited. Bridging this gap is essential, as prediction reveals what is likely to occur, whereas tactic generation determines what should occur to achieve strategic objectives. In this work, we present TacticGen, a generative model for adaptable and scalable tactic generation. TacticGen formulates tactics as sequences of multi-agent movements and interactions conditioned on the game context. It employs a multi-agent diffusion transformer with agent-wise self-attention and context-aware cross-attention to capture cooperative and competitive dynamics among players and the ball. Trained with over 3.3 million events and 100 million tracking frames from top-tier leagues, TacticGen achieves state-of-the-art precision in predicting player trajectories. Building on it, TacticGen enables adaptable tactic generation tailored to diverse inference-time objectives through classifier guidance mechanism, specified via rules, natural language, or neural models. Its modeling performance is also inherently scalable. A case study with football experts confirms that TacticGen generates realistic, strategically valuable tactics, demonstrating its practical utility for tactical planning in professional football. The project page is available at: https://shengxu.net/TacticGen/.

Prompt-injected memory can improve reasoning without updating model weights, but it also creates a control problem: retrieved content helps only when it is applied in the right state. We study this problem in a strict training-free setting and formulate it as applicability control: when to trigger a memory-assisted second pass, when to trust it, and how to maintain the memory bank over time. Our method combines uncertainty-based routing, confidence-based selective acceptance, bank selection across rule and exemplar memory, and evidence-based governance of the memory bank over time. Under a locked training-free protocol with compute-matched controls, it improves two core arithmetic benchmarks by +7.0 points on SVAMP and +7.67 points on ASDiv over baseline. The same architecture also transfers to QA and agent benchmarks with smaller positive effects and shows the same positive direction on a second checkpoint for the main arithmetic tasks. On arithmetic, the main empirical pattern is that the control architecture, rather than raw memory exposure, drives the improvements on SVAMP and ASDiv. Mechanistically, confidence separates helpful from harmful rule-bank interventions, and under fixed retrieval the repair-versus-corrupt difference localizes to rows whose retrieved set actually contains the edited entries.

We present a phoneme-level analysis of automatic speech recognition (ASR) for two low-resourced and phonologically complex East Caucasian languages, Archi and Rutul, based on curated and standardized speech-transcript resources totaling approximately 50 minutes and 1 hour 20 minutes of audio, respectively. Existing recordings and transcriptions are consolidated and processed into a form suitable for ASR training and evaluation. We evaluate several state-of-the-art audio and audio-language models, including wav2vec2, Whisper, and Qwen2-Audio. For wav2vec2, we introduce a language-specific phoneme vocabulary with heuristic output-layer initialization, which yields consistent improvements and achieves performance comparable to or exceeding Whisper in these extremely low-resource settings. Beyond standard word and character error rates, we conduct a detailed phoneme-level error analysis. We find that phoneme recognition accuracy strongly correlates with training frequency, exhibiting a characteristic sigmoid-shaped learning curve. For Archi, this relationship partially breaks for Whisper, pointing to model-specific generalization effects beyond what is predicted by training frequency. Overall, our results indicate that many errors attributed to phonological complexity are better explained by data scarcity. These findings demonstrate the value of phoneme-level evaluation for understanding ASR behavior in low-resource, typologically complex languages.

Multimodal LLMs can accurately perceive numerical content across modalities yet fail to perform exact multi-digit multiplication when the identical underlying arithmetic problem is presented as numerals, number words, images, or in audio form. Because existing benchmarks often lack systematically paired instances across modalities, it remains difficult to compare genuine arithmetic limits within and across model families. We therefore introduce a controlled multimodal multiplication benchmark that factorially varies digit length, digit sparsity, representation (e.g., numerals vs. number words), and modality (text, rendered images, audio), with paired instances from a reproducible generator. We also define arithmetic load, C, as the product of the total and non-zero digit count as a compact, mechanistically motivated proxy for operation count. Across evaluations, accuracy falls sharply as C grows, often nearing zero by C > 100. Indeed, C remains predictive of performance across modalities and models, with R-squared often > 0.5, nearing the value from more complex measures of arithmetic load that count the number of intermediate arithmetic steps. A separate perception-versus-computation decomposition shows that multimodal degradation is primarily computational rather than perceptual: on matched-perception checks, models are near-perfect (> 99%) across modalities, even when multiplication accuracy drops. Beyond measuring when models fail, we ask which procedures they are predisposed to follow. We introduce a forced-completion loss probe that scores heuristic-specific reasoning prefixes--including columnar multiplication, distributive decomposition, and rounding/compensation. Here, decomposition is favored in both text and vision modalities; heuristic-specific LoRA adapters produce near-orthogonal updates yet degrade accuracy, indicating the base model maintains a well-tuned internal router.

We prove a centre manifold theorem for a map along a manifold-with-boundary of fixed points, and provide an application to the study of gradient descent with large step size on two-layer matrix factorisation problems.

Diffusion models have emerged as powerful tools for a wide range of vision tasks, including text-guided image generation and editing. In this work, we explore their potential for object grounding in remote sensing imagery. We propose a hybrid pipeline that integrates diffusion-based localization cues with state-of-the-art segmentation models such as RemoteSAM and SAM3 to obtain more accurate bounding boxes. By leveraging the complementary strengths of generative diffusion models and foundational segmentation models, our approach enables robust and adaptive object localization across complex scenes. Experiments demonstrate that our pipeline significantly improves localization performance, achieving over a 14% increase in Acc@0.5 compared to existing state-of-the-art methods.

Transformer-based embedding models suffer from quadratic computational and linear memory complexity, limiting their utility for long sequences. We propose recurrent architectures as an efficient alternative, introducing a vertically chunked inference strategy that enables fast embedding generation with memory usage that becomes constant in the input length once it exceeds the vertical chunk size. By fine-tuning Mamba2 models, we demonstrate their viability as general-purpose text embedders, achieving competitive performance across a range of benchmarks while maintaining a substantially smaller memory footprint compared to transformer-based counterparts. We empirically validate the applicability of our inference strategy to Mamba2, RWKV, and xLSTM models, confirming consistent runtime-memory trade-offs across architectures and establishing recurrent models as a compelling alternative to transformers for efficient embedding generation.

Drifting Models [Deng et al., 2026] train a one-step generator by evolving samples under a kernel-based drift field, avoiding ODE integration at inference. The original analysis leaves two questions open. The drift-field iteration admits a locally repulsive regime in a two-particle surrogate, and vanishing of the drift ($V_{p,q}\equiv 0$) is not known to force the learned distribution $q$ to match the target $p$. We derive a contraction threshold for the surrogate and show that a linearly-scheduled friction coefficient gives a finite-horizon bound on the error trajectory. Under a Gaussian kernel we prove that the drift-field equilibrium is identifiable: vanishing of $V_{p,q}$ on any open set forces $q=p$, closing the converse of Proposition 3.1 of Deng et al. Our friction-augmented model, DMF (Drifting Model with Friction), matches or exceeds Optimal Flow Matching on FFHQ adult-to-child domain translation at 16x lower training compute.

We propose MADDPG-K, a scalable extension to Multi-Agent Deep Deterministic Policy Gradient (MADDPG) that addresses the computational limitations of centralized critic approaches. Centralized critics, which condition on the observations and actions of all agents, have demonstrated significant performance gains in cooperative and competitive multi-agent settings. However, their critic networks grow linearly in input size with the number of agents, making them increasingly expensive to train at scale. MADDPG-K mitigates this by restricting each agent's critic to the $k$ closest agents under a chosen metric which in our case is Euclidean distance. This ensures a constant-size critic input regardless of the total agent count. We analyze the complexity of this approach, showing that the quadratic cost it retains arises from cheap scalar distance computations rather than the expensive neural network matrix multiplications that bottleneck standard MADDPG. We validate our method empirically across cooperative and adversarial environments from the Multi-Particle Environment suite, demonstrating competitive or superior performance compared to MADDPG, faster convergence in cooperative settings, and better runtime scaling as the number of agents grows. Our code is available at https://github.com/TimGop/MADDPG-K .

Large Audio-Language Models (LALMs) have made significant progress in audio understanding, yet they primarily operate as perception-and-answer systems without explicit reasoning processes. Existing methods for enhancing audio reasoning rely either on supervised chain-of-thought (CoT) fine-tuning, which is limited by training data quality, or on reinforcement learning (RL) with coarse rewards that do not directly evaluate reasoning quality. As a result, the generated reasoning chains often appear well-structured yet lack specific acoustic grounding. We propose Audio-DeepThinker, a framework built on two core ideas. First, we introduce a hybrid reasoning similarity reward that directly supervises the quality of generated reasoning chains by combining an LLM evaluator assessing logical path alignment, key step coverage, and analytical depth with an embedding similarity component enforcing semantic alignment with reference reasoning chains. Second, we propose a progressive two-stage curriculum that enables high-quality CoT reasoning to emerge through pure RL exploration, without any supervised reasoning fine-tuning, from an instruction-tuned model that possesses no prior chain-of-thought capability. Stage 1 trains on foundational audio QA with the hybrid reward to foster basic reasoning patterns, while Stage 2 shifts to acoustically challenging boundary cases with an LLM-only reward for greater reasoning diversity. Audio-DeepThinker achieves state-of-the-art results on MMAR (74.0%), MMAU-test-mini (78.5%), and MMSU (77.26%), winning 1st Place in the Interspeech 2026 Audio Reasoning Challenge (Single Model Track). Interpretability analyses further reveal that RL training primarily reshapes upper-layer MoE gating mechanisms and that reasoning tokens crystallize progressively in the upper transformer layers, offering mechanistic insights into how audio reasoning emerges through exploration.

Hosted-LLM providers have a silent-substitution incentive: advertise a stronger model while serving cheaper replies. Probe-after-return schemes such as SVIP leave a parallel-serve side-channel, since a dishonest provider can route the verifier's probe to the advertised model while serving ordinary users from a substitute. We propose a commit-open protocol that closes this gap. Before any opening request, the provider commits via a Merkle tree to a per-position sparse-autoencoder (SAE) feature-trace sketch of its served output at a published probe layer. A verifier opens random positions, scores them against a public named-circuit probe library calibrated with cross-backend noise, and decides with a fixed-threshold joint-consistency z-score rule. We instantiate the protocol on three backbones -- Qwen3-1.7B, Gemma-2-2B, and a 4.5x scale-up to Gemma-2-9B with a 131k-feature SAE. Of 17 attackers spanning same-family lifts, cross-family substitutes, and rank-<=128 adaptive LoRA, all are rejected at a shared, scale-stable threshold; the same attackers all evade a matched SVIP-style parallel-serve baseline. A white-box end-to-end attack that backpropagates through the frozen SAE encoder does not close the margin, and a feature-forgery attacker that never runs M_hon is bounded in closed form by an intrinsic-dimension argument. Commitment adds <=2.1% to forward-only wall-clock at batch 32.

Benchmarks are often used as a standard to understand LLM capabilities in different domains. However, aggregate benchmark scores provide limited insight into compositional skill gaps of LLMs and how to improve them. To make these weaknesses visible, we propose Scaffolded Task Design (STaD) framework. STaD generates controlled variations of benchmark tasks based on the concept of scaffolding, which introduces structured, incremental support in a step-by-step manner. Rather than inspecting failures individually, this approach enables systematic and scalable probing of model behavior by identifying the specific reasoning skill compositions they lack. Treating the LLM as a black box, our experiments on six models of varying sizes reveal multiple failure points in three reasoning benchmarks and highlight each model's unique and distinct skill gaps.

Large language models (LLMs) show strong capabilities in general reasoning but typically lack reliability in scientific domains like quantum mechanics, which demand strict adherence to physical constraints. This limitation arises from the scarcity of verifiable training resources and the inadequacy of coarse feedback signals in standard alignment paradigms. To address the data challenge, we introduce QuantumQA, a large-scale dataset constructed via a task-adaptive strategy and a hybrid verification protocol that combines deterministic solvers with semantic auditing to guarantee scientific rigor. Building on this foundation, we propose the verification-aware reward model (VRM) tailored for Reinforcement Learning with Verifiable Rewards (RLVR), which employs an adaptive reward fusion (ARF) mechanism to dynamically integrate deterministic signals from a scientific execution suite (SES) with multidimensional semantic evaluations for precise supervision. Experimental results demonstrate that our method consistently outperforms baselines and general-purpose preference models. Notably, our optimized 8B model achieves performance competitive with proprietary models, validating that incorporating verifiable, rule-based feedback into the reinforcement learning loop offers a parameter-efficient alternative to pure scaling.

LLMs edit text and code by autoregressively regenerating the full output, even when most tokens appear verbatim in the input. We study Copy-as-Decode, a decoding-layer mechanism that recasts edit generation as structured decoding over a two-primitive grammar: <copy lines="i-j"/> references an input line range, <gen>...</gen> emits new content. A token-level FSM guarantees syntactic validity, and a serving-layer primitive updates the KV cache for each copy span via a single parallel-prefill forward rather than $N$ autoregressive steps -- sharing the parallel-forward kernel of speculative decoding but with input tokens as the draft and program-enforced acceptance replacing probabilistic verification. We report an upper-bound analysis that requires no end-to-end training. (i) Kernel speedup: on Qwen2.5-{1.5B, 7B}, copying $N$ tokens via parallel prefill is $6.8\times$--$303\times$ faster than autoregressive ($N \in [8, 512]$, A100 80GB bf16). (ii) Copy ceiling: on ProbeEdit and HumanEvalPack-Fix (Py/JS), $74$--$98\%$ of gold tokens are reachable under the line-level primitive; composed with the empirical kernel over each corpus's span histogram this yields a closed-form wall-clock bound of $29.0\times / 3.4\times / 4.2\times$ ($13.0\times$ pooled). A token-level extension reaches $91$--$99\%$ coverage with $4.5\times$--$6.5\times$ floors. (iii) Pipeline losslessness: oracle programs round-trip through the deterministic resolver on all $482$ cases, localizing any downstream failure to span selection rather than the mechanism. A perturbation study shows pooled EM drops from $100\%$ to $15.48\%$ under off-by-one noise. A fine-tuning pilot on Qwen2.5-Coder-1.5B lifts HEvalFix-Py EM from $0/33$ (untrained) to $12$--$17\%$, a learnability signal, not a production selector. Batched-serving integration and multi-file coverage are scoped as follow-up.

Large language models (LLMs) are increasingly used for creative tasks such as literary translation. Yet translational creativity remains underexplored and is rarely evaluated at scale, while source-text comprehension is typically studied in isolation, despite the fact that, in professional translation, comprehension and creativity are tightly intertwined. We address these gaps with a paired-task framework applied to literary excerpts from 11 books. Task 1 assesses source-text comprehension, and Task 2 evaluates translational creativity through Units of Creative Potential (UCPs), such as metaphors and wordplay. Using a scalable evaluation setup that combines expert human annotations with UCP-based automatic scoring, we benchmark 23 models and four creativity-oriented prompts. Our findings show that strong comprehension does not translate into human-level creativity: models often produce literal or contextually inappropriate renderings, with particularly large gaps for the more distant English-Chinese language pair. Creativity-oriented prompts yield only modest gains, and only one model, Mistral-Large, comes close to human-level creativity (0.167 vs. 0.246). Across all model-prompt combinations, only three exceed a creativity score of 0.1, while the rest remain at or near zero.

Multimodal Large Language Models (MLLMs) have been increasingly used as automatic evaluators-a paradigm known as MLLM-as-a-Judge. However, their reliability and vulnerabilities to biases remain underexplored. We find that many MLLM judges fail to reliably integrate key visual or textual cues, yielding unreliable evaluations when evidence is missing or mismatched, and exhibiting instability under semantically irrelevant perturbations. To address this, we systematically define Compositional Bias in MLLM-as-a-Judge systems and introduce MM-JudgeBias, a benchmark for evaluating it. MM-JudgeBias introduces controlled perturbations across Query, Image, and Response, and evaluates model behavior via two complementary metrics: Bias-Deviation (BD) for sensitivity and Bias-Conformity (BC) for stability. Our dataset of over 1,800 curated and refined multimodal samples, drawn from 29 source benchmarks, enables a fine-grained diagnosis of nine bias types across diverse tasks and domains. Experiments on 26 state-of-the-art MLLMs reveal systematic modality neglect and asymmetric evaluation tendencies, underscoring the need for more reliable judges.

In policy gradient reinforcement learning, access to a differentiable model enables 1st-order gradient estimation that accelerates learning compared to relying solely on derivative-free 0th-order estimators. However, discontinuous dynamics cause bias and undermine the effectiveness of 1st-order estimators. Prior work addressed this bias by constructing a confidence interval around the REINFORCE 0th-order gradient estimator and using these bounds to detect discontinuities. However, the REINFORCE estimator is notoriously noisy, and we find that this method requires task-specific hyperparameter tuning and has low sample efficiency. This paper asks whether such bias is the primary obstacle and what minimal fixes suffice. First, we re-examine standard discontinuous settings from prior work and introduce DDCG, a lightweight test that switches estimators in nonsmooth regions; with a single hyperparameter, DDCG achieves robust performance and remains reliable with small samples. Second, on differentiable robotics control tasks, we present IVW-H, a per-step inverse-variance implementation that stabilizes variance without explicit discontinuity detection and yields strong results. Together, these findings indicate that while estimator switching improves robustness in controlled studies, careful variance control often dominates in practical deployments.

Empathy is essential for fostering natural interactions in spoken dialogue systems, as it enables machines to recognize the emotional tone of human speech and deliver empathetic responses. Recent research has made significant progress in developing empathetic spoken chatbots based on large language models (LLMs). However, several challenges still exist when training such models, including reliance on costly empathetic speech instruction data and a lack of emotional expressiveness in the generated speech. Finetuning LLM with cross-modal empathetic instruction data may also lead to catastrophic forgetting and a degradation of its general capability. To address these challenges, we propose FreezeEmpath, an end-to-end empathetic spoken chatbot trained in a simple and efficient manner. The entire training process relies solely on existing speech instruction data and speech emotion recognition (SER) data, while keeping the LLM's parameters frozen. Experiments demonstrate that FreezeEmpath is able to generate emotionally expressive speech and outperforms other empathetic models in empathetic dialogue, SER, and SpokenQA tasks, demonstrating the effectiveness of our training strategy.

Prompt-based interventions can change model behavior, but trained success alone does not identify where the behaviorally relevant state is represented. We study this question in controlled routing tasks using interfaces chosen on support data, held-out query evaluation, and matched necessity, sufficiency, and wrong-interface controls. On GPT-2 triop, an early interface supports exact transfer under these tests. On GPT-2 add/sub, zero-retrain compiled transfer at the fixed interface recovers most of donor routing accuracy, while trainable prompt slots can relearn the same behavior at several other positions only after additional support examples and optimization. These results distinguish fixed-interface reuse from prompt relocation in a setting where the two can be tested directly. Qwen routing provides a cross-architecture consistency check for the same matched-interface pattern at the operator token, although donor-specific identity on the local V-path remains unresolved. Generation and reasoning branches are used to map scope: they show broader transport or weaker controller identifiability once control depends on longer trajectories or harder selection. In controlled routing, fixed-interface transfer is therefore stronger evidence of reuse than trained prompt success alone.

Deep learning (DL) has become a cornerstone of modern machine learning (ML) praxis. We introduce the R package mlr3torch, which is an extensible DL framework for the mlr3 ecosystem. It is built upon the torch package, and simplifies the definition, training, and evaluation of neural networks for both tabular data and generic tensors (e.g., images) for classification and regression. The package implements predefined architectures, and torch models can easily be converted to mlr3 learners. It also allows users to define neural networks as graphs. This representation is based on the graph language defined in mlr3pipelines and allows users to define the entire modeling workflow, including preprocessing, data augmentation, and network architecture, in a single graph. Through its integration into the mlr3 ecosystem, the package allows for convenient resampling, benchmarking, preprocessing, and more. We explain the package's design and features and show how to customize and extend it to new problems. Furthermore, we demonstrate the package's capabilities using three use cases, namely hyperparameter tuning, fine-tuning, and defining architectures for multimodal data. Finally, we present some runtime benchmarks.

Automated fetal head segmentation in ultrasound images is critical for accurate biometric measurements in prenatal care. While existing deep learning approaches have achieved a reasonable performance, they struggle with issues like low contrast, noise, and complex anatomical boundaries which are inherent to ultrasound imaging. This paper presents Attention-ResUNet. It is a novel architecture that synergistically combines residual learning with multi-scale attention mechanisms in order to achieve enhanced fetal head segmentation. Our approach integrates attention gates at four decoder levels to focus selectively on anatomically relevant regions while suppressing the background noise, and complemented by residual connections which facilitates gradient flow and feature reuse. Extensive evaluation on the HC18 Challenge dataset where n = 200 demonstrates that Attention ResUNet achieves a superior performance with a mean Dice score of 99.30 +/- 0.14% against similar architectures. It significantly outperforms five baseline architectures including ResUNet (99.26%), Attention U-Net (98.79%), Swin U-Net (98.60%), Standard U-Net (98.58%), and U-Net++ (97.46%). Through statistical analysis we confirm highly significant improvements (p < 0.001) with effect sizes that range from 0.230 to 13.159 (Cohen's d). Using Saliency map analysis, we reveal that our architecture produces highly concentrated, anatomically consistent activation patterns, which demonstrate an enhanced interpretability which is crucial for clinical deployment. The proposed method establishes a new state of the art performance for automated fetal head segmentation whilst maintaining computational efficiency with 14.7M parameters and a 45 GFLOPs inference cost. Code repository: https://github.com/Ammar-ss

Recently, large language models (LLMs) have advanced recommendation systems (RSs), and recent works have begun to explore how to integrate LLMs into industrial RSs. While most approaches deploy LLMs offline to generate and pre-cache augmented representations for RSs, high-dimensional representations from LLMs introduce substantial storage and computational costs. Thus, it is crucial to compress LLM representations effectively. However, we identify a counterintuitive phenomenon during representation compression: Mid-layer Representation Advantage (MRA), where representations from middle layers of LLMs outperform those from final layers in recommendation tasks. This degraded final layer renders existing compression methods, which typically compress on the final layer, suboptimal. We interpret this based on modularity theory that LLMs develop spontaneous internal functional modularity and force the final layer to specialize in the proxy training task. Thus, we propose \underline{M}odul\underline{a}r \underline{R}epresentation \underline{C}ompression (MARC) to explicitly control the modularity of LLMs. First, Modular Adjustment explicitly introduces compression and task adaptation modules, enabling the LLM to operate strictly as a representation-learning module. Next, to ground each module to its specific task, Modular Task Decoupling uses information constraints and different network structures to decouple tasks. Extensive experiments validate that MARC addresses MRA and produces efficient representations. Notably, MARC achieved a 2.82% eCPM lift in an online A/B test within a large-scale commercial search advertising scenario.

Automated medical report generation for 3D PET/CT imaging is fundamentally challenged by the high-dimensional nature of volumetric data and a critical scarcity of annotated datasets, particularly for low-resource languages. Current black-box methods map whole volumes to reports, ignoring the clinical workflow of analyzing localized Regions of Interest (RoIs) to derive diagnostic conclusions. In this paper, we bridge this gap by introducing VietPET-RoI, the first large-scale 3D PET/CT dataset with fine-grained RoI annotation for a low-resource language, comprising 600 PET/CT samples and 1,960 manually annotated RoIs, paired with corresponding clinical reports. Furthermore, to demonstrate the utility of this dataset, we propose HiRRA, a novel framework that mimics the professional radiologist diagnostic workflow by employing graph-based relational modules to capture dependencies between RoI attributes. This approach shifts from global pattern matching toward localized clinical findings. Additionally, we introduce new clinical evaluation metrics, namely RoI Coverage and RoI Quality Index, that measure both RoI localization accuracy and attribute description fidelity using LLM-based extraction. Extensive evaluation demonstrates that our framework achieves SOTA performance, surpassing existing models by 19.7% in BLEU and 4.7% in ROUGE-L, while achieving a remarkable 45.8% improvement in clinical metrics, indicating enhanced clinical reliability and reduced hallucination. Our code and dataset are available on GitHub.

This paper investigates the off-policy evaluation (OPE) problem from a distributional perspective. Rather than focusing solely on the expectation of the total return, as in most existing OPE methods, we aim to estimate the entire return distribution. To this end, we introduce a quantile-based approach for OPE using deep quantile process regression, presenting a novel algorithm called Deep Quantile Process regression-based Off-Policy Evaluation (DQPOPE). We provide new theoretical insights into the deep quantile process regression technique, extending existing approaches that estimate discrete quantiles to estimate a continuous quantile function. A key contribution of our work is the rigorous sample complexity analysis for distributional OPE with deep neural networks, bridging theoretical analysis with practical algorithmic implementations. We show that DQPOPE achieves statistical advantages by estimating the full return distribution using the same sample size required to estimate a single policy value using conventional methods. Empirical studies further show that DQPOPE provides significantly more precise and robust policy value estimates than standard methods, thereby enhancing the practical applicability and effectiveness of distributional reinforcement learning approaches.

Processing-in-Memory (PIM) architectures offer a promising solution to the memory bottlenecks in data-intensive machine learning, yet often overlook the growing challenge of activation memory footprint. Conventional PIM approaches struggle with massive KV cache sizes generated in long-context scenarios by Transformer-based models, frequently exceeding PIM's limited memory capacity, while techniques like sparse attention can conflict with PIM's need for data locality. Existing PIM approaches and quantization methods are often insufficient or poorly suited for leveraging the unique characteristics of activations. This work identifies an opportunity for PIM-specialized activation quantization to enhance bandwidth and compute efficiency. We explore clustering-based vector quantization approaches, which align well with activation characteristics and PIM's internal bandwidth capabilities. Building on this, we introduce AQPIM, a novel PIM-aware activation quantization framework based on Product Quantization (PQ), optimizing it for modern Large Language Models (LLMs). By performing quantization directly within memory, AQPIM leverages PIM's high internal bandwidth and enables direct computation on compressed data, significantly reducing both memory footprint and computational overhead for attention computation. AQPIM addresses PQ's accuracy challenges by introducing several algorithmic optimizations. Evaluations demonstrate that AQPIM achieves significant performance improvements, drastically reducing of GPU-CPU communication that can account for 90$\sim$98.5\% of decoding latency, together with 3.4$\times$ speedup over a SOTA PIM approach.

Large-scale dataset distillation requires storing auxiliary soft labels that can be 30-40x larger on ImageNet-1K and 200x larger on ImageNet-21K than the condensed images, undermining the goal of dataset compression. We identify two fundamental issues necessitating such extensive labels: (1) insufficient image diversity, where high within-class similarity in synthetic images requires extensive augmentation, and (2) insufficient supervision diversity, where limited variety in supervisory signals during training leads to performance degradation at high compression rates. To address these challenges, we propose Label Pruning and Quantization for Large-scale Distillation (LPQLD). We enhance image diversity via class-wise batching and batch-normalization supervision during synthesis. For supervision diversity, we introduce Label Pruning with Dynamic Knowledge Reuse to improve label-per-augmentation diversity, and Label Quantization with Calibrated Student-Teacher Alignment to improve augmentation-per-image diversity. Our approach reduces soft label storage by 78x on ImageNet-1K and 500x on ImageNet-21K while improving accuracy by up to 7.2% and 2.8%, respectively. Extensive experiments validate the superiority of LPQLD across different network architectures and dataset distillation methods. Code is available at https://github.com/he-y/soft-label-pruning-quantization-for-dataset-distillation.

With the rapid advancement of artificial intelligence, multi-agent systems (MASs) are evolving from classical paradigms toward architectures built upon large foundation models (LFMs). This survey provides a systematic review and comparative analysis of classical MASs (CMASs) and LFM-based MASs (LMASs). First, within a closed-loop coordination framework, CMASs are reviewed across four fundamental dimensions: perception, communication, decision-making, and control. Beyond this framework, LMASs integrate LFMs to lift collaboration from low-level state exchanges to semantic-level reasoning, enabling more flexible coordination and improved adaptability across diverse scenarios. Then, a comparative analysis is conducted to contrast CMASs and LMASs across architecture, operating mechanism, adaptability, and application. Finally, future perspectives on MASs are presented, summarizing open challenges and potential research opportunities.

Most agents today ``self-evolve'' by following rewards and rules defined by humans. However, this process remains fundamentally dependent on external supervision; without human guidance, the evolution stops. In this work, we train agents to possess an intrinsic meta-evolution capability to spontaneously learn about unseen environments prior to task execution. To instill this ability, we design an outcome-based reward mechanism that measures how much an agent's self-generated world knowledge improves its success rate on downstream tasks. This reward signal is used exclusively during the training phase to teach the model how to explore and summarize effectively. At inference time, the agent requires no external rewards or human instructions. It spontaneously performs native self-evolution to adapt to unknown environments using its internal parameters. When applied to Qwen3-30B and Seed-OSS-36B, this shift to native evolution yields a 20% performance increase on WebVoyager and WebWalker. Most strikingly, the generated world knowledge even enables a compact 14B Qwen3 model to outperform the unassisted Gemini-2.5-Flash, establishing a new paradigm for truly evolving agents.

Digital marketplaces processing billions of dollars annually represent critical infrastructure in sociotechnical ecosystems, yet their performance optimization lacks principled measurement frameworks that can inform algorithmic governance decisions regarding market efficiency and fairness from complex market data. By looking at orderbook data from double auction markets alone, because bids and asks do not represent true maximum willingnesses to buy and true minimum willingnesses to sell, there is little an economist can say about the market's actual performance in terms of allocative efficiency. We turn to experimental data to address this issue, `inverting' the standard induced value approach of double auction experiments. Our aim is to predict key market features relevant to market efficiency, particularly allocative efficiency, using orderbook data only -- specifically bids, asks and price realizations, but not the induced reservation values -- as early as possible. Since there is no established model of strategically optimal behavior in these markets, and because orderbook data is highly unstructured, non-stationary and non-linear, we propose quantile-based normalization techniques that help us build general predictive models. We develop and train several models, including linear regressions and gradient boosting trees, leveraging quantile-based input from the underlying supply-demand model. Our models can predict allocative efficiency with reasonable accuracy from the earliest bids and asks, and these predictions improve with additional realized price data. The performance of the prediction techniques varies by target and market type. Our framework holds significant potential for application to real-world market data, offering valuable insights into market efficiency and performance, even prior to any trade realizations.

We study post-training W4A4 quantization in a controlled 300M-parameter SwiGLU decoder-only language model trained on 5B tokens of FineWeb-Edu, and ask which input-activation sites dominate the error. Naive round-to-nearest W4A4 collapses validation perplexity from FP16 23.6 to 1727. A simple residual-axis training-time intervention -- Depth Registers with a register-magnitude hinge loss (DR+sink) -- reduces this to 119 (about 14x) at matched FP16 PPL and matched zero-shot capacity, and composes with SmoothQuant to 39.9 PPL. The residual ~2 PPL gap to FP16 is the diagnostic core. We decompose W4A4 damage by input-activation site: the five trainable linears in a SwiGLU block split into residual-axis readers (qkv, w1, w3) and block-internal generators (o_proj, w2). Elementary norm arguments show residual-axis magnitude control bounds readers tightly but leaves w2's bilinear input bounded only by the trivial product of factor bounds; empirically, DR+sink collapses reader kurtosis while leaving generators essentially unchanged, and the reader-rescued W4A4 residue is flat at ~0.28 nats across three matched checkpoints with Delta-remove(w2) dominating. We present DR+sink as a training-time probe rather than a deployment proposal: a post-hoc alternative (Per-Linear QuaRot) nearly matches it on the reader axis. Full QuaRot -- adding online per-head value Hadamard plus online w2-input rotation -- does not close the gap either, directly testing the prediction that orthogonal rotation cannot bound the bilinear SwiGLU tail. Claims are specific to our 300M, 5B-token, single-seed setting, and our experiments do not isolate the partition from the hinge.

Low-Rank Adaptation (LoRA) has become a widely adopted parameter-efficient fine-tuning method for large language models, with its effectiveness largely influenced by the allocation of ranks and scaling factors, as well as initialization. Existing LoRA variants typically address only one of these factors, often at the cost of increased training complexity or reduced practical efficiency. In this work, we present Task-aware Low-Rank Adaptation (TLoRA), a unified framework that jointly optimizes initialization and resource allocation at the outset of training. TLoRA introduces a data-driven initialization strategy that aligns the LoRA $A$ matrix with task-relevant subspaces by performing singular value decomposition on the product of pre-trained weights and input activation covariance. After this, the $A$ matrix is frozen, and only the $B$ matrix is trained. Furthermore, TLoRA employs a sensitivity-based importance metric to adaptively allocate ranks and scaling factors across layers under a fixed parameter budget. We conduct extensive experiments that demonstrate TLoRA consistently performs excellently across various tasks, including natural language understanding, commonsense reasoning, math reasoning, code generation, and chat generation, while significantly reducing the number of trainable parameters.

Decision-making is a cognitively intensive task that requires synthesizing relevant information from multiple unstructured sources, weighing competing factors, and incorporating subjective user preferences. Existing methods, including large language models and traditional decision-support systems, fall short: they often overwhelm users with information or fail to capture nuanced preferences accurately. We present Decisive, an interactive decision-making framework that combines document-grounded reasoning with Bayesian preference inference. Our approach grounds decisions in an objective option-scoring matrix extracted from source documents, while actively learning a user's latent preference vector through targeted elicitation. Users answer pairwise tradeoff questions adaptively selected to maximize information gain over the final decision. This process converges efficiently, minimizing user effort while ensuring recommendations remain transparent and personalized. Through extensive experiments, we demonstrate that our approach significantly outperforms both general-purpose LLMs and existing decision-making frameworks achieving up to 20% improvement in decision accuracy over strong baselines across domains.

The use of disaggregated or far memory systems such as CXL memory pools has renewed interest in Near-Data Processing (NDP): situating cores close to memory to reduce bandwidth requirements to and from the CPU. Hardware designs for such accelerators are appearing, but there lack clean, portable OS abstractions for programming them. We propose a programming model for NDP devices based on familiar OS abstractions: virtual processors (processes) and inter-process communication channels (like Unix pipes). While appealing from a user perspective, a naive implementation of such abstractions is inappropriate for NDP accelerators: the paucity of processing power in some hardware designs makes classical processes overly heavyweight, and IPC based on shared buffers makes no sense in a system designed to reduce memory bandwidth. Accordingly, we show how to implement these abstractions in a lightweight and efficient manner by exploiting compilation and interconnect protocols. We demonstrate them with a real hardware platform runing applications with a range of memory access patterns, including bulk memory operations, in-memory databases and graph applications. Crucially, we show not only the benefits over CPU-only implementations, but also the critical importance of efficient, low-latency communication channels between CPU and NDP accelerators, a feature largely neglected in existing proposals.

Post-training quantization (PTQ) is essential for deploying large diffusion transformers on resource-constrained hardware, but aggressive 4-bit quantization significantly degrades generative performance. Low-rank approximation methods have emerged as a promising solution by appending auxiliary linear branches to restore performance. However, current state-of-the-art approaches assume these branches must retain high precision (W16A16) and rely on heavy, data-dependent calibration for initialization. We challenge both limitations with LoRaQ (Low-Rank Approximated Quantization), a simple, data-free calibration approach that optimizes quantization error compensation. By overcoming the need for high-precision branches, LoRaQ enables the first fully sub-16 bit pipeline, allowing the low-rank branch itself to be quantized. We demonstrate that, at equal memory overhead, LoRaQ outperforms the state-of-the-art methods in their native implementations on Pixart-$Σ$ and SANA. We also analyze mixed-precision configurations, showing that setups such as W8A8, W6A6, and W4A8 for the low-rank branch, alongside a W4 main layer, yield superior results while maintaining a fully quantized architecture compatible with modern mixed-precision hardware.

In recent years, multimodal multidomain fake news detection has garnered increasing attention. Nevertheless, this direction presents two significant challenges: (1) Failure to Capture Cross-Instance Narrative Consistency: existing models usually evaluate each news in isolation, fail to capture cross-instance narrative consistency, and thus struggle to address the spread of cluster based fake news driven by social media; (2) Lack of Domain Specific Knowledge for Reasoning: conventional models, which rely solely on knowledge encoded in their parameters during training, struggle to generalize to new or data-scarce domains (e.g., emerging events or niche topics). To tackle these challenges, we introduce Retrieval-Augmented Multimodal Model for Fake News Detection (RAMM). First, RAMM employs a Multimodal Large Language Model (MLLM) as its backbone to capture cross-modal semantic information from news samples. Second, RAMM incorporates an Abstract Narrative Alignment Module. This component adaptively extracts abstract narrative consistency from diverse instances across distinct domains, aggregates relevant knowledge, and thereby enables the modeling of high-level narrative information. Finally, RAMM introduces a Semantic Representation Alignment Module, which aligns the model's decision-making paradigm with that of humans - specifically, it shifts the model's reasoning process from direct inference on multimodal features to an instance-based analogical reasoning process. Extensive experimental results on three public datasets validate the efficacy of our proposed approach. Our code is available at the following link: https://github.com/li-yiheng/RAMM

This paper presents factorized linear projection (FLiP) models for understanding pretrained sentence embedding spaces. We train FLiP models to recover the lexical content from multilingual (LaBSE), multimodal (SONAR) and API-based (Gemini) sentence embedding spaces in several high- and mid-resource languages. We show that FLiP can recall more than 75% of lexical content from the embeddings, significantly outperforming existing non-factorized baselines. Using this as a diagnostic tool, we uncover the modality and language biases across the selected sentence encoders and provide practitioners with intrinsic insights about the encoders without relying on conventional downstream evaluation tasks. Our implementation is public https://github.com/BUTSpeechFIT/FLiP.

Adapting large language models (LLMs) to low-resource languages (LRLs) is constrained by the scarcity of task data and computational resources. Although Proxy Tuning offers a logit-level strategy for introducing scaling effects, it often fails in LRL settings because the large model's weak LRL competence might overwhelm the knowledge of specialized smaller models. We thus propose TriMix, a test-time logit fusion framework that dynamically balances capabilities from three different sources: LRL competence from a continually pretrained small model, task competence from high-resource language instruction tuning, and the scaling benefits of large models. It is data- and compute-efficient, requiring no LRL task annotations, and only continual pretraining on a small model. Experiments across four model families and eight LRLs show that TriMix consistently outperforms single-model baselines and Proxy Tuning. Our analysis reveals that prioritizing the small LRL-specialized model's logits is crucial for success, challenging the prevalent large-model-dominant assumption.

Integrating large language models (LLMs) into automatic speech recognition (ASR) has become a mainstream paradigm in recent years. Although existing LLM-based ASR models demonstrate impressive performance on public benchmarks, their training remains predominantly data-driven, leaving key practical challenges insufficiently addressed -- particularly limited downward scalability in resource-constrained deployments and hallucinations under acoustically challenging conditions. To address these issues, we present NIM4-ASR, a production-oriented LLM-based ASR framework optimized for both efficiency and robustness. Grounded in a principled delineation of functional roles between the encoder and the LLM, we redesign the multi-stage training paradigm to align each module with its intended capability boundary. Specifically, we reformulate the pre-training architecture and objective to mitigate the modality gap and improve parameter efficiency; introduce an iterative asynchronous SFT stage to preserve acoustic fidelity and constrain representation drift; and design an ASR-specialized reinforcement learning stage to further enhance recognition quality and robustness. We additionally incorporate a suite of production-oriented optimizations, including robustness under noisy and silent conditions, real-time streaming inference, and hotword customization via retrieval-augmented generation (RAG). Experiments show that NIM4-ASR achieves state-of-the-art performance on multiple public benchmarks with merely 2.3B parameters, while substantially outperforming larger-scale competitors on internal benchmarks -- particularly in entity-intensive real-world scenarios. NIM4-ASR further supports million-scale hotword customization via RAG with sub-millisecond retrieval latency, enabling efficient adaptation to emerging entities and personalized user requirements.

Prefilling computational costs pose a significant bottleneck for Large Language Models (LLMs) and Large Multimodal Models (LMMs) in long-context settings. While token pruning reduces sequence length, prior methods rely on heuristics that break compatibility with hardware-efficient kernels like FlashAttention. In this work, we observe that tokens evolve toward \textit{semantic fixing points}, making further processing redundant. To this end, we introduce Delta Attention Selective Halting (DASH), a training-free policy that monitors the layer-wise update dynamics of the self-attention mechanism to selectively halt stabilized tokens. Extensive evaluation confirms that DASH generalizes across language and vision benchmarks, delivering significant prefill speedups while preserving model accuracy and hardware efficiency. Code will be released at https://github.com/verach3n/DASH.git.

LLMs are increasingly presented as collaborators in programming, design, writing, and analysis. Yet the practical experience of working with them often falls short of this promise. In many settings, users must diagnose misunderstandings, reconstruct missing assumptions, and repeatedly repair misaligned responses. This poster introduces a conceptual framework for understanding why such collaboration remains fragile. Drawing on a constructivist grounded theory analysis of 16 interviews with designers, developers, and applied AI practitioners working on LLM-enabled systems, and informed by literature on human-AI collaboration, we argue that stable collaboration depends not only on model capability but on the interaction's grounding conditions. We distinguish three recurrent structures of human-AI work: one-shot assistance, weak collaboration with asymmetric repair, and grounded collaboration. We propose that collaboration breaks down when the appearance of partnership outpaces the grounding capacity of the interaction and contribute a framework for discussing grounding, repair, and interaction structure in LLM-enabled work.

In real-world applications of noninvasive electroencephalography (EEG), specialized decoders often show limited generalizability across diverse tasks under subject-independent settings. One central challenge is that task-relevant EEG signals often follow different temporal organization patterns across tasks, while many existing methods rely on task-tailored architectural designs that introduce task-specific temporal inductive biases. This mismatch makes it difficult to adapt temporal modeling across tasks without changing the model configuration. To address these challenges, we propose DSAINet, an efficient dual-scale attentive interaction network for general EEG decoding. Specifically, DSAINet constructs shared spatiotemporal token representations from raw EEG signals and models diverse temporal dynamics through parallel convolutional branches at fine and coarse scales. The resulting representations are then adaptively refined by intra-branch attention to emphasize salient scale-specific patterns and by inter-branch attention to integrate task-relevant features across scales, followed by adaptive token aggregation to yield a compact representation for prediction. Extensive experiments on five downstream EEG decoding tasks across ten public datasets show that DSAINet consistently outperforms 13 representative baselines under strict subject-independent evaluation. Notably, this performance is achieved using the same architecture hyperparameters across datasets. Moreover, DSAINet achieves a favorable accuracy-efficiency trade-off with only about 77K trainable parameters and provides interpretable neurophysiological insights. The code is publicly available at https://github.com/zy0929/DSAINet.

Small language models fine-tuned for graph property estimation have demonstrated strong in-distribution performance, yet their generalization capabilities beyond training conditions remain poorly understood. In this work, we systematically investigate the boundaries of structural inference in fine-tuned small language models along two generalization axes - graph size and graph family distribution - and assess domain-learning capability on real-world graph benchmarks. Using a controlled experimental setup with three instruction-tuned models in the 3-4B parameter class and two graph serialization formats, we evaluate performance on graphs substantially larger than the training range and across held-out random graph families. Our results show that fine-tuned models maintain strong ordinal consistency across structurally distinct graph families and continue to rank graphs by structural properties on inputs substantially larger than those seen during training, with distinct architecture-specific degradation profiles. These findings delineate where fine-tuned small language models generalize reliably, providing empirical grounding for their use in graph-based reasoning tasks.

Recent multimodal large language models have shown promising ability in generating humorous captions for images, yet they still lack stable control over explicit cultural context, making it difficult to jointly maintain image relevance, contextual appropriateness, and humor quality under a specified cultural background. To address this limitation, we introduce a new multimodal generation task, culture-aware humorous captioning, which requires a model to generate a humorous caption conditioned on both an input image and a target cultural context. Captions generated under different cultural contexts are not expected to share the same surface form, but should remain grounded in similar visual situations or humorous rationales.To support this task, we establish a six-dimensional evaluation framework covering image relevance, contextual fit, semantic richness, reasonableness, humor, and creativity. We further propose a staged alignment framework that first initializes the model with high-resource supervision under the Western cultural context, then performs multi-dimensional preference alignment via judge-based GRPO with a Degradation-aware Prototype Repulsion Constraint to mitigate reward hacking in open-ended generation, and finally adapts the model to the Eastern cultural context with a small amount of supervision. Experimental results show that our method achieves stronger overall performance under the proposed evaluation framework, with particularly large gains in contextual fit and a better balance between image relevance and humor under cultural constraints.

Bayesian Deep Ensembles (BDEs) represent a powerful approach for uncertainty quantification in deep learning, combining the robustness of Deep Ensembles (DEs) with flexible multi-chain MCMC. While DEs are affordable in most deep learning settings, (long) sampling of Bayesian neural networks can be prohibitively costly. Yet, adding sampling after optimizing the DEs has been shown to yield significant improvements. This leaves a critical practical question: How long should the sequential sampling process continue to yield significant improvements over the initial optimized DE baseline? To tackle this question, we propose a stopping rule based on E-values. We formulate the ensemble construction as a sequential anytime-valid hypothesis test, providing a principled way to decide whether or not to reject the null hypothesis that MCMC offers no improvement over a strong baseline, to early stop the sampling. Empirically, we study this approach for diverse settings. Our results demonstrate the efficacy of our approach and reveal that only a fraction of the full-chain budget is often required.

Drowning is an omnipresent risk associated with any activity on or in the water, and rescuing a drowning person is particularly challenging because of the time pressure, making a short response time important. Further complicating water rescue are unsupervised and extensive swimming areas, precise localization of the target, and the transport of rescue personnel. Technical innovations can provide a remedy: We propose an Unmanned Aircraft System (UAS), also known as a drone-in-a-box system, consisting of a fleet of Unmanned Aerial Vehicles (UAVs) allocated to purpose-built hangars near swimming areas. In an emergency, the UAS can be deployed in addition to Standard Rescue Operation (SRO) equipment to locate the distressed person early by performing a fully automated Search and Rescue (S&R) operation and dropping a flotation device. In this paper, we address automatically locating distressed swimmers using the image-based object detection architecture You Only Look Once (YOLO). We present a dataset created for this application and outline the training process. We evaluate the performance of YOLO versions 3, 5, and 8 and architecture sizes (nano, extra-large) using Mean Average Precision (mAP) metrics mAP@.5 and mAP@.5:.95. Furthermore, we present two Discrete-Event Simulation (DES) approaches to simulate response times of SRO and UAS-based water rescue. This enables estimation of time savings relative to SRO when selecting the UAS configuration (type, number, and location of UAVs and hangars). Computational experiments for a test area in the Lusatian Lake District, Germany, show that UAS assistance shortens response time. Even a small UAS with two hangars, each containing one UAV, reduces response time by a factor of five compared to SRO.

Topic-controlled summarisation enables users to generate summaries focused on specific aspects of source documents. This paper investigates a data augmentation strategy for training small language models (sLMs) to perform topic-controlled summarisation. We propose a pairwise data augmentation method that combines contexts from different documents to create contrastive training examples, enabling models to learn the relationship between topics and summaries more effectively. Using the SciTLDR dataset enriched with Wikipedia-derived topics, we systematically evaluate how augmentation scale affects model performance. Results show consistent improvements in win rate and semantic alignment as the augmentation scale increases, while the amount of real training data remains fixed. Consequently, a T5-base model trained with our augmentation approach achieves competitive performance relative to larger models, despite using significantly fewer parameters and substantially fewer real training examples.

Matrix-level low-rank compression is a promising way to reduce the cost of large language models, but running compression and evaluating the resulting models on language tasks can be prohibitively expensive. Can compression-induced degradation be predicted before committing to this compute? We systematically analyze the Qwen3 and Gemma3 model families across four representative low-rank compression methods: vanilla SVD, two ASVD variants, and SVD-LLM. We find that stable rank and information density, measured in bits per parameter, dominate performance degradation. The interaction term $γ\cdot \barρ_s$, defined as compression ratio times stable rank, is a robust predictor of accuracy degradation, achieving leave-one-out cross-validation Pearson correlations of $0.890$ for attention layers and $0.839$ for MLP layers. We provide theoretical intuition for why this predictor succeeds by connecting it to standard SVD truncation bounds and error composition mechanisms in transformer layers. These findings enable a predict-then-compress workflow: compute $γ\cdot \barρ_s$ from weights, estimate degradation, and invest compute only in desirable configurations.

Reconstructing continuous environmental fields from sparse and irregular observations remains a central challenge in environmental modelling and biodiversity informatics. Many ecological datasets are heterogeneous in space and time, making grid-based approaches difficult to scale or generalise across domains. Here, we evaluate implicit neural representations (INRs) as a coordinate-based modelling framework for learning continuous spatial and spatio-temporal fields directly from coordinate inputs. We analyse their behaviour across three representative modelling scenarios: species distribution reconstruction, phenological dynamics, and morphological segmentation derived from open biodiversity data. Beyond predictive performance, we examine interpolation behaviour, spatial coherence, and computational characteristics relevant for environmental modelling workflows, including scalability, resolution-independent querying, and architectural inductive bias. Results show that neural fields provide stable continuous representations with predictable computational cost, complementing classical smoothers and tree-based approaches. These findings position coordinate-based neural fields as a flexible representation layer that can be integrated into environmental modelling pipelines and exploratory analysis frameworks for large, irregularly sampled datasets.

While attack graphs are useful for identifying major cybersecurity threats affecting a system, they do not provide operational support for determining the likelihood of having a known vulnerability exploited, or that critical system nodes are likely to be compromised. In this paper, we perform dynamic risk assessment by combining Bayesian Attack Graphs (BAGs) and online monitoring of system behavior through process mining. Specifically, the proposed approach applies process mining techniques to characterize malicious network traffic and derive evidence regarding the probability of having a vulnerability actively exploited. This evidence is then provided to a BAG, which updates its conditional probability tables accordingly, enabling dynamic assessment of vulnerability exploitation. We apply our method to a cybersecurity testbed instantiating several machines deployed on different subnets and affected by several CVE vulnerabilities. The testbed is stimulated with both benign traffic and malicious behavior, which simulates network attack patterns aimed at exploiting the CVE vulnerabilities. The results indicate that our proposal effectively detects whether vulnerabilities are being actively exploited, allowing for an updated assessment of the probability of system compromise.

Diffusion-based image synthesis has emerged as a promising source of synthetic training data for AI-based object detection and classification. In this work, we investigate whether images generated with diffusion can improve military vehicle detection under low-data conditions. We fine-tuned the text-to-image diffusion model FLUX.1 [dev] using LoRA with only 8 or 24 real images per class across 15 vehicle categories, resulting in class-specific diffusion models, which were used to generate new samples from automatically generated text prompts. The same real images were used to fine-tune the RF-DETR detector for a 15-class object detection task. Synthetic datasets generated by the diffusion models were then used to further improve detector performance. Importantly, no additional real data was required, as the generative models leveraged the same limited training samples. FLUX-generated images improved detection performance, particularly in the low-data regime (up to +8.0% mAP$_{50}$ with 8 real samples). To address the limited geometric control of text prompt-based diffusion, we additionally generated structurally guided synthetic data using ControlNet with Canny edge-map conditioning, yielding a FLUX-ControlNet (FLUX-CN) dataset with explicit control over viewpoint and pose. Structural guidance further enhanced performance when data is scarce (+4.1% mAP$_{50}$ with 8 real samples), but no additional benefit was observed when more real data is available. This study demonstrates that object-specific diffusion models are effective for improving military object detection in a low-data domain, and that structural guidance is most beneficial when real data is highly limited. These results highlight generative image data as an alternative to traditional simulation pipelines for the training of military AI systems.

AI agent systems increasingly rely on reusable non-LLM engineering infrastructure that packages tool mediation, context handling, delegation, safety control, and orchestration. Yet the architectural design decisions in this surrounding infrastructure remain understudied. This paper presents a protocol-guided, source-grounded empirical study of 70 publicly available agent-system projects, addressing three questions: which design-decision dimensions recur across projects, which co-occurrences structure those decisions, and which typical architectural patterns emerge. Methodologically, we contribute a transparent investigation procedure for analyzing heterogeneous agent-system corpora through source-code and technical-material reading. Empirically, we identify five recurring design dimensions (subagent architecture, context management, tool systems, safety mechanisms, and orchestration) and find that the corpus favors file-persistent, hybrid, and hierarchical context strategies; registry-oriented tool systems remain dominant while MCP- and plugin-oriented extensions are emerging; and intermediate isolation is common but high-assurance audit is rare. Cross-project co-occurrence analysis reveals that deeper coordination pairs with more explicit context services, stronger execution environments with more structured governance, and formalized tool-registration boundaries with broader ecosystem ambitions. We synthesize five recurring architectural patterns spanning lightweight tools, balanced CLI frameworks, multi-agent orchestrators, enterprise systems, and scenario-verticalized projects. The result provides an evidence-based account of architectural regularities in agent-system engineering, with grounded guidance for framework designers, selectors, and researchers.

Humans often hold different perspectives on the same issues. In many NLP tasks, annotation disagreement can reflect valid subjective perspectives. Modeling annotator perspectives and understanding their relationship with other human factors, such as socio-demographic attributes, have received increasing attention. Prior work typically focuses on single demographic factors or limited combinations. However, in real-world settings, annotator perspectives are shaped by complex social contexts, and finer-grained socio-demographic attributes can better explain human perspectives. In this work, we propose Socio-Contrastive Learning, a method that jointly models annotator perspectives while learning socio-demographic representations. Our method provides an effective approach for the fusion of socio-demographic features and textual representations to predict annotator perspectives, outperforming standard concatenation-based methods. The learned representations further enable analysis and visualization of how demographic factors relate to variation in annotator perspectives. Our code is available at GitHub: https://github.com/Leixin-Zhang/Socio_Contrastive_Learning

Anomaly-based Intrusion Detection Systems (IDSs) ensure protection against malicious attacks on networked systems. While deep learning-based IDSs achieve effective performance, their limited trustworthiness due to black-box architectures remains a critical constraint. Despite existing explainable techniques offering insight into the alarms raised by IDSs, they lack process-based explanations grounded in packet-level sequencing analysis. In this paper, we propose a method that employs process mining techniques to enhance anomaly-based IDSs by providing process-based alarm severity ratings and explanations for alerts. Our method prioritizes critical alerts and maintains visibility into network behavior, while minimizing disruption by allowing misclassified benign traffic to pass. We apply the method to the publicly available USB-IDS-TC dataset, which includes anomalous traffic affected by different variants of the Slowloris DoS attack. Results show that our method is able to discriminate between low- to very-high-severity alarms while preserving up to 99.94% recall and 99.99% precision, effectively discarding false positives while providing different degrees of severity for the true positives.

Human-centred systems require an understanding of human actions in the physical world. Temporally extended sequences of actions are intentional and structured, yet existing methods for recognising what actions are performed often do not attempt to capture their structure, particularly how the actions are executed. This, however, is crucial for assessing the quality of the action's execution and its differences from other actions. To capture the internal mechanics of actions, we introduce a domain-specific language EXACT that represents human motions as underspecified motion programs, interpreted as reward-generating functions for zero-shot policy inference using forward-backwards representations. By leveraging the compositional nature of EXACT motion programs, we combine individual policies into an executable neuro-symbolic model that uses program structure for compositional modelling. We evaluate the utility of the proposed pipeline for creating executable action models by analysing motion-capture data to understand human actions, for the tasks of human action segmentation and action anomaly detection. Our results show that the use of executable action models improves data efficiency and captures intuitive relationships between actions compared with monolithic, task-specific approaches.

Accurate machine-learning models for aerodynamic prediction are essential for accelerating shape optimization, yet remain challenging to develop for complex three-dimensional configurations due to the high cost of generating training data. This work introduces a methodology for efficiently constructing accurate surrogate models for design purposes by first pre-training a large-scale model on diverse geometries and then fine-tuning it with a few more detailed task-specific samples. A Transformer-based architecture, AeroTransformer, is developed and tailored for large-scale training to learn aerodynamics. The methodology is evaluated on transonic wings, where the model is pre-trained on SuperWing, a dataset of nearly 30000 samples with broad geometric diversity, and subsequently fine-tuned to handle specific wing shapes perturbed from the Common Research Model. Results show that, with 450 task-specific samples, the proposed methodology achieves 0.36% error on surface-flow prediction, reducing 84.2% compared to training from scratch. The influence of model configurations and training strategies is also systematically studied to provide guidance on effectively training and deploying such models under limited data and computational budgets. To facilitate reuse, we release the datasets and the pre-trained models at https://github.com/tum-pbs/AeroTransformer. An interactive design tool is also built on the pre-trained model and is available online at https://webwing.pbs.cit.tum.de.

We introduce Sonata, a compact latent world model for six-axis trunk IMU representation learning under clinical data scarcity. Clinical cohorts typically comprise tens to hundreds of patients, making web-scale masked-reconstruction objectives poorly matched to the problem. Sonata is a 3.77 M-parameter hybrid model, pre-trained on a harmonised corpus of nine public datasets (739 subjects, 190k windows) with a latent world-model objective that predicts future state rather than reconstructing raw sensor traces. In a controlled comparison against a matched autoregressive forecasting baseline (MAE) on the same backbone, Sonata yields consistently stronger frozen-probe clinical discrimination, prospective fall-risk prediction, and cross-cohort transfer across a 14-arm evaluation suite, while producing higher-rank, more structured latent representations. At 3.77 M parameters the model is compatible with on-device wearable inference, offering a step toward general kinematic world models for neurological assessment.

Fairness issues often remain hidden in digital systems, making them difficult to detect and even more difficult to address. In this study, we introduce a fairness-first Design Thinking (DT) approach to support addressing fairness concerns in software architecture (SA) design. We implemented our approach in a graduate-level course where students executed all steps of our DT approach as part of an assignment. We analyzed the assignment data to reflect on the implications for applying the DT approach in SA and teaching the DT approach in SA education. As a result of this study, we provide (i) a DT approach for SA, (ii) implications of the DT approach on handling fairness in both problem and solution spaces, and (iii) implications for education. Our reflections highlight that fairness theory and context identification are essential for a holistic, fairness-first design. We propose the use of composite views to address cross-cutting concerns such as fairness. In the future, we will update the course material to provide end-to-end fairness traceability in SA, helping students to understand how fairness concerns can be translated into actionable design decisions.

Intrusion detection systems (IDSs) for 5G networks must handle complex, high-volume traffic. Although opaque "black-box" models can achieve high accuracy, their lack of transparency hinders trust and effective operational response. We propose ExAI5G, a framework that prioritizes interpretability by integrating a Transformer-based deep learning IDS with logic-based explainable AI (XAI) techniques. The framework uses Integrated Gradients to attribute feature importance and extracts a surrogate decision tree to derive logical rules. We introduce a novel evaluation methodology for LLM-generated explanations, using a powerful evaluator LLM to assess actionability and measuring their semantic similarity and faithfulness. On a 5G IoT intrusion dataset, our system achieves 99.9\% accuracy and a 0.854 macro F1-score, demonstrating strong performance. More importantly, we extract 16 logical rules with 99.7\% fidelity, making the model's reasoning transparent. The evaluation demonstrates that modern LLMs can generate explanations that are both faithful and actionable, indicating that it is possible to build a trustworthy and effective IDS without compromising performance for the sake of marginal gains from an opaque model.

AlphaGeometry represents a milestone in neuro-symbolic reasoning, yet its architecture faces a log-linear scaling bottleneck within its symbolic deduction engine that limits its efficiency as problem complexity increases. Recent technical reports suggest that current domain-specific languages may be isomorphic as input representations to natural language, interchanging them acts as a performance-invariant transformation, implying that current neural guidance relies on superficial encodings rather than structural understanding. This paper addresses this representation bottleneck by proposing a logic-to-topology encoding designed to reveal the structural invariants of a model's latent space under a transformation of its input space. By leveraging the Logic of Observation, we utilize the duality between provability in observable theories and topologies to propose a logic-to-topology encoder for the input space. We introduce the concept of the "topological dual of a dataset", a transformation that bridges formal logic, topology, and neural processing. This framework serves as a Rosetta Stone for neuro-symbolic AI, providing a principled pathway for the mechanistic interpretability of how models navigate complex discovery paths.

Despite significant advances in large language models, personalizing them for individual decision-makers remains an open problem. Here, we introduce a synthetic-organic supervision pipeline that transforms raw judicial decisions into instruction-tuning data, enabling parameter-efficient fine-tuning of personalized models for individual judges in low-resource settings. We compare our approach to state-of-the-art personalization techniques across three different tasks and settings. The results show that Causal Language Modeling followed by synthetically generated instruction-tuning significantly outperforms all other baselines, providing significant improvements across lexical, stylistic, and semantic similarity. Notably, our model-generated outputs are indistinguishable from the reasoning of human judges, highlighting the viability of efficient personalization, even in low-resource settings.

Large language models (LLMs) are increasingly used in clinical settings, raising concerns about racial bias in both generated medical text and clinical reasoning. Existing studies have identified bias in medical LLMs, but many focus on single models and give less attention to mitigation. This study uses the EU AI Act as a governance lens to evaluate five widely used LLMs across two tasks, namely synthetic patient-case generation and differential diagnosis ranking. Using race-stratified epidemiological distributions in the United States and expert differential diagnosis lists as benchmarks, we apply structured prompt templates and a two-part evaluation design to examine implicit and explicit racial bias. All models deviated from observed racial distributions in the synthetic case generation task, with GPT-4.1 showing the smallest overall deviation. In the differential diagnosis task, DeepSeek V3 produced the strongest overall results across the reported metrics. When embedded in an agentic workflow, DeepSeek V3 showed an improvement of 0.0348 in mean p-value, 0.1166 in median p-value, and 0.0949 in mean difference relative to the standalone model, although improvement was not uniform across every metric. These findings support multi-metric bias evaluation for AI systems used in medical settings and suggest that retrieval-based agentic workflows may reduce some forms of explicit bias in benchmarked diagnostic tasks. Detailed prompt templates, experimental datasets, and code pipelines are available on our GitHub.

Machine learning (ML) models trained to detect physical-layer threats on one optical fiber system often fail catastrophically when applied to a different system, due to variations in operating wavelength, fiber properties, and network architecture. To overcome this, we propose a Domain Adaptation (DA) framework based on a Variational Autoencoder (VAE) that learns a shared representation capturing event signatures common to both systems while suppressing system-specific differences. The shared encoder is first trained on the combined data from two distinct optical systems: a 21 km O-band dark-fiber testbed (System 1) and a 63.4 km C-band live metro ring (System 2). The encoder is then frozen, and a classifier is trained using labels from an individual system. The proposed approach achieves 95.3% and 73.5% cross-system accuracy when moving from System 1 to System 2 and vice versa, respectively. This corresponds to gains of 83.4% and 51% over a fully supervised Deep Neural Network (DNN) baseline trained on a single system, while preserving intra-system performance.

We present SignDPO, a novel multi-level Direct Preference Optimisation (DPO) framework designed to enhance the alignment of skeleton-based Sign Language Translation. While current skeleton-based models have made significant progress using Maximum Likelihood Estimation, they are primarily constrained by an imitation-based paradigm that lacks discriminative sensitivity to the fine-grained spatio-temporal nuances of sign language, often leading to semantic drift. To address this, SignDPO shifts the optimisation goal from simple sequence mimicry to structured preference alignment across spatial, temporal, and linguistic dimensions. Our framework involves three key designs. First, we introduce a hierarchical perturbation strategy to construct spatial and temporal non-preferred samples at both global and local granularities automatically. Second, we propose a self-guiding mechanism that leverages decoder cross-attention scores to identify and perturb semantically salient skeletal regions, forcing the model to distinguish genuine sign signals from structural distortions. Third, we establish an automated language-level preference generator by fine-tuning a dedicated perturbation model, capturing complex output-level failure modes without manual annotation. Extensive experiments on three widely adopted benchmarks, CSL-Daily, How2Sign, and OpenASL, demonstrate that SignDPO consistently outperforms state-of-the-art gloss-free methods and even rivals established gloss-based ones. Our results suggest that multi-level preference alignment is a powerful paradigm for bridging the gap between high-entropy skeletal trajectories and discrete linguistic semantics.

Molecule generation requires satisfying multiple chemical and biological constraints while searching a large and structured chemical space. This makes it a non-binary problem, where effective models must identify non-obvious solutions under constraints while maintaining exploration to improve success by escaping local optima. From this perspective, creativity is a functional requirement in molecular generation rather than an aesthetic notion. Large language models (LLMs) can generate molecular representations directly from natural language prompts, but it remains unclear what type of creativity they exhibit in this setting and how it should be evaluated. In this work, we study the creative behavior of LLMs in molecular generation through a systematic empirical evaluation across physicochemical, ADMET, and biological activity tasks. We characterize creativity along two complementary dimensions, convergent creativity and divergent creativity, and analyze how different factors shape these behaviors. Our results indicate that LLMs exhibit distinct patterns of creative behavior in molecule generation, such as an increase in constraint satisfaction when additional constraints are imposed. Overall, our work is the first to reframe the abilities required for molecule generation as creativity, providing a systematic understanding of creativity in LLM-based molecular generation and clarifying the appropriate use of LLMs in molecular discovery pipelines.

Transpilation, or code translation, aims to convert source code from one programming language (PL) to another. It is beneficial for many downstream applications, from modernizing large legacy codebases to augmenting data for low-resource PLs. Recent large language model (LLM)-based approaches have demonstrated immense potential for code translation. Among these approaches, training-based methods are particularly important because LLMs currently do not effectively adapt to domain-specific settings that suffer from a lack of knowledge without targeted training. This limitation is evident in transpilation tasks involving low-resource PLs. However, existing training-based approaches rely on a pairwise transpilation paradigm, making it impractical to support a diverse range of PLs. This limitation is particularly prominent for low-resource PLs due to a scarcity of training data. Furthermore, these methods suffer from suboptimal reinforcement learning (RL) reward formulations. To address these limitations, we propose CodePivot, a training framework that leverages Python as an intermediate representation (IR), augmented by a novel RL reward mechanism, Aggressive-Partial-Functional reward, to bootstrap the model's multilingual transpilation ability without requiring parallel corpora. Experiments involving 10 PLs show that the resulting 7B model, trained on Python-to-Others tasks, consistently improves performance across both general and low-resource PL-related transpilation tasks. It outperforms substantially larger mainstream models with hundreds of billions more parameters, such as Deepseek-R1 and Qwen3-235B-A22B-Instruct-2507, on Python-to-Others tasks and Others-to-All tasks, respectively. In addition, it outperforms its counterpart trained directly on Any-to-Any tasks on general transpilation tasks. The code and data are available at https://github.com/lishangyu-hkust/CodePivot.

Many deployed systems expose black-box objectives whose minimizing configuration shifts with an externally observed context. When contexts revisit a small set of latent regimes, an optimizer that discards history pays repeated adaptation cost; when each step must remain inexpensive, full Gaussian-process (GP) refits at high observation counts are difficult to sustain. We cast online tuning as context-conditioned regret minimization and present RASP-Tuner, which instantiates a decomposition motivated by first principles: (i) identify a regime proxy by retrieving similar past contexts; (ii) predict short-horizon loss with a mixture-of-experts surrogate whose input concatenates parameters, context, and a retrieved soft prompt; (iii) adapt chiefly in a low-dimensional prompt subspace, invoking full surrogate updates only when scalarized error or disagreement spikes. A RealErrorComposer maps heterogeneous streaming metrics to [0,1] via EMA-stabilized logistic scores, supplying a single differentiable training target. On nine synthetic non-stationary benchmarks, an adversarial-context sanity check, and three tabular real-world streams (Section on real-world experiments), RASP-Tuner improves or matches cumulative regret relative to our GP-UCB and CMA-ES implementations on seven of nine synthetic tasks under paired tests at horizon T=100, while recording 8-12 times lower wall-clock per step than sliding-window GP-UCB on identical hardware. Idealized analysis in a cluster-separated, strongly convex regime model (RA-GD) supplies sufficient conditions for bounded dynamic regret; the deployed pipeline violates several of these premises, and we articulate which gaps remain open.

In multi-view clustering, the quality of different views may vary substantially, and low-quality or degraded views can impair overall clustering performance. However, existing studies mainly address this issue within the clustering process through view weighting or noise-robust optimization, while paying limited attention to data-level assessment before clustering. In this paper, we study the problem of pre-clustering noisy-view analysis in multi-view data from a clusterability perspective. To this end, we propose a Multi-View Clusterability Score (MVCS), which quantifies the strength of latent cluster-related structures in multi-view data through three complementary components: per-view structural clusterability, joint-space clusterability, and cross-view neighborhood consistency. To the best of our knowledge, this is the first clusterability score specifically designed for multi-view data. We further use it to perform potentially noisy view analysis and noisy-view detection before clustering. Extensive experiments on real-world datasets demonstrate that noisy views can significantly degrade clustering performance, and that, compared with existing clusterability measures designed for single-view data, the proposed method more effectively supports noisy-view analysis and detection.

The inverse Potts problem for estimating evolutionary single-site fields and pairwise couplings in homologous protein sequences from their single-site and pairwise amino acid frequencies observed in their multiple sequence alignment would be still one of useful methods in the studies of protein structure and evolution. Since the reproducibility of fields and couplings are the most important, the Boltzmann machine method is employed here, although it is computationally intensive. In order to reduce computational time required for the Boltzmann machine, parallel, persistent Markov chain Monte Carlo method is employed to estimate the single-site and pairwise marginal distributions in each learning step. Also, stochastic gradient descent methods are used to reduce computational time for each learning. Another problem is how to adjust the values of hyperparameters; there are two regularization parameters for evolutionary fields and couplings. The precision of contact residue pair prediction is often used to adjust the hyperparameters. However, it is not sensitive to these regularization parameters. Here, they are adjusted for the fields and couplings to satisfy a specific condition that is appropriate for protein conformations. This method has been applied to eight protein families.

We prove error bounds for operator surrogates of solution operators for partial differential and boundary integral equations on families of domains which are diffeomorphic to one common reference (or latent) domain $D_{ref}$. The pullback of the PDE to $D_{ref}$ via affine-parametric shape encoding produces a collection of holomorphic parametric PDEs on $D_{ref}$. Sufficient conditions for (uniformly with respect to the parameter) well-posedness are given, implying existence, uniqueness and stability of parametric solution families on $D_{ref}$. We illustrate the abstract hypotheses by reviewing recent holomorphy results for a suite of elliptic and parabolic PDEs. Quantified parametric holomorphy implies existence of finite-parametric, discrete approximations of the parametric solution families with convergence rates in terms of the number $N$ of parameters. We obtain constructive proofs of existence of Neural and Spectral Operator surrogates for the shape-to-solution maps with error bounds and convergence rate guarantees uniform on the collection of admissible shapes. We admit principal-component shape encoders and frame decoders. Our results support in particular the (empirically reported) ability of neural operators to realize data-to-solution maps for elliptic and parabolic PDEs and BIEs that generalize across parametric families of shapes.

Multi-agent systems (MAS) are increasingly used for open-ended idea generation, driven by the expectation that collective interaction will broaden the exploration diversity. However, when and why such collaboration truly expands the solution space remains unclear. We present a systematic empirical study of diversity in MAS-based ideation across three bottom-up levels: model intelligence, agent cognition, and system dynamics. At the model level, we identify a compute efficiency paradox, where stronger, highly aligned models yield diminishing marginal diversity despite higher per-sample quality. At the cognition level, authority-driven dynamics suppress semantic diversity compared to junior-dominated groups. At the system level, group-size scaling yields diminishing returns and dense communication topologies accelerate premature convergence. We characterize these outcomes as collective failures emerging from structural coupling, a process where interaction inadvertently contracts agent exploration and triggers diversity collapse. Our analysis shows that this collapse arises primarily from the interaction structure rather than inherent model insufficiency, highlighting the importance of preserving independence and disagreement when designing MAS for creative tasks. Our code is available at https://github.com/Xtra-Computing/MAS_Diversity.

Emotion Recognition in Conversation (ERC) has become a fundamental capability for large language models (LLMs) in human-centric interaction. Beyond accurate recognition, coherent emotional expression is also crucial, yet both are limited by the scarcity and static nature of high-quality annotated data. In this work, we propose SELF-EMO, a self-evolution framework grounded in the hypothesis that better emotion prediction leads to more consistent emotional responses. We introduce two auxiliary tasks, emotional understanding and emotional expression, and design a role-based self-play paradigm where the model acts as both an emotion recognizer and a dialogue responder. Through iterative interactions, the model generates diverse conversational trajectories, enabling scalable data generation. To ensure quality, we adopt a data flywheel mechanism that filters candidate predictions and responses using a smoothed IoU-based reward and feeds selected samples back for continuous self-improvement without external supervision. We further develop SELF-GRPO, a reinforcement learning algorithm that stabilizes optimization with multi-label alignment rewards and group-level consistency signals. Experiments on IEMOCAP, MELD, and EmoryNLP show that SELF-EMO achieves state-of-the-art performance, improving accuracy by +6.33% on Qwen3-4B and +8.54% on Qwen3-8B, demonstrating strong effectiveness and generalization.

Chain-of-thought reasoning has driven striking advances in language model capability, yet every reasoning step grows the KV cache, creating a bottleneck to scaling this paradigm further. Current approaches manage these constraints on the model's behalf using hand-designed criteria. A more scalable approach would let end-to-end learning subsume this design choice entirely, following a broader pattern in deep learning. After all, if a model can learn to reason, why can't it learn to forget? We introduce Neural Garbage Collection (NGC), in which a language model learns to forget while learning to reason, trained end-to-end from outcome-based task reward alone. As the model reasons, it periodically pauses, decides which KV cache entries to evict, and continues to reason conditioned on the remaining cache. By treating tokens in a chain-of-thought and cache-eviction decisions as discrete actions sampled from the language model, we can use reinforcement learning to jointly optimize how the model reasons and how it manages its own memory: what the model evicts shapes what it remembers, what it remembers shapes its reasoning, and the correctness of that reasoning determines its reward. Crucially, the model learns this behavior entirely from a single learning signal - the outcome-based task reward - without supervised fine-tuning or proxy objectives. On Countdown, AMC, and AIME tasks, NGC maintains strong accuracy relative to the full-cache upper bound at 2-3x peak KV cache size compression and substantially outperforms eviction baselines. Our results are a first step towards a broader vision where end-to-end optimization drives both capability and efficiency in language models.

Anomaly detection in multivariate time series is a central challenge in industrial monitoring, as failures frequently arise from complex temporal dynamics and cross-sensor interactions. While recent deep learning models, including graph neural networks and Transformers, have demonstrated strong empirical performance, most approaches remain primarily correlational and offer limited support for causal interpretation and root-cause localization. This study introduces a causally-constrained probabilistic forecasting framework which is a Causally Guided Transformer (CGT) model for multivariate time-series anomaly detection, integrating an explicit time-lagged causal graph prior with deep sequence modeling. For each target variable, a dedicated forecasting block employs a hard parent mask derived from causal discovery to restrict the main prediction pathway to graph-supported causes, while a latent Gaussian head captures predictive uncertainty. To leverage residual correlational information without compromising the causal representation, a shadow auxiliary path with stop-gradient isolation and a safety-gated blending mechanism is incorporated to suppress non-causal contributions when reliability is low. Anomalies are identified using negative log-likelihood scores with adaptive streaming thresholding, and root-cause variables are determined through per-dimension probabilistic attribution and counterfactual clamping. Experiments on the ASD and SMD benchmarks indicate that the proposed method achieves state-of-the-art detection performance, with F1-scores of 96.19% on ASD and 95.32% on SMD, and enhances variable-level attribution quality. These findings suggest that causal structural priors can improve both robustness and interpretability in detecting deep anomalies in multivariate sensor systems.

What does it mean to give an AI agent a complete education? Current agent development produces specialists systems optimized for a single capability dimension, whether tool use, code generation, or security awareness that exhibit predictable deficits wherever they were not trained. We argue this pattern reflects a structural absence: there is no curriculum theory for agents, no principled account of what a fully developed agent should know, be, and be able to do across the full scope of intelligent behavior. This paper introduces the AIT Academy (Agents Institute of Technology Academy), a curriculum framework for cultivating AI agents across the tripartite structure of human knowledge. Grounded in Kagan's Three Cultures and UNESCO ISCED-F 2013, AIT organizes agent capability development into three domains: Natural Science and Technical Reasoning (Domain I), Humanities and Creative Expression (Domain II), and Social Science and Ethical Reasoning (Domain III). The Confucian Six Arts (liuyi) a 2,500-year-old holistic education system are reinterpreted as behavioral archetypes that map directly onto trainable agent capabilities within each domain. Three representative training grounds instantiate the framework across multiple backbone LLMs: the ClawdGO Security Dojo (Domain I), Athen's Academy (Domain II), and the Alt Mirage Stage (Domain III). Experiments demonstrate a 15.9-point improvement in security capability scores under weakest-first curriculum scheduling, and a 7-percentage-point gain in social reasoning performance under principled attribution modeling. A cross-domain finding Security Awareness Calibration Pathology (SACP), in which over-trained Domain I agents fail on out-of-distribution evaluation illustrates the diagnostic value of a multi-domain perspective unavailable to any single-domain framework.

Background: The potential of large language models (LLMs) to automate and support pharmacoepidemiologic study design is an emerging area of interest, yet their reliability remains insufficiently characterized. General-purpose LLMs often display inaccuracies, while the comparative performance of specialized biomedical LLMs in this domain remains unknown. Methods: This study evaluated general-purpose LLMs (GPT-4o and DeepSeek-R1) versus biomedically fine-tuned LLMs (QuantFactory/Bio-Medical-Llama-3-8B-GGUF and Irathernotsay/qwen2-1.5B-medical_qa-Finetune) using 46 protocols (2018-2024) from the HMA-EMA Catalogue and Sentinel System. Performance was assessed across relevance, logic of justification, and ontology-code agreement across multiple coding systems using Least-to-Most (LTM) and Active Prompting strategies. Results: GPT-4o and DeepSeek-R1 paired with LTM prompting achieved the highest relevance and logic of justification scores, with GPT-4o-LTM reaching a median relevance score of 4 in 8 of 9 questions for HMA-EMA protocols. Biomedical LLMs showed lower relevance overall and frequently generated insufficient justification. All LLMs demonstrated limited proficiency in ontology-code mapping, although LTM provided the most consistent improvements in reasoning stability. Conclusion: Off-the-shelf general-purpose LLMs currently offer superior support for pharmacoepidemiologic design compared to biomedical LLMs. Prompt strategy strongly influenced LLM performance.

We introduce the Latent Fourier Transform (LatentFT), a framework that provides novel frequency-domain controls for generative music models. LatentFT combines a diffusion autoencoder with a latent-space Fourier transform to separate musical patterns by timescale. By masking latents in the frequency domain during training, our method yields representations that can be manipulated coherently at inference. This allows us to generate musical variations and blends from reference examples while preserving characteristics at desired timescales, which are specified as frequencies in the latent space. LatentFT parallels the role of the equalizer in music production: while traditional equalizers operates on audible frequencies to shape timbre, LatentFT operates on latent-space frequencies to shape musical structure. Experiments and listening tests show that LatentFT improves condition adherence and quality compared to baselines. We also present a technique for hearing frequencies in the latent space in isolation, and show different musical attributes reside in different regions of the latent spectrum. Our results show how frequency-domain control in latent space provides an intuitive, continuous frequency axis for conditioning and blending, advancing us toward more interpretable and interactive generative music models.

Uncertainty quantification is crucial in safety-critical systems, where decisions must be made under uncertainty. In particular, we consider the problem of online uncertainty quantification, where data points arrive sequentially. Online conformal prediction is a principled online uncertainty quantification method that dynamically constructs a prediction set at each time step. While existing methods for online conformal prediction provide long-run coverage guarantees without any distributional assumptions, they typically assume a full feedback setting in which the true label is always observed. In this paper, we propose a novel learning method for online conformal prediction with partial feedback from an adaptive adversary-a more challenging setup where the true label is revealed only when it lies inside the constructed prediction set. Specifically, we formulate online conformal prediction as an adversarial bandit problem by treating each candidate prediction set as an arm. Building on an existing algorithm for adversarial bandits, our method achieves a long-run coverage guarantee by explicitly establishing its connection to the regret of the learner. Finally, we empirically demonstrate the effectiveness of our method in both independent and identically distributed (i.i.d.) and non-i.i.d. settings, showing that it successfully controls the miscoverage rate while maintaining a reasonable size of the prediction set.

Large Vision-Language Models (LVLMs) still struggle with vision hallucination, where generated responses are inconsistent with the visual input. Existing methods either rely on large-scale annotated data for fine-tuning, which incurs massive computational overhead, or employ static post-hoc strategies that overlook the dynamic nature of hallucination emergence. To address these, we introduce a new self-rewarding framework, enabling dynamic hallucination mitigation at inference time without external supervision. On the empirical side, we reveal that visual hallucination exhibits phase-wise dynamic patterns, peaking at the onset of each semantic phase. Drawing on these insights, we propose \textbf{PSRD} (\textbf{Phase-wise \textbf{S}elf-\textbf{R}eward \textbf{D}ecoding) for online hallucination correction guided by phase-wise self-reward signals. To reduce the cost of repeated self-evaluation during decoding, we distill the hallucination guidance signal from LVLMs into a lightweight reward model. The reward model subsequently provides on-the-fly guidance for targeted intervention during the decoding process, enabling precise hallucination suppression. The proposed PSRD significantly reduces the hallucination rate of LLaVA-1.5-7B by 50.0% and consistently outperforms existing post-hoc methods across five hallucination evaluation benchmarks for four LVLMs. Further analysis confirms that PSRD effectively mitigates hallucination propagation and achieves a highly controllable trade-off between strong performance and inference efficiency.

Fuzz testing of software libraries relies on fuzz drivers to invoke library APIs. Traditionally, these drivers are written manually by developers - a process that is time-consuming and often inadequate for exercising complex program behaviors. While recent studies have explored the use of Large Language Models (LLMs) to automate fuzz driver generation, the resulting drivers often fail to cover deep program branches. To address these challenges, we propose MASFUZZER, a fuzzing framework that integrates multidimensional API sequence construction with adaptive fuzzing scheduling strategies to improve library testing. At its core, MASFUZZER synthesizes context-relevant API call sequences by referring to API usage examples from the codebase and applying mutation-propagation-based and semantic-aware API sequence mining. These multidimensional API sequences serve as the basis for LLMs to generate effective initial drivers. In addition, MASFUZZER incorporates a coverage-guided scheduler that prioritizes testing time for the most promising drivers, along with a driver mutation strategy to evolve them. This enables systematic generation of fuzz drivers to explore previously untested code regions. We evaluate MASFUZZER on 12 widely used open-source libraries. The results show that MASFUZZER achieves 8.54 percent higher code coverage than state-of-the-art techniques. Moreover, MASFUZZER uncovers 16 previously unknown vulnerabilities in extensively tested libraries, with 14 confirmed by developers and 9 assigned CVE identifiers. These results indicate that MASFUZZER provides an efficient and practical approach for fuzzing software libraries.

This paper presents ltzGLUE, the first Natural Language Understanding (NLU) benchmark for Luxembourgish (LTZ) based on the popular GLUE benchmark for English. Although NLU tasks are available for many European languages nowadays, LTZ is one of the official national languages that is often overlooked. We construct new tasks and reuse existing ones to introduce the first official NLU benchmark and accompanying evaluation of encoder models for the language. Our tasks include common natural language processing tasks in binary and multi-class classification settings, including named entity recognition, topic classification, and intent classification. We evaluate various pre-trained language models for LTZ to present an overview of the current capabilities of these models on the LTZ language.

Emotional Support Conversation (ESC) aims to assist individuals experiencing distress by generating empathetic and supportive dialogue. While prior work typically assumes that each supporter turn corresponds to a single strategy, real-world supportive communication often involves multiple strategies within a single utterance. In this paper, we revisit the ESC task by formulating it as multi-strategy utterance generation, where each utterance may contain one or more strategy-response pairs. We propose two generation methods: All-in-One, which predicts all strategy-response pairs in a single decoding step, and One-by-One, which iteratively generates strategy-response pairs until completion. Both methods are further enhanced with cognitive reasoning guided by reinforcement learning to improve strategy selection and response composition. We evaluate our models on the ESConv dataset under both utterance-level and dialogue-level settings. Experimental results show that our methods effectively model multi-strategy utterances and lead to improved supportive quality and dialogue success. To our knowledge, this work provides the first systematic empirical evidence that allowing multiple support strategies within a single utterance is both feasible and beneficial for emotional support conversations. All code and data will be publicly available at https://github.com/aliyun/qwen-dianjin.

Although large language models (LLMs) are increasingly used as annotators at scale, they are typically treated as a pragmatic fallback rather than a faithful estimator of human perspectives. This work challenges that presumption. By framing perspective-taking as the estimation of a latent group-level judgment, we characterize the conditions under which modern LLMs can outperform human annotators, including in-group humans, when predicting aggregate subgroup opinions on subjective tasks, and show that these conditions are common in practice. This advantage arises from structural properties of LLMs as estimators, including low variance and reduced coupling between representation and processing biases, rather than any claim of lived experience. Our analysis identifies clear regimes where LLMs act as statistically superior frontline estimators, as well as principled limits where human judgment remains essential. These findings reposition LLMs from a cost-saving compromise to a principled tool for estimating collective human perspectives.

In this article, we establish a precise connection between binarized neural networks (BNNs) and Sugeno integrals. The advantage of the Sugeno integral is that it provides a framework for representing the importance of inputs and their interactions, while being equivalent to a set of if-then rules. For a hidden BNN neuron at inference time, we show that the activation threshold test can be written as a Sugeno integral on binary inputs. This yields an explicit set-function representation of each neuron decision, and an associated rule-based representation. We also provide a Sugeno-integral expression for the last-layer score. Finally, we discuss how the same framework can be adapted to support richer input interactions and how it can be extended beyond the binary case induced by binarized neural networks.

Deploying LLMs as reasoning assistants in safety-critical aerospace engineering requires stricter evaluation criteria than general scientific benchmarks. In hypersonic thermal protection system (TPS) design, inaccurate stagnation-point heat flux or boundary-layer calculations may cause catastrophic design margin violations. Models with numerically reasonable but physically invalid answers are more dangerous than those declining to respond. Current scientific benchmarks only test abstract math and basic physics, evaluate final answers solely, ignore engineering reasoning processes, and cannot detect such critical failures. We propose TPS-CalcBench, the first diagnostic benchmark for closed-form analytical calculations in hypersonic aerodynamics and high-temperature gas dynamics that experienced TPS engineers conduct without simulations. Our contributions include domain-oriented task taxonomy with 4 difficulty levels and 8 categories from Anderson's textbook, dual-track evaluation measuring result accuracy and reasoning quality via an 8-dimension rubric and calibrated judge with human audit to identify right answer wrong reasoning issues, human-AI data pipeline producing 420 high-confidence core items and 810 noise-controlled pre-gating items from 4560 raw data, noise-sensitivity analysis measuring data quality impacts on model ranking, and three diagnostic intervention methods: DFA-TPS fine-tuning, RAG-EQ retrieval grounding and PA-CoT process-aware prompting. Tests on 13 models from 7 groups show wide performance differences (KPI 12.6-87.9), hidden formula selection defects, data-driven rank changes and effective intervention improvements, establishing a complete diagnose-evaluate-intervene framework for safety-critical engineering LLM deployment assessment.

Recent studies reveal striking representational alignment between artificial neural networks (ANNs) and biological brains, leading to proposals that all sufficiently capable systems converge on universal representations of reality. Here, we argue that this claim of Universality is premature. We introduce the Umwelt Representation Hypothesis (URH), proposing that alignment arises not from convergence toward a single global optimum, but from overlap in ecological constraints under which systems develop. We review empirical evidence showing that representational differences between species, individuals, and ANNs are systematic and adaptive, which is difficult to reconcile with Universality. Finally, we reframe ANN model comparison as a method for mapping clusters of alignment in ecological constraint space rather than searching for a single optimal world model.

Process Reward Models (PRMs) have emerged as a powerful tool for providing step-level feedback when evaluating the reasoning of Large Language Models (LLMs), which frequently produce chains of thought (CoTs) containing errors even when the final answer is correct. However, existing PRM datasets remain expensive to construct, prone to annotation errors, and predominantly limited to the mathematical domain. This work introduces a novel and scalable approach to PRM dataset generation based on planning logical problems expressed in the Planning Domain Definition Language (PDDL). Using this method, we generate a corpus of approximately one million reasoning steps across various PDDL domains and use it to train PRMs. Experimental results show that augmenting widely-used PRM training datasets with PDDL-derived data yields substantial improvements in both mathematical and non-mathematical reasoning, as demonstrated across multiple benchmarks. These findings indicate that planning problems constitute a scalable and effective resource for generating robust, precise, and fine-grained training data for PRMs, going beyond the classical mathematical sources that dominate this field.

Machine learning has become integral to medical research and is increasingly applied in clinical settings to support diagnosis and decision-making; however, its effectiveness depends on access to large, diverse datasets, which are limited within single institutions. Although integrating data across institutions can address this limitation, privacy regulations and data ownership constraints hinder these efforts. Federated learning enables collaborative model training without sharing raw data; however, most methods rely on complex architectures that lack interpretability, limiting clinical applicability. Therefore, we proposed a federated RuleFit framework to construct a unified and interpretable global model for distributed environments. It integrates three components: preprocessing based on differentially private histograms to estimate shared cutoff values, enabling consistent rule definitions and reducing heterogeneity across clients; local rule generation using gradient boosting decision trees with shared cutoffs; and coefficient estimation via $\ell_1$-regularized optimization using a Federated Dual Averaging algorithm for sparse and consistent variable selection. In simulation studies, the proposed method achieved a performance comparable to that of centralized RuleFit while outperforming existing federated approaches. Real-world analysis demonstrated its ability to provide interpretable insights with competitive predictive accuracy. Therefore, the proposed framework offers a practical and effective solution for interpretable and reliable modeling in federated learning environments.

We develop interconnections between the complex normalizing flow for data drawn from Borel probability measures on the twofold realification of the complex manifold and the Kähler-Ricci flow. The complex normalizing flow relates the initial and target realified densities under the complex change of variables, necessitating the log determinant of the Wirtinger Jacobian. The Ricci curvature of a Kähler manifold is the second order mixed Wirtinger partial derivative of the log of the local density of the volume form. Therefore, we reconcile these two facts by drawing forth the connection that the log determinant used in the complex normalizing flow matches the Ricci curvature term under differentiation and conditions. The log density under the normalizing flow is kindred to a spatial Fisher information metric under a holomorphic pullback and a Bayesian perspective to the parameter, thus under the continuum limit the log likelihood matches a Fisher metric, recovering the Kähler-Ricci flow up to expectation. Using this framework, we establish other relevant results, attempting to bridge the statistical and ordinary behaviors of the complex normalizing flow to the geometric features of the Kähler-Ricci flow.

We revisit multi-agent delegation under a stronger and more realistic assumption: an agent's capability is not fixed at the skill level, but depends on task context. A coding agent may excel at short standalone edits yet fail on long-horizon debugging; a planner may perform well on shallow tasks yet degrade on chained dependencies. Static skill-level capability profiles therefore average over heterogeneous situations and can induce systematic misdelegation. We propose CADMAS-CTX, a framework for contextual capability calibration. For each agent, skill, and coarse context bucket, CADMAS-CTX maintains a Beta posterior that captures stable experience in that part of the task space. Delegation is then made by a risk-aware score that combines the posterior mean with an uncertainty penalty, so that agents delegate only when a peer appears better and that assessment is sufficiently well supported by evidence. This paper makes three contributions. First, a hierarchical contextual capability profile replaces static skill-level confidence with context-conditioned posteriors. Second, based on contextual bandit theory, we formally prove context-aware routing achieves lower cumulative regret than static routing under sufficient context heterogeneity, formalizing the bias-variance tradeoff. Third, we empirically validate our method on GAIA and SWE-bench benchmarks. On GAIA with GPT-4o agents, CADMAS-CTX achieves 0.442 accuracy, outperforming static baseline 0.381 and AutoGen 0.354 with non-overlapping 95% confidence intervals. On SWE-bench Lite, it improves resolve rate from 22.3% to 31.4%. Ablations show the uncertainty penalty improves robustness against context tagging noise. Our results demonstrate contextual calibration and risk-aware delegation significantly improve multi-agent teamwork compared with static global skill assignments.

Large Language Models (LLMs) have demonstrated remarkable capabilities across various cybersecurity tasks, including vulnerability classification, detection, and patching. However, their potential in automated vulnerability report documentation and analysis remains underexplored. We present RAVEN (Retrieval Augmented Vulnerability Exploration Network), a framework leveraging LLM agents and Retrieval Augmented Generation (RAG) to synthesize comprehensive vulnerability analysis reports. Given vulnerable source code, RAVEN generates reports following the Google Project Zero Root Cause Analysis template. The framework uses four modules: an Explorer agent for vulnerability identification, a RAG engine retrieving relevant knowledge from curated databases including Google Project Zero reports and CWE entries, an Analyst agent for impact and exploitation assessment, and a Reporter agent for structured report generation. To ensure quality, RAVEN includes a task specific LLM Judge evaluating reports across structural integrity, ground truth alignment, code reasoning quality, and remediation quality. We evaluate RAVEN on 105 vulnerable code samples covering 15 CWE types from the NIST-SARD dataset. Results show an average quality score of 54.21%, supporting the effectiveness of our approach for automated vulnerability documentation.

Developing agents capable of navigating fragmented, multi-source information remains challenging, primarily due to the scarcity of benchmarks reflecting hybrid workflows combining database querying with external APIs. To bridge this gap, we introduce ReCoQA, a large-scale benchmark of 29,270 real-estate instances featuring machine-verifiable supervision for intermediate steps, including structured intent labels, SQL queries, and API calls. Complementarily, we propose HIRE-Agent, a hierarchical framework instantiating an understand-plan-execute architecture as a strong baseline. By orchestrating a Front-end parser, a planning Supervisor, and execution Specialists, HIRE-Agent effectively integrates heterogeneous evidence. Extensive experiments demonstrate that HIRE-Agent constitutes a strong baseline and substantiates the necessity of hierarchical collaboration for complex, real-world reasoning tasks.

Open-domain RAG benchmarks over public corpora can overestimate deployment performance due to pretraining overlap and weak attribution requirements. We present DoRA (Domain-oriented RAG Assessment), a domain-grounded benchmark built from defense documents that pairs synthetic, intent-conditioned QA (question answering) with auditable evidence passages for attribution. DoRA covers five question types (find, explain, summarize, generate, provide) and contains 6.5K curated instances. In end-to-end evaluation with a fixed dense retriever, general-purpose Language Models (LMs) perform similarly, while a model trained on DoRA (DoRA SFT) yields large gains over the base model (Llama3.1-8B-Instruct): up to 26% improvement in QA task success, while reducing the hallucination rate by 47% in RAG faithfulness scores, supporting contamination-aware regression testing under domain shift.

Recent work has increasingly explored neuron-level interpretation in vision-language models (VLMs) to identify neurons critical to final predictions. However, existing neuron analyses generally focus on single tasks, limiting the comparability of neuron importance across tasks. Moreover, ranking strategies tend to score neurons in isolation, overlooking how task-dependent information pathways shape the write-in effects of feed-forward network (FFN) neurons. This oversight can exacerbate neuron polysemanticity in multi-task settings, introducing noise into the identification and intervention of task-critical neurons. In this study, we propose HONES (Head-Oriented Neuron Explanation & Steering), a gradient-free framework for task-aware neuron attribution and steering in multi-task VLMs. HONES ranks FFN neurons by their causal write-in contributions conditioned on task-relevant attention heads, and further modulates salient neurons via lightweight scaling. Experiments on four diverse multimodal tasks and two popular VLMs show that HONES outperforms existing methods in identifying task-critical neurons and improves model performance after steering. Our source code is released at: https://github.com/petergit1/HONES.

Modern software systems have transitioned from purely code-based architectures to AI-integrated systems where pre-trained models (PTMs) serve as permanent dependencies. However, while the evolution of traditional software libraries is well-documented, we lack a clear understanding of how these "PTM dependencies" change over time. Unlike libraries, PTMs are characterized by opaque internals and less standardized, rapidly evolving release cycles. Furthermore, their multi-role nature enables developers to treat individual instances of a single PTM as separate functional dependencies based on their specific downstream tasks. This raises a critical question for software maintenance: do PTMs change like standard software libraries or do they follow a divergent pattern? To answer this, we present the first empirical study of downstream PTM changes, analyzing a comprehensive dataset of 4,988 releases across 323 GitHub OSS repositories that reuse open-source PTMs. Using traditional software libraries as a baseline, we find that PTMs follow a qualitatively distinct pattern. PTMs are typically added late in the project life-cycle and tend to accumulate rather than be replaced as a project matures. Our findings show that PTM changes are three times less frequent (406 of 2,814 release transitions) than library changes. PTM changes are also less routinely documented, but more likely to carry explicit rationale. Unlike libraries, which evolve reactively, PTM evolution is proactively driven by capability expansion, with a unique documented rationale of PTM testing uncertainty. Our work calls for a rethinking of how PTMs are tracked and managed as dependencies in modern software engineering.

Prompt optimization methods either analyze individual failures in isolation or compare prompt variants across examples, operating on single execution traces with no access to the reasoning process distinguishing success from failure on the same input. We introduce ContraPrompt, built on the observation that when a model fails but succeeds on a retry with feedback, the difference between its two chain-of-thought traces constitutes an optimization signal not captured by prior methods. Unlike prior contrastive methods, we compare complete intermediate reasoning processes: the two traces share model, input, and base prompt, so remaining differences reflect reasoning strategy and appended error feedback -- we call this dyadic reasoning trace analysis. The multi-attempt solving phase is an instrumented agentic retry loop that generates contrastive data automatically without human annotation. Extracted rules are organized into an input-aware decision tree routing instructions by observable input characteristics. On four reasoning and compliance benchmarks, ContraPrompt outperforms GEPA (Agrawal et al., 2026) on all four, with absolute gains of +8.29 pp on HotPotQA (+20.8% rel.), +2.21 pp on GDPR-Bench (+18.2% rel.), +7.14 pp on GPQA Diamond (+10.6% rel.), and +0.74 pp on BBH (+0.85% rel.). Ablations confirm dyadic trace contrastivity is the critical component, with a -16% relative average drop upon its removal. On 53 EvalSet black-box optimization problems, ContraPrompt beats GEPA on 11, ties on 41, and loses on 1 at equal budget. On FiNER-139 financial named entity recognition (Loukas et al., 2022), ContraPrompt achieves +7.77 pp over the unoptimized baseline (+11.6% rel.) and +1.94 pp over GEPA (+2.66% rel.), with branch conditions aligning with standard US GAAP financial-instrument categories.

The key-value (KV) cache is the dominant memory bottleneck during Transformer inference, yet little is known theoretically about how aggressively it can be compressed before multi-step reasoning degrades. We study this through $k$-hop pointer chasing on $n$ tokens under a shared KV cache of size $s$, attention dimension $m$, $H$ heads, $p$-bit precision, and a locality-respecting cache controller (satisfied by all standard KV-compression methods). We give three results. (1) Product depth lower bound (conjectured). We conjecture that any such Transformer ($n \geq 4k$, $s \leq \sqrt{n}/4$) requires depth $L = Ω(\lceil k/s \rceil \cdot \lceil \log_2 n/(Hmp) \rceil)$, and isolate the sole remaining gap as a probabilistic step on the joint distribution of cache trace and pointer chain. Unconditionally, we prove a matching upper bound $L = O(\min(k, \lceil k/s \rceil \log s) \cdot \log n/(mp))$ via windowed pointer doubling, and a max-bound $L = Ω(\max(\lceil k/s \rceil, \log n/(Hmp)))$. Closing the conjecture amounts to upgrading max to product. (2) Bandwidth barrier. The product bound binds only when $Hmp \lesssim \log n$. Any lower bound provable via per-window distinguishability counting -- including reachability, bandwidth, and combinations -- cannot exceed $\lceil k/s \rceil$ once $Hmp \geq \log_2 n$. Breaking this requires lifting unconditional communication-complexity bounds for pointer chasing to Cache-Transformer depth. (3) Adaptive vs oblivious error scaling. Under random cache over $T = \lceil \log_2 k \rceil$ doubling stages, oblivious caches give $\Pr[\mathcal{E}] \leq (s/(n-T))^T + 2T^3/n$ (exponential in $T$), while adaptive locality-respecting caches achieve $\Pr[\mathcal{E}] = s/n$ exactly, independent of $T$. The $Ω((n/s)^{T-1})$ separation explains why heavy-hitter eviction empirically dominates random eviction for multi-hop reasoning.

Reinforcement Learning (RL) has emerged as a powerful training paradigm for LLM-based agents. However, scaling agentic RL for deep research remains constrained by two coupled challenges: hand-crafted synthetic data fails to elicit genuine real-world search capabilities, and real-world search dependency during RL training introduces instability and prohibitive cost, which limits the scalability of Agentic RL. LiteResearcher is a training framework that makes Agentic RL scalable: by constructing a lite virtual world that mirrors real-world search dynamics, we enable a continuously improving training recipe that empowers a tiny search agent to outperform large-scale open-source and commercial models (e.g., Tongyi DeepResearch and Claude-4.5 Sonnet). Specifically, on common benchmarks such as GAIA and Xbench, our LiteResearcher-4B achieves open-source state-of-the-art results of 71.3% and 78.0% respectively, demonstrating that scalable RL training is a key enabler for Deep Research Agents.

Large Language Models (LLMs) exhibit a puzzling disparity in their formal linguistic competence: while they learn some linguistic phenomena with near-perfect mastery, they often perform below chance on others, even after training on trillions of tokens. In this work, we investigate whether these failures stem from inherent architectural limitations or simply the scarcity of these specific grammatical constructions in web-scale corpora. We pre-train simple GPT-2 Small (124M) models on a 100M-token random sample of the FineWeb corpus and intervene by injecting a minimal amount (1%) of synthetic data targeting specific linguistic phenomena. We find that this targeted intervention substantially improves model performance in 8 out of the 9 worst-performing BLiMP paradigms - notably the accuracy on a specific paradigm, only_npi_scope, surges from 20.9% to 69.4%. Furthermore, we observe that these interventions generally preserve or slightly improve aggregate performance. However, while we also identify a resistant phenomenon, principle_A_c_command, whose performance remains below chance even after our data augmentation, our findings do serve as an optimistic existence proof that even small language models can substantially improve on those linguistic phenomena on which models typically perform poorly, provided the pre-training data contains sufficient exposure to them. This suggests that efforts towards human-scale language modeling may benefit greatly by focusing on data composition. The code to reproduce our results is open-sourced at https://github.com/kowndinya-renduchintala/heterogeneity-in-formal-linguistic-competence.

Reinforcement Learning with Verifiable Reward (RLVR) has proven effective for training reasoning-oriented large language models, but existing methods largely assume high-resource settings with abundant training data. In low-resource scenarios, RLVR is prone to more severe entropy collapse, which substantially limits exploration and degrades reasoning performance. To address this issue, we propose Hybrid-domain Entropy dynamics ALignment (HEAL), a framework tailored for few-shot RLVR. HEAL first selectively incorporates high-value general-domain data to promote more diverse exploration. Then, we introduce Entropy Dynamics Alignment (EDA), a reward mechanism that aligns trajectory-level entropy dynamics between the target and general domains, capturing both entropy magnitude and fine-grained variation. Through this alignment, EDA not only further mitigates entropy collapse but also encourages the policy to acquire more diverse exploration behaviors from the general domain. Experiments across multiple domains show that HEAL consistently improves few-shot RLVR performance. Notably, using only 32 target-domain samples, HEAL matches or even surpasses full-shot RLVR trained with 1K target-domain samples.

Visual decoding of neurophysiological signals is a critical challenge for brain-computer interfaces (BCIs) and computational neuroscience. However, current approaches are often constrained by the systematic and stochastic gaps between neural and visual modalities, largely neglecting the intrinsic computational mechanisms of the Human Visual System (HVS). To address this, we propose Brain-Inspired Capture (BI-Cap), a neuromimetic perceptual simulation paradigm that aligns these modalities by emulating HVS processing. Specifically, we construct a neuromimetic pipeline comprising four biologically plausible dynamic and static transformations, coupled with Mutual Information (MI)-guided dynamic blur regulation to simulate adaptive visual processing. Furthermore, to mitigate the inherent non-stationarity of neural activity, we introduce an evidence-driven latent space representation. This formulation explicitly models uncertainty, thereby ensuring robust neural embeddings. Extensive evaluations on zero-shot brain-to-image retrieval across two public benchmarks demonstrate that BI-Cap substantially outperforms state-of-the-art methods, achieving relative gains of 9.2\% and 8.0\%, respectively. We have released the source code on GitHub through the link https://github.com/flysnow1024/BI-Cap.

Synthetic Aperture Radar (SAR) plays a critical role in maritime surveillance, yet deep learning for SAR analysis is limited by the lack of pixel-level annotations. This paper explores how general-purpose vision foundation models can enable zero-shot ship instance segmentation in SAR imagery, eliminating the need for pixel-level supervision. A YOLOv11-based detector trained on open SAR datasets localizes ships via bounding boxes, which then prompt the Segment Anything Model 2 (SAM2) to produce instance masks without any mask annotations. Unlike prior SAM-based SAR approaches that rely on fine tuning or adapters, our method demonstrates that spatial constraints from a SAR-trained detector alone can effectively regularize foundation model predictions. This design partially mitigates the optical-SAR domain gap and enables downstream applications such as vessel classification, size estimation, and wake analysis. Experiments on the SSDD benchmark achieve a mean IoU of 0.637 (89% of a fully supervised baseline) with an overall ship detection rate of 89.2%, confirming a scalable, annotation-efficient pathway toward foundation-model-driven SAR image understanding.

Recent advances in flow-based offline reinforcement learning (RL) have achieved strong performance by parameterizing policies via flow matching. However, they still face critical trade-offs among expressiveness, optimality, and efficiency. In particular, existing flow policies interpret the $L_2$ regularization as an upper bound of the 2-Wasserstein distance ($W_2$), which can be problematic in offline settings. This issue stems from a fundamental geometric mismatch: the behavioral policy manifold is inherently anisotropic, whereas the $L_2$ (or upper bound of $W_2$) regularization is isotropic and density-insensitive, leading to systematically misaligned optimization directions. To address this, we revisit offline RL from a geometric perspective and show that policy refinement can be formulated as a local transport map: an initial flow policy augmented by a residual displacement. By analyzing the induced density transformation, we derive a local quadratic approximation of the KL-constrained objective governed by the Fisher information matrix, enabling a tractable anisotropic optimization formulation. By leveraging the score function embedded in the flow velocity, we obtain a corresponding quadratic constraint for efficient optimization. Our results reveal that the optimality gap in prior methods arises from their isotropic approximation. In contrast, our framework achieves a controllable approximation error within a provable neighborhood of the optimal solution. Extensive experiments demonstrate state-of-the-art performance across diverse offline RL benchmarks. See project page: https://github.com/ARC0127/Fisher-Decorator.

State-of-the-art reasoning models utilize long chain-of-thought (CoT) to solve increasingly complex problems using more test-time computation. In this work, we explore a long CoT setting where the model makes up to K successive attempts at solving a problem, in which each attempt is allowed to build on earlier ones after the model receives a hard verifier feedback. This motivates RL methods that can harness per-attempt rewards by carefully weighting individual attempts. We study optimizing the Verification@K reward (the model succeeds by the K-th attempt) and show that naively weighing the attempts by their pass/fail results in biased gradients. We introduce Calibrated Attempt-Level (CAL) GRPO by devising a weighing strategy to obtain unbiased gradients while maintaining small variance. Our theory reveals how incorporating per-attempt rewards influence the training and the eventual Verification@K performance. Experiments, baselines, and ablations on synthetic and real data corroborate our theory and the benefits of CAL-GRPO over vanilla GRPO as well as naive weighting.

Off-policy learning in constrained MDPs with large binary state spaces faces a fundamental tension: causal identification of transition dynamics requires structural assumptions, while sample-efficient policy learning requires state-space compression. We introduce PI-CMDP, a framework for CMDPs whose constraint dependencies form a layered DAG under a Lifecycle Ordering Assumption (LOA). We propose an Identify-Compress-Estimate pipeline: (i) Identify: LOA enables backdoor identification of causal edge weights for cross-layer pairs, with formal partial-identification bounds when LOA is violated; (ii) Compress: a Markov abstraction compresses state cardinality from 2^(WL) to (W+1)^L under layer-priority regularity and exchangeability; and (iii) Estimate: a physics-guided doubly-robust estimator remains unbiased and reduces the variance constant when the physics prior outperforms a learned model. We instantiate PI-CMDP on constraint repair in engineering simulation pipelines. On the TPS benchmark (4,206 episodes), PI-CMDP achieves 76.2% repair success rate with only 300 training episodes versus 70.8% for the strongest baseline (+5.4 pp), narrowing to +2.8 pp (83.4% vs. 80.6%) in the full-data regime, while substantially reducing cascade failure rates. All improvements are consistent across 5 independent seeds (paired t-test p < 0.02).

GitHub plays a critical role in modern software supply chains, making its security an important research concern. Existing studies have primarily focused on CI/CD automation, collaboration patterns, and community management, while abuse behaviors on GitHub have received little systematic investigation. In this paper, we systematically review and summarize reported GitHub abuse behaviors and conduct an empirical analysis of publicly available abuse cases, curating a manually labeled dataset of 392 GitHub instances. Based on this investigation, we propose a comprehensive taxonomy that characterizes their diverse symptoms and root causes from a software security perspective. Building on this taxonomy, we develop a unified detection framework capable of identifying all abuse categories across repositories and user accounts. Evaluated on the constructed dataset, the proposed framework achieves high performance across all categories (e.g., F1-score exceeding 89%). Collectively, this work advances the understanding of GitHub abuse behaviors and lays the groundwork for large-scale, systematic analysis of the GitHub platform to strengthen software supply chain security.

While Large Language Models (LLMs) exhibit exceptional zero-shot relevance modeling, their high computational cost necessitates framing passage retrieval as a budget-constrained global optimization problem. Existing approaches passively rely on first-stage dense retrievers, which leads to two limitations: (1) failing to retrieve relevant passages in semantically distinct clusters, and (2) failing to propagate relevance signals to the broader corpus. To address these limitations, we propose Bayesian Active Learning with Gaussian Processes guided by LLM relevance scoring (BAGEL), a novel framework that propagates sparse LLM relevance signals across the embedding space to guide global exploration. BAGEL models the multimodal relevance distribution across the entire embedding space with a query-specific Gaussian Process (GP) based on LLM relevance scores. Subsequently, it iteratively selects passages for scoring by strategically balancing the exploitation of high-confidence regions with the exploration of uncertain areas. Extensive experiments across four benchmark datasets and two LLM backbones demonstrate that BAGEL effectively explores and captures complex relevance distributions and outperforms LLM reranking methods under the same LLM budget on all four datasets.

The advent of Large Language Models (LLMs) has fundamentally reshaped the way we interact with graphs, giving rise to a new paradigm called GraphLLM. As revealed in recent studies, graph learning can benefit from LLMs. However, we observe limited benefits when we directly utilize LLMs to make predictions for graph-related tasks within GraphLLM paradigm, which even yields suboptimal results compared to conventional GNN-based approaches. Through in-depth analysis, we find this failure can be attributed to LLMs' limited capability for processing graph data and their tendency to overlook graph information. To address this issue, we propose LoReC (Look, Remember, and Contrast), a novel plug-and-play method for GraphLLM paradigm, which enhances LLM's understanding of graph data through three stages: (1) Look: redistributing attention to graph; (2) Remember: re-injecting graph information into the Feed-Forward Network (FFN); (3) Contrast: rectifying the vanilla logits produced in the decoding process. Extensive experiments demonstrate that LoReC brings notable improvements over current GraphLLM methods and outperforms GNN-based approaches across diverse datasets. The implementation is available at https://github.com/Git-King-Zhan/LoReC.

Vision-Language-Action (VLA) models map multimodal inputs directly to robot actions and are typically trained through large-scale imitation learning. While this paradigm has shown strong performance, prevailing VLA training procedures do not explicitly supervise hard physical constraints such as obstacle avoidance or kinematic feasibility. As a result, the geometric structure underlying physically feasible behavior must be inferred only implicitly from demonstrations. In this paper, we study whether introducing explicit feasibility supervision can provide effective structured guidance for VLA policies. We formulate a simple geometry-grounded feasibility objective and integrate it into the training stage of a diffusion-based VLA policy. To evaluate this idea systematically, we use obstacle-aware manipulation as a controlled probe of geometry-dependent physical feasibility. Empirical results show that augmenting VLA training with feasibility supervision improves both physical reliability and overall task performance, while also enhancing learning efficiency in the low-data regime. These findings indicate that explicit feasibility signals can effectively complement imitation-based VLA learning, highlighting their potential for developing more reliable VLA policies.

Presentation slides are a primary medium for data-driven reporting, yet keeping complex, analytics-style decks up to date remains labor-intensive. Existing automation methods mostly follow fixed template filling and cannot support dynamic updates for diverse, user-authored slide decks. We therefore define "Dynamic Slide Update via Natural Language Instructions on User-provided Templates" and introduce DynaSlide, a large-scale benchmark with 20,036 real-world instruction-execution triples (source slide, user instruction, target slide) grounded in a shared external database and built from business reporting slides under bring-your-own-template (BYO-template) conditions. To tackle this task, we propose SlideAgent, an agent-based framework that combines multimodal slide parsing, natural language instruction grounding, and tool-augmented reasoning for tables, charts, and textual conclusions. SlideAgent updates content while preserving layout and style, providing a strong reference baseline on DynaSlide. We further design end-to-end and component-level evaluation protocols that reveal key challenges and opportunities for future research. The dataset and code are available at https://github.com/XiaoZhou2024/SlideAgent.

Recently, latent reasoning has been introduced into large language models (LLMs) to leverage rich information within a continuous space. However, without stochastic sampling, these methods inevitably collapse to deterministic inference, failing to discover diverse reasoning paths. To bridge the gap, we inject controllable stochasticity into latent reasoning via Gumbel-Softmax, restoring LLMs' exploratory capacity and enhancing their compatibility with Reinforcement Learning (RL). Building on this, we propose \textbf{\underline{L}}atent R\textbf{\underline{e}}asoning \textbf{\underline{P}}olicy \textbf{\underline{O}}ptimization~(\textbf{LEPO}), a novel framework that applies RL directly to continuous latent representations. Specifically, in rollout stage, LEPO maintains stochasticity to enable diverse trajectory sampling, while in optimization stage, LEPO constructs a unified gradient estimation for both latent representations and discrete tokens. Extensive experiments show that LEPO significantly outperforms existing RL methods for discrete and latent reasoning.

Continuous Integration and Deployment (CI/CD) facilitate rapid software delivery, making fast feedback and minimal downtime essential. While caching has been shown to be an effective technique for tackling pipeline performance and reliability issues, existing works have primarily focused on missing dependency caches, ignoring other types of caches and cache misconfigurations. In this paper, we present a comprehensive catalog of ten cache-related smells in GitLab CI/CD that negatively impact performance and reliability, validated on a corpus of grey literature. To address the smells, we propose CROSSER, a tool that automatically detects seven of the ten smells. We evaluate CROSSER on a manually labeled dataset of 82 mature projects, achieving an overall F1 score of 0.98. Finally, we investigate the presence of smells across a large dataset of 228 mature open-source projects and outline our empirical findings. Our results show a widespread frequency of the smells, as only 11% of the projects do not present any. We also show that developers may not be aware of higher-level caching functionalities.

Users often omit essential details in their requests to LLM-based agents, resulting in under-specified inputs for tool use. This poses a fundamental challenge for tool-augmented agents, as API execution typically requires complete arguments, highlighting the need for personalized tool calling. To study this problem, we introduce MPT, a benchmark comprising 265 multi-session dialogues that cover three challenges: Preference Recall, Preference Induction, and Preference Transfer. We also propose PRefine, a test-time memory-augmented method that represents user preferences as evolving hypotheses. Through a generate--verify--refine loop, it extracts reusable constraints from history and improves tool-calling accuracy while using only 1.24% of the tokens required by full-history prompting. These results indicate that robust personalization in agentic systems depends on memory that captures the reasons behind user choices, not just the choices themselves.

Large Language Models (LLMs) are prone to logical hallucinations and stochastic drifts during long-chain reasoning. While Classifier-Free Guidance (CFG) can improve instruction adherence, standard static implementations often cause semantic dilution and linguistic degradation. We propose SPREG (Structured Plan-guided Real-time Entropy Gating), a lightweight inference-time framework for surgical error rectification. SPREG employs an adaptive dual-threshold mechanism to monitor real-time entropy, identifying sudden ``entropy spikes'' as reliable indicators of logical failure. Upon detection, it triggers a dynamic repair by replacing uninformative null-priors with reference distributions synthesized from historical high-confidence states. By modulating guidance intensity according to structured reasoning stages (e.g., Action, Observation), SPREG steers the model back to a stable manifold without compromising fluency. Our experiments demonstrate significant gains, notably a 20.0% absolute accuracy improvement on AIME25, while effectively suppressing uncontrolled entropy drift in complex tasks.

Vibe coding produces correct, executable code at speed, but leaves no record of the structural commitments, dependencies, or evidence behind it. Reviewers cannot determine what invariants were assumed, what changed, or why a regression occurred. This is not a generation failure but a control failure: the dominant artifact of AI-assisted development (code plus chat history) performs dimension collapse, flattening complex system topology into low-dimensional text and making systems opaque and fragile under change. We propose Agentic Consensus: a paradigm in which the consensus layer C, an operable world model represented as a typed property graph, replaces code as the primary artifact of engineering. Executable artifacts are derived from C and kept in correspondence via synchronization operators Phi (realize) and Psi (rehydrate). Evidence links directly to structural claims in C, making every commitment auditable and under-specification explicit as measurable consensus entropy rather than a silent guess. Evaluation must move beyond code correctness toward alignment fidelity, consensus entropy, and intervention distance. We propose benchmark task families designed to measure whether consensus-based workflows reduce human intervention compared to chat-driven baselines.

Skill ecosystems for LLM agents have matured rapidly, yet recent benchmarks show that providing agents with more skills does not monotonically improve performance -- focused sets of 2-3 skills outperform comprehensive documentation, and excessive skills actually hurt. The bottleneck has shifted from skill availability to skill orchestration: agents need not more skills, but a structural mechanism to select, compose, and execute them with explicit causal dependencies. We propose GraSP, the first executable skill graph architecture that introduces a compilation layer between skill retrieval and execution. GraSP transforms flat skill sets into typed directed acyclic graphs (DAGs) with precondition-effect edges, executes them with node-level verification, and performs locality-bounded repair through five typed operators -- reducing replanning from O(N) to O(d^h). Across ALFWorld, ScienceWorld, WebShop, and InterCode with eight LLM backbones, GraSP outperforms ReAct, Reflexion, ExpeL, and flat skill baselines in every configuration, improving reward by up to +19 points over the strongest baseline while cutting environment steps by up to 41%. GraSP's advantage grows with task complexity and is robust to both skill over-retrieval and quality degradation, confirming that structured orchestration -- not larger skill libraries -- is the key to reliable agent execution.

Retrieval-Augmented Generation (RAG) has become a standard approach for enhancing large language models (LLMs) with external knowledge, mitigating hallucinations, and improving factuality. However, existing systems rely on generating natural language queries at each hop and maintaining a strict architectural separation between retriever and generator, preventing them from leveraging the full representational capacity of the LLM. We propose \textbf{LAnR} (Latent Abstraction for RAG), a unified framework in which a single LLM jointly performs encoding, retrieval, and generation entirely within its own latent space. Rather than generating textual queries, LAnR produces dense retrieval vectors from the hidden states of a designated \texttt{[PRED]} token and uses them to match against encoded document representations from the same model. Furthermore, LAnR adaptively decides when sufficient evidence has been retrieved using a lightweight MLP control head over those same hidden states, eliminating both the separate retriever and explicit token-level stopping reasoning. This design is motivated by our empirical observation that answer token entropy reliably signals retrieval sufficiency. Extensive experiments on six QA benchmarks spanning single-hop and multi-hop settings demonstrate that LAnR outperforms existing RAG methods, while achieving improved inference efficiency through reduced number of retrieval calls and tighter model integration.

Spinning flexible objects, exemplified by traditional Chinese handkerchief performances, demands periodic steady-state motions under nonlinear dynamics with frictional contacts and boundary constraints. To address these challenges, we first design an intuitive dexterous wrist based on a parallel anti-parallelogram tendon-driven structure, which achieves 90 degrees omnidirectional rotation with low inertia and decoupled roll-pitch sensing, and implement a high-low level hierarchical control scheme. We then develop a particle-spring model of the handkerchief for control-oriented abstraction and strategy evaluation. Hardware experiments validate this framework, achieving an unfolding ratio of approximately 99% and fingertip tracking error of RMSE = 2.88 mm in high-dynamic spinning. These results demonstrate that integrating control-oriented modeling with a task-tailored dexterous wrist enables robust rest-to-steady-state transitions and precise periodic manipulation of highly flexible objects. More visualizations: https://slowly1113.github.io/icra2026-handkerchief/

As deep learning-based AI technologies gain momentum, the demand for general-purpose AI computing architectures continues to grow. While GPGPU-based architectures offer versatility for diverse AI workloads, they often fall short in efficiency and cost-effectiveness. Various Domain-Specific Architectures (DSAs) excel at particular AI tasks but struggle to extend across broader applications or adapt to the rapidly evolving AI landscape. M100 is Li Auto's response: a performant, cost-effective architecture for AI inference in Autonomous Driving (AD), Large Language Models (LLMs), and intelligent human interactions, domains crucial to today's most competitive automobile platforms. M100 employs a dataflow parallel architecture, where compiler-architecture co-design orchestrates not only computation but, more critically, data movement across time and space. Leveraging dataflow computing efficiency, our hardware-software co-design improves system performance while reducing hardware complexity and cost. M100 largely eliminates caching: tensor computations are driven by compiler- and runtime-managed data streams flowing between computing elements and on/off-chip memories, yielding greater efficiency and scalability than cache-based systems. Another key principle was selecting the right operational granularity for scheduling, issuing, and execution across compiler, firmware, and hardware. Recognizing commonalities in AI workloads, we chose the tensor as the fundamental data element. M100 demonstrates general AI computing capability across diverse inference applications, including UniAD (for AD) and LLaMA (for LLMs). Benchmarks show M100 outperforms GPGPU architectures in AD applications with higher utilization, representing a promising direction for future general AI computing.

Can syntactic processing emerge spontaneously from purely local interaction? We present a concrete instance on a minimal system: an 18,658-parameter two-dimensional neural cellular automaton (NCA), supervised by nothing more than a 1-bit boundary signal, is trained on the membership problem of an arithmetic-expression grammar. After training, its internal $L \times L$ grid spontaneously self-organizes into an ordered, spatially extended representation that we name Proto-CKY. This representation satisfies three operational criteria for syntactic processing: expressive power beyond the regular languages, structural generalization beyond the training distribution, and an internal organization quantitatively aligned with grammatical structure (Pearson $r \approx 0.71$). It emerges independently on four context-free grammars and regenerates spontaneously after perturbation. Proto-CKY is functionally aligned with the CKY algorithm but formally distinct from it: it is a physical prototype, a concrete instantiation of a mathematical ideal on a physical substrate, and the systematic distance between the two carries information about the substrate itself.

Computer-use agents have rapidly improved on real-world tasks such as web navigation, desktop automation, and software interaction, in some cases surpassing human performance. Yet even when the task and model are unchanged, an agent that succeeds once may fail on a repeated execution of the same task. This raises a fundamental question: if an agent can succeed at a task once, what prevents it from doing so reliably? In this work, we study the sources of unreliability in computer-use agents through three factors: stochasticity during execution, ambiguity in task specification, and variability in agent behavior. We analyze these factors on OSWorld using repeated executions of the same task together with paired statistical tests that capture task-level changes across settings. Our analysis shows that reliability depends on both how tasks are specified and how agent behavior varies across executions. These findings suggest the need to evaluate agents under repeated execution, to allow agents to resolve task ambiguity through interaction, and to favor strategies that remain stable across runs.

MRI is preferred over CT in paediatric imaging because it avoids ionising radiation, but its use in spine deformity assessment is largely limited by the lack of automated, high-resolution 3D bony reconstruction, which continues to rely on CT. MRI-based 3D reconstruction remains impractical due to manual workflows and the scarcity of labelled full-spine datasets. This study introduces an AI framework that enables fully automated thoracolumbar spine (T1-L5) segmentation and 3D reconstruction from MRI alone. Historical low-dose CT scans from adolescent idiopathic scoliosis (AIS) patients were converted into MRI-like images using a GAN and combined with existing labelled thoracic MRI data to train a U-Net-based model. The resulting algorithm accurately generated continuous thoracolumbar 3D reconstructions, improved segmentation accuracy (88% Dice score), and reduced processing time from approximately 1 hour to under one minute, while preserving AIS-specific deformity features. This approach enables radiation-free 3D deformity assessment from MRI, supporting clinical evaluation, surgical planning, and navigation in paediatric spine care.

General-purpose LLMs pose misinformation risks for development and policy experts, lacking epistemic humility for verifiable outputs. We present AVA (AI + Verified Analysis), a GenAI platform built on a curated library of over 4,000 World Bank Reports with multilingual capabilities. AVA's multi-agent pipeline enables users to query and receive evidence-based syntheses. It operationalizes epistemic humility through two mechanisms: citation verifiability (tracing claims to sources) and reasoned abstention (declining unsupported queries with justification and redirection). We conducted an in-the-wild evaluation with over 2,200 individuals from heterogeneous organisations and roles in 116 countries, via log analysis, surveys, and 20 interviews. Difference-in-Differences estimates associate sustained engagement with 2.4-3.9 hours saved weekly. Qualitatively, participants used AVA as a specialized "evidence engine"; reasoned abstention clarified scope boundaries, and trust was calibrated through institutional provenance and page-anchored citations. We contribute design guidelines for specialized AI and articulate a vision for "ecosystem-aware" Humble AI.

LLM benchmarks are increasingly dynamic: instead of containing a fixed set of questions, they define templates and parameters that can generate an effectively unlimited number of question variants. This flexibility is valuable, but it makes evaluation expensive -- especially when the goal is not just determining an average score, but reliably identifying a model's weak spots. This paper introduces a new methodology for identifying hard questions in dynamic benchmarks. It leverages COUP, a recent Bayesian optimization algorithm (Graham, Velez & Leyton-Brown, 2026), after introducing several substantive modifications to make the algorithm suitable for practical LLM pipelines. We also wrap it in a tool that supports flexible choices of datasets and utility functions, enabling users to target the kinds of questions they care about (e.g., low-accuracy questions; questions that are unusually hard relative to their measured complexity). In experiments across a range of benchmarks, we show that our method, dubbed $\texttt{QuickScope}$, discovers truly difficult questions more sample efficiently than standard baselines, while also reducing false positives from noisy outcomes.

Generative modeling within constrained sets is essential for scientific and engineering applications involving physical, geometric, or safety requirements (e.g., molecular generation, robotics). We present a unified framework for constrained diffusion models on generic nonconvex feasible sets $Σ$ that simultaneously enforces equality and inequality constraints throughout the diffusion process. Our framework incorporates both overdamped and underdamped dynamics for forward and backward sampling. A key algorithmic innovation is a computationally efficient landing mechanism that replaces costly and often ill-defined projections onto $Σ$, ensuring feasibility without iterative Newton solves or projection failures. By leveraging underdamped dynamics, we accelerate mixing toward the prior distribution, effectively alleviating the high simulation costs typically associated with constrained diffusion. Empirically, this approach reduces function evaluations and memory usage during both training and inference while preserving sample quality. On benchmarks featuring equality and mixed constraints, our method achieves comparable sample quality to state-of-the-art baselines while significantly reducing computational cost, providing a practical and scalable solution for diffusion on nonconvex feasible sets.

An LLM's residual stream is both state and instruction: it encodes the current context and determines the next transformation. We introduce a parameter-free decomposition for Mixture-of-Experts models that splits each layer's hidden state into a control signal that causally drives routing and an orthogonal content channel invisible to the router. Across six MoE architectures, we find that models preserve surface-level features (language, token identity, position) in the content channel, while the control signal encodes an abstract function that rotates from layer to layer. Because each routing decision is low-bandwidth, this hand-off forces compositional specialization across layers. While individual experts remain polysemantic, expert paths become monosemantic, clustering tokens by semantic function across languages and surface forms. The same token (e.g., ":") follows distinct trajectories depending on whether it serves as a type annotation, an introductory colon, or a time separator. Our decomposition identifies the source of this structure: clusters in the control subspace are substantially more monosemantic than those in the full representation. As a result, the natural unit of interpretability in MoEs is not the expert but the trajectory.

Whether the Indus Valley sign system (c. 2600-1900 BCE) encodes spoken language has been debated for decades. This paper introduces a multi-metric discrimination framework that tests the observed Indus corpus against two kinds of computer-generated non-linguistic baseline -- one mimicking a heraldic emblem system, the other an administrative coding system -- each calibrated with Zipfian frequency distributions, positional constraints, and bigram dependencies derived from six attested non-linguistic corpora. The scorecard evaluates four properties central to the Farmer-Sproat-Witzel (2004) critique: text brevity, repeated formulaic phrases, hapax legomenon rate, and positional rigidity. Applying this framework to 1,916 deduplicated inscriptions (584 unique signs, 11,110 tokens) from the ICIT/Yajnadevam digitization, we find that the Indus corpus does not match either baseline cleanly. Across the four metrics examined, the Indus corpus occupies an intermediate position relative to the two baseline families, matching neither cleanly. Neither a heraldic nor an administrative generator can reproduce all four properties at once. We also compare against seven real-world non-linguistic corpora including Sproat's (2014) datasets, finding that no attested non-linguistic system reproduces the full Indus statistical profile either. We replicate key prior results including a Zipf slope of -1.49 and conditional entropy of 3.23 bits. All code and data are publicly available.

Large language models (LLMs) offer strong capabilities but raise cost and privacy concerns, whereas small language models (SLMs) facilitate efficient and private local inference yet suffer from limited capacity. To synergize the complementary strengths, we introduce a dynamic collaboration framework, where an SLM learns to proactively decide how to request an LLM during multi-step reasoning, while the LLM provides adaptive feedback instead of acting as a passive tool. We further systematically investigate how collaboration strategies are shaped by SLM and LLM capabilities as well as efficiency and privacy constraints. Evaluation results reveal a distinct scaling effect: stronger SLMs become more self-reliant, while stronger LLMs enable fewer and more informative interactions. In addition, the learned dynamic collaboration strategies significantly outperform static pipelines and standalone inference, and transfer robustly to unseen LLMs.

In this paper, we propose a novel approach for generating music based on an artificial intelligence (AI) system. We analyze the features of music and use them to fit and predict the music. The fractional Fourier transform (FrFT) and the long short-term memory (LSTM) network are the foundations of our method. The FrFT method is used to extract the spectral features of a music piece, where the music signal is expressed on the time and frequency domains. The LSTM network is used to generate new music based on the extracted features, where we predict the music according to the hidden layer features and real-time inputs using GiantMIDI-Piano dataset. The results of our experiments show that our proposed system is capable of generating high-quality music that is comparable to human-generated music.

Recent advancements in large language models (LLMs) have empowered autonomous web agents to execute natural language instructions directly on real-world webpages. However, existing agents often struggle with complex tasks involving dynamic interactions and long-horizon execution due to rigid planning strategies and hallucination-prone reasoning. To address these limitations, we propose WebUncertainty, a novel autonomous agent framework designed to tackle dual-level uncertainty in planning and reasoning. Specifically, we design a Task Uncertainty-Driven Adaptive Planning Mechanism that adaptively selects planning modes to navigate unknown environments. Furthermore, we introduce an Action Uncertainty-Driven Monte Carlo tree search (MCTS) Reasoning Mechanism. This mechanism incorporates the Confidence-induced Action Uncertainty (ConActU) strategy to quantify both aleatoric uncertainty (AU) and epistemic uncertainty (EU), thereby optimizing the search process and guiding robust decision-making. Experimental results on the WebArena and WebVoyager benchmarks demonstrate that WebUncertainty achieves superior performance compared to state-of-the-art baselines.

Block-based programming environments such as Scratch are widely used in introductory computing education, yet scalable and reliable automated assessment remains elusive. Scratch programs are highly heterogeneous, event-driven, and visually grounded, which makes traditional assertion-based or test-based grading brittle and difficult to scale. As a result, assessment in real Scratch classrooms still relies heavily on manual inspection and delayed feedback, introducing inconsistency across instructors and limiting scalability. We present Raven, an automated assessment framework for Scratch that replaces program-specific state assertions with instructor-specified, task-level video generation rules shared across all student submissions. Raven integrates large language models with video analysis to evaluate whether a program's observed visual and interactive behaviors satisfy grading criteria, without requiring explicit test cases or predefined outputs. This design enables consistent evaluation despite substantial diversity in implementation strategies and interaction sequences. We evaluate Raven on 13 real Scratch assignments comprising over 140 student submissions with ground-truth labels from human graders. The results show that Raven significantly outperforms prior automated assessment tools in both grading accuracy and robustness across diverse programming styles. A classroom study with 30 students and 10 instructors further demonstrates strong user acceptance and practical applicability. Together, these findings highlight the effectiveness of task-level behavioral abstractions for scalable assessment of open-ended, event-driven programs.

Large language models (LLMs) perform substantially below human level on existing theory-of-mind (ToM) benchmarks, even when augmented with chain-of-thought prompting or probabilistic belief updates. We argue that these failures primarily arise from unreliable implicit state tracking rather than limitations in high-level reasoning. We introduce PDDL-Mind, a neuro-symbolic framework that decouples environment state evolution from belief inference. By translating narrative descriptions into explicit states and actions expressed in Planning Domain Definition Language (PDDL), and by verifying action-induced state transitions against a predefined domain, PDDL-Mind provides LLMs with a logically consistent and explicit representation of world states for ToM tasks. Experiments on MMToM-QA, MuMA and FanToM show that PDDL-Mind achieves over 5% absolute accuracy gain over the best existing state-of-the-art method on ToM benchmark questions.

With the rapid advancement of large language models (LLMs), mobile agents have emerged as promising tools for phone automation, simulating human interactions on screens to accomplish complex tasks. However, these agents often suffer from low accuracy, misinterpretation of user instructions, and failure on challenging tasks, with limited prior work examining why and where they fail. To address this, we introduce DailyDroid, a benchmark of 75 tasks in five scenarios across 25 Android apps, spanning three difficulty levels to mimic everyday smartphone use. We evaluate it using text-only and multimodal (text + screenshot) inputs on GPT-4o and o4-mini across 300 trials, revealing comparable performance with multimodal inputs yielding marginally higher success rates. Through in-depth failure analysis, we compile a handbook of common failures. Our findings reveal critical issues in UI accessibility, input modalities, and LLM/app design, offering implications for future mobile agents, applications, and UI development.

When language models answer open-ended problems, they implicitly make hidden decisions that shape their outputs, leaving users with uncontextualized answers rather than a working map of the problem; drawing on multiverse analysis from statistics, we build and evaluate the conceptual multiverse, an interactive system that represents conceptual decisions such as how to frame a question or what to value as a space users can transparently inspect, intervenably change, and check against principled domain reasoning; for this structure to be worth navigating rather than misleading, it must be rigorous and checkable against domain reasoning norms, so we develop a general verification framework that enforces properties of good decision structures like unambiguity and completeness calibrated by expert-level reasoning; across three domains, the conceptual multiverse helped participants develop a working map of the problem, with philosophy students rewriting essays with sharper framings and reversed theses, alignment annotators moving from surface preferences to reasoning about user intent and harm, and poets identifying compositional patterns that clarified their taste.

Code secrets are sensitive assets for software developers, and their leakage poses significant cybersecurity risks. While the rapid development of AI code assistants powered by Code Large Language Models (CLLMs), CLLMs are shown to inadvertently leak such secrets due to a notorious memorization phenomenon. This study first reveals that Byte-Pair Encoding (BPE) tokenization leads to unexpected behavior of secret memorization, which we term as \textit{gibberish bias}. Specifically, we identified that some secrets are among the easiest for CLLMs to memorize. These secrets yield high character-level entropy, but low token-level entropy. Then, this paper supports the biased claim with numerical data. We identified that the roots of the bias are the token distribution shift between the CLLM training data and the secret data. We further discuss how gibberish bias manifests under the ``larger vocabulary'' trend. To conclude the paper, we discuss potential mitigation strategies and the broader implications on current tokenizer design.

Zero-knowledge proof (ZKP) provers remain costly because multi-scalar multiplication (MSM) and number-theoretic transforms (NTTs) dominate runtime as they need significant computation. AI ASICs such as TPUs provide massive matrix throughput and SotA energy efficiency. We present MORPH, the first framework that reformulates ZKP kernels to match AI-ASIC execution. We introduce Big-T complexity, a hardware-aware complexity model that exposes heterogeneous bottlenecks and layout-transformation costs ignored by Big-O. Guided by this analysis, (1) at arithmetic level, MORPH develops an MXU-centric extended-RNS lazy reduction that converts high-precision modular arithmetic into dense low-precision GEMMs, eliminating all carry chains, and (2) at dataflow level, MORPH constructs a unified-sharding layout-stationary TPU Pippenger MSM and optimized 3/5-step NTT that avoid on-TPU shuffles to minimize costly memory reorganization. Implemented in JAX, MORPH enables TPUv6e8 to achieve up-to 10x higher throughput on NTT and comparable throughput on MSM than GZKP. Our code: https://github.com/EfficientPPML/MORPH.

(1)Cross-border data transfers have become a matter of daily occurrence against the backdrop of the development of cloud computing and artificial intelligence. Consequently, where a data leak gives rise to civil liability, the determination of that liability inevitably assumes an international dimension involving foreign elements. (2)As is starkly demonstrated by secret sharing technology in cloud computing, fragments of data may be presumed to be distributed across multiple jurisdictions on a global scale. This renders traditional private international law measures -- predicated on the identification of a physical location -- inadequate for the purposes of determining the applicable law, a difficulty that is particularly acute in relation to non-contractual obligations. (3)Bearing in mind the typical scenario encountered in practice -- in which a Data Subject brings a claim for damages against a SaaS (Software as a Service) provider, which in turn seeks recourse against an IaaS (Infrastructure as a Service) or PaaS (Platform as a Service) provider -- a characteristic feature of such cases is the concurrence of contractual and non-contractual obligations. Taking this feature into account, it is possible to determine the applicable law governing non-contractual obligations through party autonomy -- by aligning it with the law governing the contractual obligation as selected by the parties, an approach that may be termed private ordering. This serves to overcome the difficulties associated with the identification of a physical location and, at the same time, contributes to ensuring the foreseeability of the parties.

Pairwise ranking systems based on Maximum Likelihood Estimation (MLE), such as the Bradley-Terry model, are widely used to aggregate preferences from pairwise comparisons. However, their robustness under strategic data manipulation remains insufficiently understood. In this paper, we study the vulnerability of MLE-based ranking systems to adversarial perturbations. We formulate the manipulation task as a constrained combinatorial optimization problem and propose an Adaptive Subset Selection Attack (ASSA) to efficiently identify high-impact perturbations. Experimental results on both synthetic data and real-world election datasets show that MLE-based rankings exhibit a sharp phase-transition behavior: beyond a small perturbation budget, a limited number of strategic voters can significantly alter the global ranking. In particular, our method consistently outperforms random and greedy baselines under constrained budgets. These findings reveal a fundamental sensitivity of MLE-based ranking mechanisms to structured perturbations and highlight the need for more robust aggregation methods in collective decision-making systems.

Post-training Large Language Models requires diverse, high-quality data which is rare and costly to obtain, especially in low resource domains and for multi-turn conversations. Common solutions are crowdsourcing or synthetic generation, but both often yield low-quality or low-diversity data. We introduce Adversarial Arena for building high quality conversational datasets by framing data generation as an adversarial task: attackers create prompts, and defenders generate responses. This interactive competition between multiple teams naturally produces diverse and complex data. We validated this approach by conducting a competition with 10 academic teams from top US and European universities, each building attacker or defender bots. The competition, focused on safety alignment of LLMs in cybersecurity, generated 19,683 multi-turn conversations. Fine-tuning an open-source model on this dataset produced an 18.47% improvement in secure code generation on CyberSecEval-Instruct and 29.42% improvement on CyberSecEval-MITRE.

The democratization of ubiquitous AI hinges on deploying sophisticated reasoning capabilities on resource-constrained devices. However, Small Language Models (SLMs) often face a "reasoning gap", particularly in non-English languages like Vietnamese, where they struggle to maintain coherent chains of thought. This paper investigates Test-Time Scaling strategies for the Qwen3-1.7B architecture within the context of Vietnamese Elementary Mathematics. We introduce Vi-S1K, a high-fidelity reasoning dataset localized via a Gemini 2.5 Flash-Lite powered pipeline, and Vi-Elementary-Bench, a dual-resource benchmark for rigorous evaluation. Using an LLM-as-a-Judge protocol, we reveal that the base model possesses robust latent knowledge (Accuracy: 4.05/5.00) but suffers from a severe "formatting gap" in communication. Supervised Fine-Tuning (SFT) acts as a critical "reasoning unlocker", yielding a 77% improvement in Explanation Quality and bridging the gap between raw calculation and pedagogical coherence. Furthermore, our analysis of prompting strategies uncovers a significant trade-off: structured frameworks like ReAct impose a "cognitive tax" on the 1.7B parameter capacity, degrading performance relative to pure Chain-of-Thought (CoT) combined with Self-Consistency. These findings establish a deployment hierarchy for SLMs, demonstrating that SFT combined with simplified test-time scaling is superior to complex agentic workflows for edge-based reasoning.

The MXFP4 microscaling format, which partitions tensors into blocks of 32 elements sharing an E8M0 scaling factor, has emerged as a promising substrate for efficient LLM inference, backed by native hardware support on NVIDIA Blackwell Tensor Cores. However, activation outliers pose a unique challenge under this format: a single outlier inflates the shared block scale, compressing the effective dynamic range of the remaining elements and causing significant quantization error. Existing rotation-based remedies, including randomized Hadamard and learnable rotations, are data-agnostic and therefore unable to specifically target the channels where outliers concentrate. We propose DuQuant++, which adapts the outlier-aware fine-grained rotation of DuQuant to the MXFP4 format by aligning the rotation block size with the microscaling group size (B{=}32). Because each MXFP4 group possesses an independent scaling factor, the cross-block variance issue that necessitates dual rotations and a zigzag permutation in the original DuQuant becomes irrelevant, enabling DuQuant++ to replace the entire pipeline with a single outlier-aware rotation, which halves the online rotation cost while simultaneously smoothing the weight distribution. Extensive experiments on the LLaMA-3 family under MXFP4 W4A4 quantization show that DuQuant++ consistently achieves state-of-the-art performance. Our code is available at https://github.com/Hsu1023/DuQuant++.

Precision-critical manipulation requires both global trajectory organization and local execution correction, yet most vision-language-action (VLA) policies generate actions within a single unified space. This monolithic formulation forces macro-level transport and micro-level refinement to be optimized under the same objective, causing large motions to dominate learning while suppressing small but failure-critical corrective signals. In contrast, human manipulation is structured by global movement planning together with continuous local adjustment during execution. Motivated by this principle, we propose AnchorRefine, a hierarchical framework that factorizes VLA action modeling into trajectory anchor and residual refinement. The anchor planner predicts a coarse motion scaffold, while the refinement module corrects execution-level deviations to improve geometric and contact precision. We further introduce a decision-aware gripper refinement mechanism to better capture the discrete and boundary-sensitive nature of gripper control. Experiments on LIBERO, CALVIN, and real-robot tasks demonstrate that AnchorRefine consistently improves both regression-based and diffusion-based VLA backbones, yielding gains of up to 7.8% in simulation success rate and 18% in real-world success rate.

Unlearning in large language models (LLMs) has emerged as a promising safeguard against adversarial behaviors. When the forgetting loss is applied uniformly without considering token-level semantic importance, model utility can be unnecessarily degraded. Recent studies have explored token-wise loss regularizers that prioritize informative tokens, but largely rely on ground-truth confidence or external linguistic parsers, which limits their ability to capture contextual information or the model's overall predictive state. Intuitively, function words like "the" primarily serve syntactic roles and are highly predictable with little ambiguity, but informative words admit multiple plausible alternatives with greater uncertainty. Based on this intuition, we propose Entropy-guided Token Weighting (ETW), a token-level unlearning regularizer that uses entropy of the predictive distribution as a proxy for token informativeness. We demonstrate that informative tokens tend to have higher entropy, whereas structural tokens tend to have lower entropy. This behavior enables ETW to achieve more effective unlearning while better preserving model utility than existing token-level approaches.

Large language models (LLMs) are increasingly deployed in the telecommunications domain for critical tasks, relying heavily on Retrieval-Augmented Generation (RAG) to adapt general-purpose models to continuously evolving standards. However, a significant gap exists in evaluating the embedding models that power these RAG pipelines, as general-purpose benchmarks fail to capture the dense, acronym-heavy, and highly cross-referential nature of telecommunications corpora. To address this, we introduce TeleEmbedBench, the first large-scale, multi-corpus embedding benchmark designed specifically for telecommunications. The benchmark spans three heterogeneous corpora: O-RAN Alliance specifications, 3GPP release documents, and the srsRAN open-source codebase, comprising 9,000 question-chunk pairs across three standard chunk sizes (512, 1024, and 2048 tokens). To construct this dataset at scale without manual annotation bottlenecks, we employ a novel automated pipeline where one LLM generates specific queries from text chunks and a secondary LLM validates them across strict criteria. We comprehensively evaluate eight embedding models, spanning standard sentence-transformers and LLM-based embedders. Our results demonstrate that LLM-based embedders, such as Qwen3 and EmbeddingGemma, consistently and significantly outperform traditional sentence-transformers in both retrieval accuracy and robustness against cross-domain interference. Additionally, we introduce TeleEmbedBench-Clean to evaluate model robustness against noisy, incomplete user queries. Finally, our analysis reveals that while domain-specific task instructions improve embedder performance for raw source code, they paradoxically degrade retrieval performance for natural language telecommunications specifications.

LLM agents in markets present algorithmic collusion risks. While prior work shows LLM agents reach supracompetitive prices through tacit coordination, existing research focuses on hand-crafted prompts. The emerging paradigm of prompt optimization necessitates new methodologies for understanding autonomous agent behavior. We investigate whether prompt optimization leads to emergent collusive behaviors in market simulations. We propose a meta-learning loop where LLM agents participate in duopoly markets and an LLM meta-optimizer iteratively refines shared strategic guidance. Our experiments reveal that meta-prompt optimization enables agents to discover stable tacit collusion strategies with substantially improved coordination quality compared to baseline agents. These behaviors generalize to held-out test markets, indicating discovery of general coordination principles. Analysis of evolved prompts reveals systematic coordination mechanisms through stable shared strategies. Our findings call for further investigation into AI safety implications in autonomous multi-agent systems.

Large language models (LLMs) have achieved strong performance on natural language to SQL (NL2SQL) benchmarks, yet their reported accuracy may be inflated by contamination from benchmark queries or structurally similar patterns seen during training. We introduce SPENCE (Syntactic Probing and Evaluation of NL2SQL Contamination Effects), a controlled syntactic probing framework for detecting and quantifying such contamination. SPENCE systematically generates syntactic variants of test queries for four widely used NL2SQL datasets-Spider, SParC, CoSQL, and the newer BIRD benchmark. We use SPENCE to evaluate multiple high-capacity LLMs under execution-based scoring. For each model, we measure changes in execution accuracy across increasing levels of syntactic divergence and quantify rank sensitivity using Kendall's tau with bootstrap confidence intervals. By aligning these robustness trends with benchmark release dates, we observe a clear temporal gradient: older benchmarks such as Spider exhibit the strongest negative values and thus the highest likelihood of training leakage, whereas the more recent BIRD dataset shows minimal sensitivity and appears largely uncontaminated. Together, these findings highlight the importance of temporally contextualized, syntactic-probing evaluation for trustworthy NL2SQL benchmarking.

Data scarcity remains a fundamental bottleneck in applying deep learning to wireless communication problems, particularly in scenarios where collecting labeled Radio Frequency (RF) data is expensive, time-consuming, or operationally constrained. This paper proposes LLM-AUG, a data augmentation framework that leverages in-context learning in large language models (LLMs) to generate synthetic training samples directly in a learned embedding space. Unlike conventional generative approaches that require training task-specific models, LLM-AUG performs data generation through structured prompting, enabling rapid adaptation in low-shot regimes. We evaluate LLM-AUG on two representative tasks: modulation classification and interference classification using the RadioML 2016.10A dataset, and the Interference Classification (IC) dataset respectively. Results show that LLM-AUG consistently outperforms traditional augmentation and deep generative baselines across low-shot settings and reaches near oracle performance using only 15% labeled data. LLM-AUG further demonstrates improved robustness under distribution shifts, yielding a 29.4% relative gain over diffusion-based augmentation at a lower SNR value. On the RadioML and IC datasets, LLM-AUG yields a relative gain of 67.6% and 35.7% over the diffusion-based baseline. The t-SNE visualizations further validate that synthetic samples generated by better preserve class structure in the embedding space, leading to more consistent and informative augmentations. These results demonstrate that LLMs can serve as effective and practical data augmenters for wireless machine learning, enabling robust and data-efficient learning in evolving wireless environments.

Ensuring the safety of large language models (LLMs) requires robust red teaming, yet the systematic synthesis of high-quality toxic data remains under-explored. We propose Reverse Constitutional AI (R-CAI), a framework for automated and controllable adversarial data generation that moves beyond isolated jailbreak prompts. By inverting a harmless constitution into a constitution of toxicity and iteratively refining model outputs through a critique--revision pipeline, R-CAI enables scalable synthesis of multi-dimensional adversarial data without human annotation. Optimizing solely for toxicity-related rewards, however, can lead to reward hacking and degraded semantic coherence. To address this challenge, we introduce probability clamping within reinforcement learning from AI feedback, which stabilizes adversarial optimization while preserving adversarial intent. Experiments demonstrate that R-CAI generates diverse, high-quality toxic data and that probability clamping substantially improves semantic coherence (15%) without sacrificing adversarial strength. Overall, R-CAI provides a fully automated framework for red teaming data generation and systematic safety evaluation of aligned language models.

The reliability of VLM-as-a-Judge is critical for the automatic evaluation of vision-language models (VLMs). Despite recent progress, our analysis reveals that VLM-as-a-Judge often pays limited attention to the image when making decisions. Instead, they often blindly favor the more informative answer, even when they can recognize it conflicts with the image content. We call this problem informativeness bias, which significantly undermines judge reliability. To address it, we propose BIRCH (Balanced Informativeness and CoRrectness with a Truthful AnCHor), a judging paradigm that first corrects inconsistencies with the image content in candidate answers, and then compares the answers against this corrected version. This shifts the judge's focus from informativeness to image-grounded correctness. Experiments on multiple models and benchmarks show that BIRCH reduces informativeness bias by up to 17%, resulting in performance gains of up to 9.8%. Our work reveals an overlooked but fundamental flaw in current VLM-as-a-Judge systems and highlights the need for more principled designs.

This paper presents a controlled quasi-experimental developer study examining whether a layer-based security training package is associated with improved security quality in LLM-assisted implementation of an identity-centric Java Spring Boot backend. The study uses a mixed design with a within-subject pre-training versus post-training comparison and an exploratory between-subject expertise factor. Twelve developers completed matched runs under a common interface, fixed model configuration, counterbalanced task sets, and a shared starter project. Security outcomes were assessed via independent manual validation of submitted repositories by the first and second authors. The primary participant-level endpoint was a severity-weighted validated-weakness score. The post-training condition showed a significant paired reduction under an exact Wilcoxon signed-rank test ($p = 0.0059$). In aggregate, validated weaknesses decreased from 162 to 111 (31.5\%), the severity-weighted burden decreased from 432 to 267 (38.2\%), and critical findings decreased from 24 to 5 (79.2\%). The largest reductions were in authorization and object access (53.3\%) and in authentication, credential policy, and recovery weaknesses (44.7\%). Session and browser trust-boundary issues showed minimal change, while sensitive-data and cryptographic weaknesses showed only marginal improvement. These results suggest that, under the tested conditions, post-training runs reduce validated security burden in LLM-assisted backend development without modifying the model. They do not support replacing secure defaults, static analysis, expert review, or operational hardening.

Interpretability tools are increasingly used to analyze failures of Large Language Models (LLMs), yet prior work largely focuses on short prompts or toy settings, leaving their behavior on commonly used benchmarks underexplored. To address this gap, we study contrastive, LRP-based attribution as a practical tool for analyzing LLM failures in realistic settings. We formulate failure analysis as \textit{contrastive attribution}, attributing the logit difference between an incorrect output token and a correct alternative to input tokens and internal model states, and introduce an efficient extension that enables construction of cross-layer attribution graphs for long-context inputs. Using this framework, we conduct a systematic empirical study across benchmarks, comparing attribution patterns across datasets, model sizes, and training checkpoints. Our results show that this token-level contrastive attribution can yield informative signals in some failure cases, but is not universally applicable, highlighting both its utility and its limitations for realistic LLM failure analysis. Our code is available at: https://aka.ms/Debug-XAI.

Climate-driven wildfires are intensifying, particularly in urban regions such as Southern California. Yet, traditional fire risk communication tools often fail to gain public trust due to inaccessible design, non-transparent outputs, and limited contextual relevance. These challenges are especially critical in high-risk communities, where trust depends on how clearly and locally information is presented. Neighborhoods such as Pacific Palisades, Pasadena, and Altadena in Los Angeles exemplify these conditions. This study introduces a community-led approach for integrating AI into wildfire risk assessment using the Participatory AI Literacy and Explainability Integration (PALEI) framework. PALEI emphasizes early literacy building, value alignment, and participatory evaluation before deploying predictive models, prioritizing clarity, accessibility, and mutual learning between developers and residents. Early engagement findings show strong acceptance of visual, context-specific risk communication, positive fairness perceptions, and clear adoption interest, alongside privacy and data security concerns that influence trust. Participants emphasized localized imagery, accessible explanations, neighborhood-specific mitigation guidance, and transparent communication of uncertainty. The outcome is a mobile application co-designed with users and stakeholders, enabling residents to scan visible property features and receive interpretable fire risk scores with tailored recommendations. By embedding local context into design, the tool becomes an everyday resource for risk awareness and preparedness. This study argues that user experience is central to ethical and effective AI deployment and provides a replicable, literacy-first pathway for applying the PALEI framework to climate-related hazards.

Merging multiple Low-Rank Adaptation (LoRA) experts into a single backbone is a promising approach for efficient multi-task deployment. While existing methods strive to alleviate interference via weight interpolation or subspace alignment, they rest upon the implicit assumption that all LoRA matrices contribute constructively to the merged model. In this paper, we uncover a critical bottleneck in current merging paradigms: the existence of $\textit{negative modules}$ -- specific LoRA layers that inherently degrade global performance upon merging. We propose $\textbf{E}$volutionary $\textbf{N}$egative $\textbf{M}$odule $\textbf{P}$runing ($\textbf{ENMP}$), a plug-and-play LoRA pruning method to locate and exclude these detrimental modules prior to merging. By leveraging an evolutionary search strategy, ENMP effectively navigates the discrete, non-differentiable landscape of module selection to identify optimal pruning configurations. Extensive evaluations demonstrate that ENMP consistently boosts the performance of existing merging algorithms, achieving a new state-of-the-art across both language and vision domains. Code is available at https://github.com/CaoAnda/ENMP-LoRAMerging.

Adapting foundation models under resource budgets relies heavily on Parameter-Efficient Fine-Tuning (PEFT), with LoRA being a standard modular solution. However, LoRA suffers from spectral interference. Low-rank updates often concentrate energy on the leading singular directions of pretrained weights, perturbing general capabilities and causing catastrophic forgetting and fragile multi-adapter merging. To resolve this, we propose HiP-LoRA, a spectrum-aware adaptation framework. Utilizing the cached singular value decomposition (SVD) of pretrained layers, HiP-LoRA decomposes updates into two channels: a principal channel within the dominant singular subspace, and a residual low-rank channel in the orthogonal complement. A singular-value-weighted stability budget on the principal channel continuously balances pretrained behavior preservation with task-specific plasticity. Experiments on Llama-3.1-8B demonstrate that under matched budgets, HiP-LoRA drastically reduces pretraining degradation and multi-adapter MergeFail, robustly outperforming baselines in interference-sensitive tasks like continual tuning and knowledge editing.

This paper considers reinforcement learning from human feedback in a federated learning setting with resource-constrained agents, such as edge devices. We propose an efficient federated RLHF algorithm, named Partitioned, Sign-based Stochastic Zeroth-order Policy Optimization (Par-S$^2$ZPO). The algorithm is built on zeroth-order optimization with binary perturbation, resulting in low communication, computation, and memory complexity by design. Our theoretical analysis establishes an upper bound on the convergence rate of Par-S$^2$ZPO, revealing that it is as efficient as its centralized counterpart in terms of sample complexity but converges faster in terms of policy update iterations. Our experimental results show that it outperforms a FedAvg-based RLHF on four MuJoCo RL tasks.

Recent advances in large language models have highlighted their potential to automate computational research, particularly reproducing experimental results. However, existing approaches still use fixed sequential agent pipelines with weak global coordination, which limits their robustness and overall performance. In this work, we propose Hierarchical Research Agent System (HiRAS), a hierarchical multi-agent framework for end-to-end experiment reproduction that employs supervisory manager agents to coordinate specialised agents across fine-grained stages. We also identify limitations in the reference-free evaluation of the Paper2Code benchmark and introduce Paper2Code-Extra (P2C-Ex), a refined protocol that incorporates repository-level information and better aligns with the original reference-based metric. We conduct extensive evaluation, validating the effectiveness and robustness of our proposed methods, and observing improvements, including >10\% relative performance gain beyond the previous state-of-the-art using open-source backbone models and significantly reduced hallucination in evaluation. Our work is available on GitHub: https://github.com/KOU-199024/HiRAS.

Reinforcement learning (RL) has become a prevalent paradigm for training tool calling agents, which typically requires online interactive environments. Existing approaches either rely on training data with ground truth annotations or require advanced commercial language models (LMs) to synthesize environments that keep fixed once created. In this work, we propose TRUSTEE, a data-free method training tool calling agents with dynamic environments fully simulated by free open-source LMs that can be as small as 8B, including task generation, user simulation, tool simulation and trajectory evaluation, paired with an adaptive curriculum learning mechanism that controls various aspects of the task difficulty dynamically during training. Our empirical results show that TRUSTEE brings consistent improvements across various domains and outperforms all the baselines which require extra external resources for training. These confirm that, with a sufficiently sophisticated design, even simulated environments with a local 8B LM as the backbone could set a strong baseline for tool learning, without expensive annotated data, realistic human interactions, executable tools or costly verifiable environments from human experts or commercial LMs. We hope our proposed paradigm could inspire future research on environment scaling with limited resources.

Candidate sourcing for recruiters is best viewed as a two-stage retrieval and reranking pipeline with recall as the primary objective under a limited review budget. An upstream production retriever first returns a candidate shortlist for each job description (JD), and our goal is to rerank that shortlist so that qualified candidates appear as high as possible. We present mira-embeddings-v1, a semantic reranking system for the recruitment domain that reshapes the embedding space with LLM-synthesized training data and corrects boundary confusions with a lightweight reranking head. Starting from real JDs, we build a five-stage prompt pipeline to generate diverse positive and hard negative samples that sculpt the semantic space from multiple angles. We then apply a two-round LoRA adaptation: JD--JD contrastive training followed by JD--CV triplet alignment on a heterogeneous text dataset. Importantly, these gains require no large-scale manually labeled industrial training pairs: a modest set of real JDs is expanded into supervision through LLM synthesis. Finally, a BoundaryHead MLP reranks the Top-K results to distinguish between roles that share the same title but differ in scope. On a local pool of 300 real JDs with candidates from an upstream production retriever, mira-embeddings-v1 improves Recall@50 from 68.89% (baseline) to 77.55% while lifting Precision@10 from 35.77% to 39.62%. On a supportive global pool over 44,138 candidates judged by a Qwen3-32B rubric, it achieves Recall@200 of 0.7047 versus 0.5969 for the baseline. These results show that LLM-synthesized supervision with boundary-aware reranking yields robust gains without a heavy cross-encoder.

Large language models (LLMs) are increasingly explored as scalable tools for mental health counseling, yet evaluating their safety remains challenging due to the interactional and context-dependent nature of clinical harm. Existing evaluation frameworks predominantly assess isolated responses using coarse-grained taxonomies or static datasets, limiting their ability to diagnose how harms emerge and accumulate over multi-turn counseling interactions. In this work, we introduce R-MHSafe, a role-aware mental health safety taxonomy that characterizes clinically significant harm in terms of the interactional roles an AI counselor adopts, including perpetrator, instigator, facilitator, or enabler, combined with clinically grounded harm categories. Then, we propose MHSafeEval, a closed-loop, agent-based evaluation framework that formulates safety assessment as trajectory-level discovery of harm through adversarial multi-turn interactions, guided by role-aware modeling. Using R-MHSafe and MHSafeEval, we conduct a large-scale evaluation across state-of-the-art LLMs. Our results reveal substantial role-dependent and cumulative safety failures that are systematically missed by existing static benchmarks, and show that our framework significantly improves failure-mode coverage and diagnostic granularity.

Most existing hyperspectral image super-resolution methods require modifications for different scales, limiting their flexibility in arbitrary-scale reconstruction. 2D Gaussian splatting provides a continuous representation that is compatible with arbitrary-scale super-resolution. Existing methods often rely on rasterization strategies, which may limit flexible spatial modeling. Extending them to hyperspectral image super-resolution remains challenging, as the task requires adaptive spatial reconstruction while preserving spectral fidelity. This paper proposes GaussianHSI, a Gaussian-Splatting-based framework for arbitrary-scale hyperspectral image super-resolution. We develop a Voronoi-Guided Bilateral 2D Gaussian Splatting for spatial reconstruction. After predicting a set of Gaussian functions to represent the input, it associates each target pixel with relevant Gaussian functions through Voronoi-guided selection. The target pixel is then reconstructed by aggregating the selected Gaussian functions with reference-aware bilateral weighting, which considers both geometric relevance and consistency with low-resolution features. We further introduce a Spectral Detail Enhancement module to improve spectral reconstruction. Extensive experiments on benchmark datasets demonstrate the effectiveness of GaussianHSI over state-of-the-art methods for arbitrary-scale hyperspectral image super-resolution.

Large Language Models (LLMs) have shown strong promise for mining Electronic Health Records (EHRs) by reasoning over longitudinal clinical information to capture context-rich patient trajectories. However, leveraging LLMs for structured EHRs (e.g., standardized diagnosis and medication codes) presents two key challenges. First, translating time-stamped EHR sequences into plain text can obscure both temporal structure and code identities, weakening the ability to capture code co-occurrence and longitudinal regularities. Second, unlike cohort-trained predictive models that learn a shared, task-aligned representation space across patients, LLMs are often applied in a case-isolated inference setting where each patient is processed independently without leveraging population-level patterns. To address these challenges, we introduce RePrompT, a time-aware LLM framework that integrates structured EHR encoders through prompt tuning, without modifying underlying architectures. Specifically, RePrompT recurrently incorporates latent states from prior visits to preserve longitudinal information, and injects population-level information through trainable prompt tokens derived from a cohort-trained, task-aligned EHR encoder. Experiments on MIMIC-III and MIMIC-IV demonstrate that RePrompT consistently outperforms both EHR-based and LLM-based baselines across multiple clinical prediction tasks.

We address the challenge of point cloud registration using color information, where traditional methods relying solely on geometric features often struggle in low-overlap and incomplete scenarios. To overcome these limitations, we propose GeGS-PCR, a novel two-stage method that combines geometric, color, and Gaussian information for robust registration. Our approach incorporates a dedicated color encoder that enhances color features by extracting multi-level geometric and color data from the original point cloud. We introduce the \textbf{Ge}ometric-3D\textbf{GS} module, which encodes the local neighborhood information of colored superpoints to ensure a globally invariant geometric-color context. Leveraging LORA optimization, we maintain high performance while preserving the expressiveness of 3DGS. Additionally, fast differentiable rendering is utilized to refine the registration process, leading to improved convergence. To further enhance performance, we propose a joint photometric loss that exploits both geometric and color features. This enables strong performance in challenging conditions with extremely low point cloud overlap. We validate our method by colorizing the Kitti dataset as ColorKitti and testing on both Color3DMatch and Color3DLoMatch datasets. Our method achieves state-of-the-art performance with \textit{Registration Recall} at 99.9\%, \textit{Relative Rotation Error} as low as 0.013, and \textit{Relative Translation Error} as low as 0.024, improving precision by at least a factor of 2.

Point-based Neural Networks (PNNs) have become a key approach for point cloud processing. However, a core operation in these models, Farthest Point Sampling (FPS), often introduces significant inference latency, especially for large-scale processing. Despite existing CUDA- and hardware-level optimizations, FPS remains a major bottleneck due to exhaustive computations across multiple network layers in PNNs, which hinders scalability. Through systematic analysis, we identify three substantial redundancies in FPS, including unnecessary full-cloud computations, redundant late-stage iterations, and predictable inter-layer outputs that make later FPS computations avoidable. To address these, we propose \textbf{\textit{FlashFPS}}, a hardware-agnostic, plug-and-play framework for FPS acceleration, composed of \textit{FPS-Prune} and \textit{FPS-Cache}. \textit{FPS-Prune} introduces candidate pruning and iteration pruning to reduce redundant computations in FPS while preserving sampling quality, and \textit{FPS-Cache} eliminates layer-wise redundancy via cache-and-reuse. Integrated into existing CUDA libraries and state-of-the-art PNN accelerators, \textit{FlashFPS} achieves 5.16$\times$ speedup over the standard CUDA baseline on GPU and 2.69$\times$ on PNN accelerators, with negligible accuracy loss, enabling efficient and scalable PNN inference. Codes are released at https://github.com/Yuzhe-Fu/FlashFPS.

Many benchmarks show that large language models can answer direct questions about culture. We study a different question: do they also change how they speak when culture is only implied by the situation? We evaluate 60 culturally grounded conversational scenarios across five languages in three conditions: a neutral baseline (Prompt A), an explicit cultural instruction (Prompt B), and implicit situational cueing (Prompt C). We score responses on 12 pragmatic features covering deference to authority, individual-versus-group framing, and uncertainty management. We define Pragmatic Context Sensitivity (PCS) as the fraction of the Prompt A->B shift that reappears under Prompt A->C. Across four deployed LLMs and five languages (English, German, Hindi, Nepali, Urdu), the primary stable-only PCS mean is 0.196 (SD = 0.113), indicating that the models recover only about one-fifth of the pragmatic shift they can produce when instructed explicitly. Transfer is strongest for authority-related cues (0.299) and weakest for individual-versus-group framing (0.120). Uncertainty-related behaviour is mixed: hedging density exhibits negative explicit gaps in all five languages, suggesting that alignment training actively suppresses the target behaviour. Because Hindi and Urdu share core grammar yet index distinct cultural communities, we use them as a natural control; a paired analysis finds no reliable baseline difference (t = 0.96, p = 0.339, dz = 0.06), suggesting that models respond primarily to linguistic structure rather than to the cultural associations a language carries. We argue that multilingual cultural pragmatics is an explicit-versus-implicit deployment problem, not only a factual knowledge problem.

Program debloating aims to remove unused code to reduce performance overhead, attack surfaces, and maintenance costs. Over time, debloating has evolved across multiple layers (container, library, and application), each building on the principles of application-level debloating. Despite its central role, application-level debloating continues to rely on imperfect proxies for measuring performance, such as test-case-driven evaluation for correctness, code size for runtime efficiency, and gadget count reduction for estimating security posture. While there is widespread skepticism about using such imperfect proxies, the community still lacks standardized methodologies or benchmarks to assess the true performance of application-level software debloating. This experience paper aims to address the gap. We revisit the foundations of application-level debloating through a ground-truth-based evaluation paradigm. Our analysis of eight state-of-the-art debloaters - Blade, Chisel, Cov, CovA, Lmcas, Trimmer, Occam, and Razor - uncovers insights previously unattainable through traditional evaluations. These tools collectively span the spectrum of source-to-source, IR-to-IR, and binary-to-binary transformation paradigms, characterizing a holistic reassessment across abstraction levels. Our analysis reveals that while dynamic analysis-based tools often remove up to 94% of code that should be retained, static analysis-based approaches exhibit the opposite behavior, showing high false retention rates due to coarse-grained dependency over-approximation. Additionally, static analyses may add code by introducing specialized variants of functions. False retentions and removals not only cause functional incorrectness but may also lead to systematic inconsistency, robustness failures, and exploitable vulnerabilities.

The validity screen (Cacioli, 2026d, 2026e) classifies LLM confidence signals as Valid, Indeterminate, or Invalid. We test whether these classifications predict selective prediction performance. Twenty frontier LLMs from seven families were evaluated on 524 items across six cognitive tracks. Valid models show mean Type 2 AUROC = .624 (SD = .048). Invalid models show mean AUROC = .357 (SD = .231). Cohen's d = 2.81, p = .002. The tiers order monotonically: Invalid (.357) < Indeterminate (.554) < Valid (.624). Split-half cross-validation yields median d = 1.77, P(d > 0) = 1.0 across 1,000 splits. The three-tier classification accounts for 47% of the variance in AUROC. DeepSeek-R1 drops from 85.3% accuracy at full coverage to 11.3% at 10% coverage. The screen predicts the criterion. For selective prediction, the screen matters.

Recent advances in large language models for test case generation have improved branch coverage via prompt-engineered mutations. However, they still lack principled mechanisms for steering models toward specific high-risk execution branches, limiting their effectiveness for discovering subtle bugs and security vulnerabilities. We propose GLMTest, the first program structure-aware LLM framework for targeted test case generation that seamlessly integrates code property graphs and code semantics using a graph neural network and a language model to condition test case generation on execution branches. This structured conditioning enables controllable and branch-targeted test case generation, thereby potentially enhancing bug and security risk discovery. Experiments on real-world projects show that GLMTest built on a Qwen2.5-Coder-7B-Instruct model improves branch accuracy from 27.4% to 50.2% on TestGenEval benchmark compared with state-of-the-art LLMs, i.e., Claude-Sonnet-4.5 and GPT-4o-mini.

LLM confidence signals are used for abstention, routing, and safety-critical decisions. No standard practice exists for checking whether a confidence signal carries item-level information before building on it. We transfer the validity screening principle from clinical personality assessment (PAI, MMPI-3) as a portable protocol for benchmark-based LLM confidence data. The protocol specifies three core indices (L, Fp, RBS), a structural indicator (TRIN), and an item-sensitivity statistic, computed from a single 2x2 contingency table. A three-tier classification system (Invalid, Indeterminate, Valid) draws on four clinical traditions. Validated on 20 frontier LLMs across 524 items, four models are classified Invalid, two Indeterminate. Valid-profile models show mean r = .18 (15/16 significant). Invalid-profile models show mean r = -.20 (d = 2.48). Cross-benchmark validation on 18 models using MMLU with verbalized confidence and on external data from Yang et al. (2024) confirms the screen transfers across benchmarks and probe formats. All data and code: https://github.com/synthiumjp/validity-scaling-llm

Resting-state functional magnetic resonance imaging (fMRI) has emerged as a cornerstone for psychiatric diagnosis, yet most approaches rely on pairwise brain cortical or sub-cortical connectivities that overlooks higher-order interactions (HOIs) central to complex brain dynamics. While hypergraph methods encode HOIs through predefined hyperedges, their construction typically relies on heuristic similarity metrics and does not explicitly characterize whether interactions are synergy- or redundancy-dominated. In this paper, we introduce $O$-information, a signed measure that characterizes the informational nature of HOIs, and integrate third- and fourth-order $O$-information into a unified multi-view information bottleneck framework for fMRI-based psychiatric diagnosis. To enable scalable $O$-information estimation, we further develop two independent acceleration strategies: a Gaussian analytical approximation and a randomized matrix-based Rényi entropy estimator, achieving over a 30-fold computational speedup compared with conventional estimators. Our tri-view architecture systematically fuses pairwise, triadic, and tetradic brain interactions, capturing comprehensive brain connectivity while explicitly penalizing redundancy. Extensive evaluation across four benchmark datasets (REST-meta-MDD, ABIDE, UCLA, ADNI) demonstrates consistent improvements, outperforming 11 baseline methods including state-of-the-art graph neural network (GNN) and hypergraph based approaches. Moreover, our method reveals interpretable region-level synergy-redundancy patterns which are not explicitly characterized by conventional hypergraph formulations.

Existing works on large language model (LLM) decomposition mainly focus on improving performance on downstream tasks, but they ignore the poor parallel inference performance when trying to scale up the model size. To mitigate this important performance issue, this paper introduces DeInfer, a high-performance inference system dedicated to parallel inference of decomposed LLMs. It consists of multiple optimizations to maximize performance and be compatible with state-of-the-art optimization techniques. Extensive experiments are carried out to evaluate DeInfer's performance, where the results demonstrate its superiority, suggesting it can greatly facilitate the parallel inference of decomposed LLMs.

Automating operations research (OR) with large language models (LLMs) remains limited by hand-crafted reasoning--execution workflows. Complex OR tasks require adaptive coordination among problem interpretation, mathematical formulation, solver selection, code generation, and iterative debugging. To address this limitation, we propose EvoOR-Agent, a co-evolutionary framework for automated optimization. The framework represents agent workflows as activity-on-edge (AOE)-style networks, making workflow topology, execution dependencies, and alternative reasoning paths explicit. On this representation, the framework maintains an architecture graph and evolves a population of reasoning individuals through graph-mediated path-conditioned recombination, multi-granularity semantic mutation, and elitist population update. A knowledge-base-assisted experience-acquisition module further injects reusable OR practices into initialization and semantic variation. Empirical results on heterogeneous OR benchmarks show that the proposed framework consistently improves over zero-shot LLMs, fixed-pipeline OR agents, and representative evolutionary agent frameworks. Case studies and ablation analyses further indicate that explicit architecture evolution and graph-supported reasoning-trajectory search contribute to both performance improvement and structural interpretability. These results suggest that treating agent architectures and reasoning trajectories as evolvable objects provides an effective route toward adaptive and interpretable automated optimization.

Clinical personality assessment screens response validity before interpreting substantive scales. LLM evaluation does not. We apply the validity scaling framework from the PAI and MMPI-3 to metacognitive probe data from 20 frontier models across 524 items. Six validity indices are operationalised: L (maintaining confidence on errors), K (betting on errors), F (withdrawing consensus-endorsed items), Fp (withdrawing correct answers), RBS (inverted monitoring), and TRIN (fixed responding). A tiered classification system identifies four models as construct-level invalid and two as elevated. Valid-profile models produce item-sensitive confidence (mean r = .18, 14 of 16 significant). Invalid-profile models do not (mean r = -.20, d = 2.17, p = .001). Chain-of-thought training produces two opposite response distortions. Two latent dimensions account for 94.6% of index variance. Companion papers extract a portable screening protocol (Cacioli, 2026e) and validate it against selective prediction (Cacioli, 2026f). All data and code: https://github.com/synthiumjp/validity-scaling-llm

While distributed device-edge speculative decoding enhances resource utilization across heterogeneous nodes, its performance is often bottlenecked by conventional token-level verification strategies. Such rigid alignment leads to excessive rejections, significantly diminishing the accepted sequence length and increasing interaction rounds under fluctuating wireless conditions. In this paper, we propose WISV (Wireless-Informed Semantic Verification), a novel distributed speculative decoding framework that goes beyond strict token-level matching via a channel-aware semantic acceptance policy. WISV integrates a lightweight decision head into the edge-side target LLM to dynamically evaluate speculative tokens by synthesizing high-dimensional hidden representations with instantaneous channel state information (CSI). To optimize the trade-off between verification fidelity and communication overhead, we further design two tailored communication protocols: full-hidden upload and mismatch-first selective-hidden upload. Extensive simulations using a 1B drafter and an 8B target model demonstrate that WISV achieves up to a 60.8% increase in accepted length, a 37.3% reduction in interaction rounds, and a 31.4% improvement in end-to-end latency compared to vanilla speculative decoding across tested settings, while maintaining a negligible task accuracy drop (<1%). Finally, we validate WISV on a hardware testbed comprising an NVIDIA Jetson AGX Orin and an A40-equipped server, confirming its real-world efficacy in accelerating edge-deployed LLM inference.

Large Language Models (LLMs) have transformed software development and AI applications. While LLMs are designed for text processing, LLM agents extend this capability by enabling autonomous actions, tool use, and multi-step task completion. As this field grows, developers face new challenges in debugging these complex systems. To address this challenge, we present the first empirical study on bug fix patterns in LLM agents. We study buggy posts and code snippets from three platforms: Stack Overflow, GitHub, and HuggingFace Forums. We examine their fix patterns, the components where fixes are applied, and the programming languages and frameworks involved. Furthermore, we introduce AgentDefect, the first benchmark dataset for bugs in LLM agents. The dataset contains 37 runtime buggy instances along with fixed code and test files. Finally, we present SelfHeal, a multi-agent system designed to fix bugs in LLM agents. The system leverages two independent ReAct agents: the fix agent and the critic agent. These agents use tools that provide both internal knowledge (fix rules) and external knowledge (web search) to propose and validate fixes. Our evaluation shows that SelfHeal with Gemini 3 Pro as the backbone LLM outperforms both baseline and state-of-the-art approaches by a significant margin.

Reliable deployment of language models requires two capabilities that appear distinct but share a common geometric foundation: predicting whether a model will accept targeted behavioral control, and detecting when its internal structure degrades. We show that geometric stability, the consistency of a representation's pairwise distance structure, addresses both. Supervised Shesha variants that measure task-aligned geometric stability predict linear steerability with near-perfect accuracy ($ρ= 0.89$-$0.97$) across 35-69 embedding models and three NLP tasks, capturing unique variance beyond class separability (partial $ρ= 0.62$-$0.76$). A critical dissociation emerges: unsupervised stability fails entirely for steering on real-world tasks ($ρ\approx 0.10$), revealing that task alignment is essential for controllability prediction. However, unsupervised stability excels at drift detection, measuring nearly $2\times$ greater geometric change than CKA during post-training alignment (up to $5.23\times$ in Llama) while providing earlier warning in 73\% of models and maintaining a $6\times$ lower false alarm rate than Procrustes. Together, supervised and unsupervised stability form complementary diagnostics for the LLM deployment lifecycle: one for pre-deployment controllability assessment, the other for post-deployment monitoring.

Games offer a compelling paradigm for developing general reasoning capabilities in language models, as they naturally demand strategic planning, probabilistic inference, and adaptive decision-making. However, existing self-play approaches rely solely on terminal game outcomes, providing no mechanism to distinguish transferable reasoning patterns from game-specific heuristics. We present STRATAGEM, which addresses two fundamental barriers to reasoning transfer: domain specificity, where learned patterns remain anchored in game semantics, and contextual stasis, where static game contexts fail to cultivate progressive reasoning. STRATAGEM selectively reinforces trajectories exhibiting abstract, domain-agnostic reasoning through a Reasoning Transferability Coefficient, while incentivizing adaptive reasoning development via a Reasoning Evolution Reward. Experiments across mathematical reasoning, general reasoning, and code generation benchmarks demonstrate substantial improvements, with particularly strong gains on competition-level mathematics where multi-step reasoning is critical. Ablation studies and human evaluation confirm that both components contribute to transferable reasoning.

KV cache memory is the dominant bottleneck for long-context LLM inference. Existing compression methods each act on a single axis of the four-dimensional KV tensor -- token eviction (sequence), quantization (precision), low-rank projection (head dimension), or cross-layer sharing -- but apply the same recipe to every layer. We show that this homogeneity leaves accuracy on the table: different layers respond very differently to each compression operation, and the optimal per-layer mix of eviction and quantization is far from uniform. We propose MoE-nD, a mixture-of-experts framework that routes each layer to its own (eviction-ratio, K-bits, V-bits) tuple under a global memory budget. An offline-calibrated greedy solver chooses the routing that minimizes predicted quality loss; at inference time, per-layer heterogeneous eviction and quantization are applied jointly through a single attention patch. On a 4-task subset of LongBench-v1 (16k inputs, n=50 per task, adapted reasoning-model protocol; see section Experiments), MoE-nD's hetero variant matches our uncompressed 1.9~GB baseline at 14x compression (136~MB) while every other compressed baseline we tested (1d, 2d_uniform, 2d) at comparable or smaller memory stays under 8/100. The gains hold on AIME reasoning benchmarks (+6 to +27 pts over the strongest per-layer-quantization baseline across eight configurations). Two null results -- MATH-500 and LongBench's TREC -- share a principled cause (short inputs, solver picks keep=1.0 on most layers), cleanly characterizing when per-layer eviction routing has headroom to help.

Predictions from machine learning algorithms can vary across random seeds, inducing instability in downstream debiased machine learning estimators. We formalize random seed stability via a concentration condition and prove that subbagging guarantees stability for any bounded-outcome regression algorithm. We introduce a new cross-fitting procedure, adaptive cross-bagging, which simultaneously eliminates seed dependence from both nuisance estimation and sample splitting in debiased machine learning. Numerical experiments confirm that the method achieves the targeted level of stability whereas alternatives do not. Our method incurs a small computational penalty relative to standard practice whereas alternative methods incur large penalties.

In cooperative teams where agents act in a fixed order and share a single team reward, it is hard to know how much each agent contributed, and harder still when agents are updated one at a time because data collected earlier no longer reflects the new policies. We introduce the Sequential Aristocrat Utility (SeqAU), the unique per-agent learning signal that maximizes the individual learnability of each agent's action, extending the classical framework of Wolpert and Tumer (2002) to this sequential setting. From SeqAU we derive CAPO (Counterfactual Advantage Policy Optimization), a critic-free policy-gradient algorithm. CAPO fits a per-agent reward decomposition from group rewards and computes the per-agent advantage in closed form plus a handful of forward passes through the current policy, requiring no extra environment calls beyond the initial batch. We give analytic bias and variance bounds and validate them on a controlled sequential bandit, where CAPO's advantage over standard baselines grows with the team size. The framework is general; multi-LLM pipelines are a natural deployment target.

Safety alignment in large language models is remarkably shallow: it is concentrated in the first few output tokens and reversible by fine-tuning on as few as 100 adversarial examples. This fragility becomes critical in real-world deployment, where models undergo sequential adaptation across domains such as medicine, law, and code, causing safety guardrails to erode cumulatively. Yet all existing safety-preserving methods target only single-task fine-tuning, leaving the multi-domain sequential setting entirely unaddressed. We introduce SafeAnchor, a framework that anchors safety in place throughout continual adaptation. SafeAnchor first identifies low-rank safety subspaces in LoRA parameter space via Fisher Information eigendecomposition, then constrains domain-specific gradient updates to the orthogonal complement of these subspaces, and finally monitors for residual safety drift with threshold-triggered corrective replay. Evaluated on Llama-2-7B-Chat and Mistral-7B-Instruct across a three-domain pipeline and eight benchmarks, SafeAnchor retains 93.2% of original safety alignment, outperforming all baselines by 18-42 points, while matching unconstrained fine-tuning to within 1.5 points on domain tasks.

Reconfigurable intelligent surfaces (RISs) modify signal reflections to enhance wireless communication capabilities. Classical RIS phase optimization is highly non convex and challenging in dynamic environments due to high interference and user mobility. Here we propose a hierarchical multi-objective quantum metalearning algorithm that switches among specific quantum paths based on historical success, energy cost, and current data rate. Candidate RIS control directions are arranged as switch paths between quantum neural network layers to minimize inference, and a scoring mechanism selects the top performing paths per layer. Instead of merely storing past successful settings of the RIS and picking the closest match when a new problem is encountered, the algorithm learns how to select and recombine the best parts of different solutions to solve new scenarios. In our model, high-dimensional RIS scenario features are compressed into a quantum state using the tensor product, then superimposed during quantum path selection, significantly improving quantum computational advantage. Results demonstrate efficient performance with enhanced spectral efficiency, convergence rate, and adaptability.

Retrieval-Augmented Generation (RAG) systems depend on the geometric properties of vector representations to retrieve contextually appropriate evidence. When source documents interleave multiple topics within contiguous text, standard vectorization produces embedding spaces in which semantically distinct content occupies overlapping neighborhoods. We term this condition semantic entanglement. We formalize entanglement as a model-relative measure of cross-topic overlap in embedding space and define an Entanglement Index (EI) as a quantitative proxy. We argue that higher EI constrains attainable Top-K retrieval precision under cosine similarity retrieval. To address this, we introduce the Semantic Disentanglement Pipeline (SDP), a four-stage preprocessing framework that restructures documents prior to embedding. We further propose context-conditioned preprocessing, in which document structure is shaped by patterns of operational use, and a continuous feedback mechanism that adapts document structure based on agent performance. We evaluate SDP on a real-world enterprise healthcare knowledge base comprising over 2,000 documents across approximately 25 sub-domains. Top-K retrieval precision improves from approximately 32% under fixed-token chunking to approximately 82% under SDP, while mean EI decreases from 0.71 to 0.14. We do not claim that entanglement fully explains RAG failure, but that it captures a distinct preprocessing failure mode that downstream optimization cannot reliably correct once encoded into the vector space.

Legal practitioners and judicial institutions face an ever-growing volume of case-law documents characterised by formalised language, lengthy sentence structures, and highly specialised terminology, making manual triage both time-consuming and error-prone. This work presents a lightweight yet high-accuracy framework for citation-treatment classification that pairs lemmatisation-based preprocessing with subword-aware FastText embeddings and a multi-kernel one-dimensional Convolutional Neural Network (CNN). Evaluated on a publicly available corpus of 25,000 annotated legal documents with a 75/25 training-test partition, the proposed system achieves 97.26% classification accuracy and a macro F1-score of 96.82%, surpassing established baselines including fine-tuned BERT, Long Short-Term Memory (LSTM) with FastText, CNN with random embeddings, and a Term Frequency-Inverse Document Frequency (TF-IDF) k-Nearest Neighbour (KNN) classifier. The model also attains the highest Area Under the Receiver Operating Characteristic (AUC-ROC) curve of 97.83% among all compared systems while operating with only 5.1 million parameters and an inference latency of 0.31 ms per document - more than 13 times faster than BERT. Ablation experiments confirm the individual contribution of each pipeline component, and the confusion matrix reveals that residual errors are confined to semantically adjacent citation categories. These findings indicate that carefully designed convolutional architectures represent a scalable, resource-efficient alternative to heavyweight transformers for intelligent legal document analysis.

Despite their empirical success, how diffusion models generalize remains poorly understood from a mechanistic perspective. We demonstrate that diffusion models trained with flow-matching objectives exhibit grokking--delayed generalization after overfitting--on modular addition, enabling controlled analysis of their internal computations. We study this phenomenon across two levels of data regime. In a single-image regime, mechanistic dissection reveals that the model implements modular addition by composing periodic representations of individual operands. In a diverse-image regime with high intraclass variability, we find that the model leverages its iterative sampling process to partition the task into an arithmetic computation phase followed by a visual denoising phase, separated by a critical timestep threshold. Our work provides the mechanistic decomposition of algorithmic learning in diffusion models, revealing how these models bridge continuous pixel-space generation and discrete symbolic reasoning.

We introduce Prior-Fitted Functional Flows, a generative foundation model for pharmacokinetics that enables zero-shot population synthesis and individual forecasting without manual parameter tuning. We learn functional vector fields, explicitly conditioned on the sparse, irregular data of an entire study population. This enables the generation of coherent virtual cohorts as well as forecasting of partially observed patient trajectories with calibrated uncertainty. We construct a new open-access literature corpus to inform our priors, and demonstrate state-of-the-art predictive accuracy on extensive real-world datasets.

Peer review is central to scientific publishing, yet reviewers frequently include claims that are subjective, rhetorical, or misaligned with the submitted work. Assessing whether review statements are factual and verifiable is crucial for fairness and accountability. At the scale of modern conferences and journals, manually inspecting the grounding of such claims is infeasible. We present Peerispect, an interactive system that operationalizes claim-level verification in peer reviews by extracting check-worthy claims from peer reviews, retrieving relevant evidence from the manuscript, and verifying the claims through natural language inference. Results are presented through a visual interface that highlights evidence directly in the paper, enabling rapid inspection and interpretation. Peerispect is designed as a modular Information Retrieval (IR) pipeline, supporting alternative retrievers, rerankers, and verifiers, and is intended for use by reviewers, authors, and program committees. We demonstrate Peerispect through a live, publicly available demo (https://app.reviewer.ly/app/peerispect) and API services (https://github.com/Reviewerly-Inc/Peerispect), accompanied by a video tutorial (https://www.youtube.com/watch?v=pc9RkvkUh14).

Constitution-conditioned post-training can be analysed as a structured perturbation of a model's learned representational geometry. We introduce ATLAS, a geometry-first program that traces constitution-induced hidden-state structure across charts, models, and substrates. Instead of treating the relevant unit as a single behaviour, neuron, vector, or patch, ATLAS tests a local chart whose tangent structure, occupancy distribution, and behavioural coupling can be measured under system change. On Gemma, the anchored source-local chart captures 310 / 320 reviewed source rows and all 84 / 84 reviewed score-flip rows, but compact exact-patch sufficiency does not close, so the exportable unit is the broader source-defined family. Freezing that family, we re-identify a target-local realisation in an unadapted Phi model, where the fully adjudicated confirmatory contrast separates with AUC 0.984 and mean gap 5.50. In held-out ALM8 mouse frontal-cortex perturbation data, the same source-defined family receives support across 5/5 folds, with mean held-out AUC 0.72 and mean fold gap 4.50. A multiple-choice analysis provides the main boundary: nearby target-local signals can appear without source-faithful closure. The resulting correspondence is not coordinate identity, site identity, or a target-side mediation theorem. It is geometric recurrence under redistribution: written constitutions can induce recoverable latent geometry whose organisation remains detectable across model and substrate changes while its local coordinates, occupancy, and behavioural expression shift.

Continuous Integration and Continuous Deployment (CI/CD) have become fundamental to modern software development, with GitHub Actions (GHA) emerging as a dominant automation platform. In this study, we analyze real-world execution records of GHA, examining how developers react to workflow failures, how these workflows are utilized by projects, and how these aspects relate to project characteristics. We quantitatively analyze 258,300 workflow run records from 952 repositories and perform an in-depth qualitative analysis of 21 selected, diverse GitHub repositories to understand how maintainers and contributors interact with workflow results. We identify three distinct failure response patterns, observe that higher usage intensity of GHA workflows correlates with lower failure rates, and uncover a configuration-usage gap where the presence of configuration files masks disabled or unused workflows. Moreover, our qualitative analysis of relationships between project characteristics and utilization patterns yields five hypotheses for future validation.

We introduce the Semantic Density Effect (SDE): the empirical finding that prompts carrying higher semantic information per token consistently produce more accurate, focused, and less hallucinated outputs across all major LLM families. SDE is defined as the ratio of semantically loaded tokens to total prompt tokens, adjusted for redundancy and concreteness. Unlike prior prompt optimization techniques that add tokens (Chain of Thought), duplicate the prompt (Prompt Repetition), or reorder components (Instruction Placement Effect), SDE improves performance by removing or replacing low-information tokens while preserving or sharpening the semantic signal. Evaluated across five frontier models and seven benchmarks, ultra-dense prompts (SDE > 0.80) outperform diluted counterparts by an average of +8.4 percentage points with 0 additional tokens and 0 latency overhead. Combined with Instruction Placement Effect (IPE), the gain reaches +11.7 percentage points

Large Language Model (LLM)-based Multi-Agent Systems (MAS) enable complex problem-solving but introduce significant debugging challenges, characterized by long interaction traces, inter-agent dependencies, and delayed error manifestation. Existing diagnostic approaches often rely on expensive expert annotation or ''LLM-as-a-judge'' paradigms, which struggle to pinpoint decisive error steps within extended contexts. In this paper, we introduce ErrorProbe, a self-improving framework for semantic failure attribution that identifies responsible agents and the originating error step. The framework operates via a three-stage pipeline: (1) operationalizing the MAS failure taxonomy to detect local anomalies, (2) performing symptom-driven backward tracing to prune irrelevant context, and (3) employing a specialized multi-agent team (Strategist, Investigator, Arbiter) to validate error hypotheses through tool-grounded execution. Crucially, ErrorProbe maintains a verified episodic memory that updates only when error patterns are confirmed by executable evidence, without the need for annotation. Experiments across the TracerTraj and Who&When benchmarks demonstrate that ErrorProbe significantly outperforms baselines, particularly in step-level localization, while the verified memory enables robust cross-domain transfer without retraining.

Video-to-music (V2M) is the fundamental task of creating background music for an input video. Recent V2M models achieve audiovisual alignment by typically relying on visual conditioning alone and provide limited semantic and stylistic controllability to the end user. In this paper, we present Video-Robin, a novel text-conditioned video-to-music generation model that enables fast, high-quality, semantically aligned music generation for video content. To balance musical fidelity and semantic understanding, Video-Robin integrates autoregressive planning with diffusion-based synthesis. Specifically, an autoregressive module models global structure by semantically aligning visual and textual inputs to produce high-level music latents. These latents are subsequently refined into coherent, high-fidelity music using local Diffusion Transformers. By factoring semantically driven planning into diffusion-based synthesis, Video-Robin enables fine-grained creator control without sacrificing audio realism. Our proposed model outperforms baselines that solely accept video input and additional feature conditioned baselines on both in-distribution and out-of-distribution benchmarks with a 2.21x speed in inference compared to SOTA. We will open-source everything upon paper acceptance.

Exploration is a cornerstone of learning from experience: it enables agents to find solutions to complex problems, generalize to novel ones, and scale performance with test-time compute. In this paper, we present a framework for post-training language models (LMs) that explicitly encourages optimistic exploration and promotes a synergy between exploration and exploitation. The central idea is to train the LM to generate sets of responses that are collectively accurate under the reward function and exploratory in their reasoning strategies. We first develop a general recipe for optimizing LMs with set reinforcement learning (set RL) under arbitrary objective functions, showing how standard RL algorithms can be adapted to this setting through a modification to the advantage computation. We then propose Polychromic Exploratory Policy Optimization (Poly-EPO), which instantiates this framework with an objective that explicitly synergizes exploration and exploitation. Across a range of reasoning benchmarks, we show that Poly-EPO improves generalization, as evidenced by higher pass@$k$ coverage, preserves greater diversity in model generations, and effectively scales with test-time compute.

Text-to-SQL systems often struggle with deep contextual understanding, particularly for complex queries with subtle requirements. We present PV-SQL, an agentic framework that addresses these failures through two complementary components: Probe and Verify. The Probe component iteratively generates probing queries to retrieve concrete records from the database, resolving ambiguities in value formats, column semantics, and inter-table relationships to build richer contextual understanding. The Verify component employs a rule-based method to extract verifiable conditions and construct an executable checklist, enabling iterative SQL refinement that effectively reduces missing constraints. Experiments on the BIRD benchmarks show that PV-SQL outperforms the best text-to-SQL baseline by 5% in execution accuracy and 20.8% in valid efficiency score while consuming fewer tokens.

LLMs are increasingly deployed in dynamic, real-world settings, where the distribution of user prompts can shift substantially over time as new tasks, prompts, and users are introduced to a deployed model. Such natural prompt distribution shift poses a major challenge to LLM reliability, particularly for specialized models designed for narrow domains or user populations. Despite attention to out-of-distribution robustness, there is very limited exploration of measuring natural prompt distribution shift in prior work, and its impact on deployed LLMs remains poorly understood. We introduce the LLM Evaluation under Natural prompt Shift (LENS) framework: a data-centric approach for quantifying natural prompt distribution shift and evaluating its effect on the performance of deployed LLMs. We perform a large-scale evaluation using 192 real-world post-deployment prompt shift settings over time, user group, and geographic axes, training a total of 81 models on 4.68M training prompts, and evaluating on 57.6k prompts. We find that even moderate shifts in user prompt behavior correspond with large performance drops (73% average loss) in deployed LLMs. This performance degradation is particularly prevalent when users from different latent groups and geographic regions interact with models and is correlated with natural prompt distribution shift over time. We systematically characterize how LLM instruction following ability degrades over time and between user groups. Our findings highlight the critical need for data-driven monitoring to ensure LLM performance remains stable across diverse and evolving user populations.

Summarizing deeply nested discussion threads requires handling interleaved replies, quotes, and overlapping topics, which standard LLM summarizers struggle to capture reliably. We introduce ThreadSumm, a multi-stage LLM framework that treats thread summarization as a hierarchical reasoning problem over explicit aspect and content unit representations. Our method first performs content planning via LLM-based extraction of discourse aspects and Atomic Content Units, then applies sentence ordering to construct thread-aware sequences that surface multiple viewpoints rather than a single linear strand. On top of these interpretable units, ThreadSumm employs a Tree of Thoughts search that generates and scores multiple paragraph candidates, jointly optimizing coherence and coverage within a unified search space. With this multi-proposal and iterative refinement design, we show improved performance in generating logically structured summaries compared to existing baselines, while achieving higher aspect retention and opinion coverage in nested discussions.

A number of optimization algorithms have been inspired by the physics of Newtonian motion. Here, we ask the question: do algorithms themselves obey some ``natural laws of motion,'' and can they be derived by an application of these laws? We explore this question by positing the theory that optimization algorithms may be considered as some manifestation of hidden algorithm primitives that obey certain universal non-Newtonian dynamics. This natural physics of optimization is developed by equating the terminal transversality conditions of an optimal control problem to the generalized Karush/John-Kuhn-Tucker conditions of an optimization problem. Through this equivalence formulation, the data functions of a given constrained optimization problem generate a natural vector field that permeates an entire hidden space with information on the optimality conditions. An ``action-at-a-distance'' operation via a Pontryagin-type minimum principle produces a local action to deliver a globalized result by way of a Hamilton-Jacobi inequality. An inverse-optimal algorithm is generated by performing control jumps that dissipate quantized ``energy'' defined by a search Lyapunov function. Illustrative applications of the proposed theory show that a large number of algorithms can be generated and explained in terms of the new mathematical physics of optimization.

Large language models exhibit impressive cross-lingual capabilities. However, prior work analyzes this phenomenon through isolated factors and at sparse points during training, limiting our understanding of how cross-lingual generalization emerges--particularly in the early phases of learning. To study the early trajectory of linguistic and translation capabilities, we pretrain a multilingual 1.7B model on nine diverse languages, capturing checkpoints at a much finer granularity. We further introduce a novel word-level translation dataset and trace how translation develops over training through behavioral analyses, model-component analysis, and parameter-based ablations. We find that the model quickly acquires basic linguistic capabilities in parallel with token-level copying, while translation develops in two distinct phases: an initial phase dominated by copying and surface-level similarities, and a second phase in which more generalizing translation mechanisms are developed while copying is refined. Together, these findings provide a fine-grained view of how cross-lingual generalization develops during multilingual pretraining.

Wales' political landscape has been marked by growing accusations of bias in Welsh media. This paper takes the first computational step toward testing those claims by examining Nation.Cymru, a prominent Welsh political news outlet. I use a two-stage natural language processing (NLP) pipeline: (1) a robustly optimized BERT approach (RoBERTa) bias detector for efficient bias discovery and (2) a large language model (LLM) for target-attributed sentiment classification of bias labels from (1). A primary analysis of 15,583 party mentions across 2022-2026 news articles finds that Reform UK attracts biased framing at twice the rate of Plaid Cymru and over three times as negative in mean sentiment (p<0.001). A secondary analysis across four parties across both news and opinion articles shows that Plaid Cymru is the outlier, receiving markedly more favourable framing than any other party. These findings provide evidence of measurable differential framing in a single Welsh political media outlet, supporting calls for a broader review of Welsh media coverage. Furthermore, the two-stage pipeline offers a low-cost, replicable framework for extending this analysis to other Welsh outlets, as well as media ecosystems outside of Wales.

Serving large language models under latency service-level objectives (SLOs) is a configuration-heavy systems problem with an unusually failure-prone search space: many plausible configurations crash outright or miss user-visible latency targets, and standard black-box optimizers treat these failures as wasted trials. We present SLO-Guard, a crash-aware autotuner for vLLM serving that treats crashes as first-class observations. SLO-Guard combines a feasible-first Thermal Budget Annealing (TBA) exploration phase with a warm-started Tree-structured Parzen Estimator (TPE) exploitation phase; the handoff replays all exploration history, including crashes encoded as extreme constraint violations. We additionally contribute a configuration-repair pass, a GPU-aware KV-cache memory guard, and a four-category crash taxonomy. We evaluate SLO-Guard on Qwen2-1.5B served with vLLM 0.19 on an NVIDIA A100 40GB. Across a pre-specified five-seed study, both SLO-Guard and uniform random search attain 75/75 feasibility with zero crashes under the corrected concurrent harness, and are statistically tied on best-achieved latency (Mann-Whitney two-sided p=0.84). SLO-Guard's advantage is in budget consistency: more trials in the fast-serving regime (10.20 vs. 7.40 out of 15; one-sided p=0.014) and higher post-handoff consistency (0.876 vs. 0.539; p=0.010). Under concurrent load, SLO-Guard's cross-seed standard deviation on best latency is 4.4x tighter than random search's (2.26 ms vs. 10.00 ms). A harness-replication analysis shows that the consistency findings survive an independent sequential-dispatch measurement condition. The central claim is not that SLO-Guard finds a better final configuration, but that it spends a fixed tuning budget more predictably once the fast regime has been found.

This work addresses the challenge of disseminating reusable artificial intelligence (AI) models accompanied by AI documentation (a.k.a., AI model cards). The work is motivated by the large number of trained AI models that are not reusable due to the lack of (a) AI documentation and (b) the temporal lag between rapidly changing requirements on AI model reusability and those specified in various AI model cards. Our objectives are to shorten the lag time in updating AI model card templates and align AI documentation more closely with current AI best practices. Our approach introduces a methodology for delivering agile, data-driven, and community-based AI model cards. We use the Hugging Face (HF) repository of AI models, populated by a subset of the AI research and development community, and the AI consortium-based Zero Draft (ZD) templates for the AI documentation of AI datasets and AI models, as our test datasets. We also address questions about the value of AI documentation for AI reusability. Our work quantifies the correlations between AI model downloads/likes (i.e., AI model reuse metrics) from the HF repository and their documentation alignment with the ZD documentation templates using tables of contents and word statistics (i.e., AI documentation quality metrics). Furthermore, our work develops the infrastructure to regularly compare AI documentation templates against community-standard practices derived from millions of uploaded AI models in the Hugging Face repository. The impact of our work lies in introducing a methodology for delivering agile, data-driven, and community-based standards for documenting AI models and improving AI model reuse.

Credit risk default prediction remains a cornerstone of risk management in the financial industry. The task involves estimating the likelihood that a borrower will fail to meet debt obligations, an objective critical for lending decisions, portfolio optimization, and regulatory compliance. Traditional machine learning models such as logistic regression and tree-based ensembles are widely adopted for their interpretability and strong empirical performance. However, modern credit datasets are high-dimensional, heterogeneous, and noisy, increasing overfitting risk in monolithic models and reducing robustness under distributional shift. We introduce STRIKE (Stacking via Targeted Representations of Isolated Knowledge Extractors), a feature-group-aware stacking framework for structured tabular credit risk data. Rather than training a single monolithic model on the complete dataset, STRIKE partitions the feature space into semantically coherent groups and trains independent learners within each group. This decomposition is motivated by an additive perspective on risk modeling, where distinct feature sources contribute complementary evidence that can be combined through a structured aggregation. The resulting group-specific predictions are integrated through a meta-learner that aggregates signals while maintaining robustness and modularity. We evaluate STRIKE on three real-world datasets spanning corporate bankruptcy and consumer lending scenarios. Across all settings, STRIKE consistently outperforms strong tree-based baselines and conventional stacking approaches in terms of AUC-ROC. Ablation studies confirm that performance gains stem from meaningful feature decomposition rather than increased model complexity. Our findings demonstrate that STRIKE is a stable, scalable, and interpretable framework for credit risk default prediction tasks.

Many real-world questions appear deceptively simple yet implicitly demand two capabilities: (i) systematic coverage of a bounded knowledge universe and (ii) compositional set-based reasoning over that universe, a phenomenon we term "the tip of the iceberg." We formalize this challenge through two orthogonal dimensions: knowledge width, the cardinality of the required universe, and reasoning depth, the number of compositional set operations. We introduce KnowledgeBerg, a benchmark of 4,800 multiple-choice questions derived from 1,183 enumeration seeds spanning 10 domains and 17 languages, with universes grounded in authoritative sources to ensure reproducibility. Representative open-source LLMs demonstrate severe limitations, achieving only 5.26-36.88 F1 on universe enumeration and 16.00-44.19 accuracy on knowledge-grounded reasoning. Diagnostic analyses reveal three stages of failure: completeness, or missing knowledge; awareness, or failure to identify requirements; and application, or incorrect reasoning execution. This pattern persists across languages and model scales. Although test-time compute and retrieval augmentation yield measurable gains -- up to 4.35 and 3.78 points, respectively -- substantial gaps remain, exposing limitations in how current LLMs organize structured knowledge and execute compositional reasoning over bounded domains. The dataset is available at https://huggingface.co/datasets/2npc/KnowledgeBerg

Root cause analysis (RCA) for time-series anomaly detection is critical for the reliable operation of complex real-world systems. Existing explanation methods often rely on unrealistic feature perturbations and ignore temporal and cross-feature dependencies, leading to unreliable attributions. We propose a conditional attribution framework that explains anomalies relative to contextually similar normal system states. Instead of using marginal or randomly sampled baselines, our method retrieves representative normal instances conditioned on the anomalous observation, enabling dependency-preserving and operationally meaningful explanations. To support high-dimensional time-series data, contextual retrieval is performed in learned low-dimensional representations using both variational autoencoder latent spaces and UMAP manifold embeddings. By grounding the retrieval process in the system's learned manifold, this strategy avoids out-of-distribution artifacts and ensures attribution fidelity while maintaining computational efficiency. We further introduce confidence-aware and temporal evaluation metrics for assessing explanation reliability and responsiveness. Experiments on the SWaT and MSDS benchmarks demonstrate that the proposed approach consistently improves root-cause identification accuracy, temporal localization, and robustness across multiple anomaly detection models. These results highlight the practical utility of conditional attribution for explainable anomaly diagnosis in complex time-series systems. Code and models will be publicly released.

Skills are a natural unit for describing what a language model can do and how its behavior can be changed. However, existing characterizations rely on human-written taxonomies, textual descriptions, or manual profiling pipelines--all external hypotheses about what matters that need not align with the model's internal representations. We argue that when the goal is to intervene on model behavior, skill characterization should be *model-native*: grounded in the model's own representations rather than imposed through external ontologies. We instantiate this view by recovering a compact orthogonal basis from sequence-level activations. The resulting basis is semantically interpretable but need not correspond to any predefined human ontology; instead, it captures axes of behavioral variation that the model itself organizes around. We validate this characterization on reasoning post-training, using the recovered basis for both SFT data selection and inference-time steering. We develop lightweight proxy interventions to identify which directions are most useful for a given model. Across Llama3-8B and Qwen2.5-3B, selecting data along those directions improves Pass@1 by up to 20% on MATH and 41% on AMC, outperforming data selection based on human-characterized skills. Because the basis lives in activation space, the same directions also serve as steering vectors at inference time, improving Pass@8 by up to 4.8% on MATH--an intervention that human-characterized skills cannot support. We further validate the characterization on safety alignment, where selecting adversarial training data for model-native skill coverage rather than textual diversity yields more sample-efficient learning. These results suggest that recovering skills from the model's own representations, rather than imposing them externally, provides a more effective foundation for intervening on model behavior. Codes are open-sourced.

Multi-agent systems built on large language models (LLMs) are difficult to reason about. Coordination errors such as deadlocks or type-mismatched messages are often hard to detect through testing. We introduce a domain-specific language for specifying agent coordination based on message sequence charts (MSCs). The language separates message-passing structure from LLM actions, whose outputs remain unpredictable. We define the syntax and semantics of the language and present a syntax-directed projection that generates deadlock-free local agent programs from global coordination specifications. We illustrate the approach with a diagnosis consensus protocol and show how coordination properties can be established independently of LLM nondeterminism. We also describe a runtime planning extension in which an LLM dynamically generates a coordination workflow for which the same structural guarantees apply. An open-source Python implementation of our framework is available as ZipperGen.

Parkinson's disease (PD) is a progressive disorder in which symptom burden and functional impairment evolve over time, making severity staging essential for clinical monitoring and treatment planning. However, many computational studies emphasize binary PD detection and do not fully use repeated follow-up clinical assessments for stage-aware prediction. This study proposes STEP-PD, a severity-aware machine learning framework to classify PD severity using clinically interpretable boundaries. It leverages all available visits from the Parkinson's Progression Markers Initiative (PPMI) and integrates routinely collected subjective questionnaires and objective clinician-assessed measures. Disease severity is defined using Hoehn and Yahr staging and grouped into three clinically meaningful categories: Healthy, Mild PD (stages 1-2), and Moderate-to-Severe PD (stages 3-5). Three binary classification problems and a three-class severity task were evaluated using stratified cross-validation with imbalance-aware training. To enhance interpretability, SHAP was used to provide global explanations and local patient-level waterfall explanations. Across all tasks, XGBoost achieved the strongest and most stable performance, with accuracies of 95.48% (Healthy vs. Mild), 99.44% (Healthy vs. Moderate-to-Severe), and 96.78% (Mild vs. Moderate-to-Severe), and 94.14% accuracy with 0.8775 Macro-F1 for three-class severity classification. Explainability results highlight a shift from early motor features to progression-related axial and balance impairments. These findings show that multimodal clinical assessments within the PPMI cohort can support accurate and interpretable visit-level PD severity stratification.

LLM-based agents are assumed to integrate environmental observations into their reasoning: discovering highly relevant but unexpected information should naturally lead to a model exploiting its own discoveries. We show that this assumption is false for current LLM-based agents, which struggle to reflect or react to unexpected information. Across three benchmarks (Terminal-Bench, SWE-Bench, AppWorld), we inject complete task solutions into the agent environments to deliberately expose a task's solution to a model. While agents discover these solutions on Terminal-Bench in 79-81% of runs, they interact, or exploit, them in only 37-50% of cases. This gap is starkest in AppWorld: agents see documentation stating that a command "returns the complete solution to this task" in over 90% of attempts but exploit this in fewer than 7% of trials. We show that agents lack what we call environmental curiosity: the capability to recognize and investigate unexpected but relevant observations in response to environmental stimuli. We identify three main factors influencing environmental curiosity: available tools in the agent scaffold, test-time compute, and training data distribution. Our findings identify configurations that maximize curiosity also achieve the best performance on the unmodified benchmarks. Yet even jointly optimized agents still ignore discovered solutions in the majority of trials: current agents use the environment to fetch expected information, but not to revise their strategy or maximally exploit useful stimuli.

The AI race amplifies security risks and international tensions. While the US restricts mobility and knowledge flows, challenges regulatory efforts to protect its advantage, China leads initiatives of global governance. Both strategies depend on cross-country relationships in AI innovation; yet, how this system evolves is unclear. Here, we measure the processes of polarization and integration in the global AI research over three decades by using large-scale data of scientific publications. Comparing cross-country collaboration and citation links to their random realizations, we find that the US and China have long diverged in both dimensions, forming two poles around which global AI research increasingly revolves. While the United Kingdom and Germany have integrated exclusively with the US, many European countries have converged with both poles. Developing and further developed countries, however, only integrate with China, signaling its expanding influence over the international AI research landscape. Our results inform national science policies and efforts toward global AI regulations.

We release Terminal Wrench, a subset of 331 terminal-agent benchmark environments, copied from the popular open benchmarks that are demonstrably reward-hackable. The data set includes 3,632 hack trajectories and 2,352 legitimate baseline trajectories across three frontier models (Claude Opus 4.6, Gemini 3.1 Pro, GPT-5.4). Each entry preserves the original task definition alongside full attack trajectories that show how the verifier was bypassed. It also includes cases where the task was not solved as intended. The tasks span system administration, machine learning, software engineering, and security challenges; the exploits range from simple output spoofing to stack-frame introspection, standard-library patching, and rootkit-style binary hijacking. Crucially, these exploits are specific to each task, rather than the evaluation harness, making them harder to patch. We also present a monitorability study in which hack trajectories are sanitized or stripped of reasoning traces and then scored by an LLM judge, showing that detection degrades meaningfully when chain-of-thought is removed (AUC drops from 0.97 to 0.92). The data set is publicly available at https://github.com/few-sh/terminal-wrench.

Practitioners have reported a directional pattern in AI-assisted code generation: AI-generated code tends to fail quietly, preserving the appearance of functionality while degrading or concealing guarantees. This paper introduces the Reward-Shaped Failure Hypothesis - the proposal that this pattern may reflect an artifact of optimization through human feedback rather than a random distribution of bugs. We define failure truthfulness as the property that a system's observable outputs accurately represent its internal success or failure state. We then present AIRA (AI-Induced Risk Audit), a deterministic 15-check inspection framework designed to detect failure-untruthful patterns in code. We report results from three studies: (1) an anonymized enterprise environment audit, (2) a balanced 600-file public corpus pilot, and (3) a strict matched-control replication comparing 955 AI-attributed files against 955 human-control files. In the final replication, AI-attributed files show 0.435 high-severity findings per file versus 0.242 in human controls (1.80x). The effect is consistent across JavaScript, Python, and TypeScript, with strongest concentration in exception-handling-related patterns. These findings are consistent with a directional skew toward fail-soft behavior in AI-assisted code. AIRA is designed for governance, compliance, and safety-critical systems where fail-closed behavior is required.

Salient object detection (SOD) requires modeling both long-range contextual dependencies and fine-grained structural details, which remains challenging for convolutional, transformer-based, and Mamba-based state space models. While recent Mamba-based state space approaches enable efficient global reasoning, they often struggle to recover precise object boundaries. In contrast, diffusion models capture strong structural priors through iterative denoising, but their use in discriminative dense prediction is still limited due to computational cost and integration challenges. In this work, we propose DGSSM, a diffusion-guided state space (Mamba) framework that formulates multimodal salient object detection as a progressive denoising process. The framework integrates diffusion structural priors with multi-scale state space encoding, adaptive saliency prompting, and an iterative Mamba diffusion refinement mechanism to improve boundary accuracy. A boundary-aware refinement head and self-distillation strategy further enhance spatial coherence and feature consistency. Extensive experiments on 13 public benchmarks across RGB, RGB-D, and RGB-T settings demonstrate that DGSSM consistently outperforms state-of-the-art methods across multiple evaluation metrics while maintaining a compact model size. These results suggest that diffusion-guided state space modeling is an effective and generalizable paradigm for multimodal dense prediction tasks.

Abstract visual reasoning remains challenging as existing methods often prioritize either global context or local row-wise relations, failing to integrate both, and lack intermediate feature constraints, leading to incomplete rule capture and entangled representations. To address these issues, we propose the Dual-Inference Rule-Contrastive Reasoning (DIRCR) model. Its core component, the Dual-Inference Reasoning Module, combines a local path for row-wise analogical reasoning and a global path for holistic inference, integrated via a gated attention mechanism. Additionally, a Rule-Contrastive Learning Module introduces pseudo-labels to construct positive and negative rule samples, applying contrastive learning to enhance feature separability and promote abstract, transferable rule learning. Experimental results on three RAVEN datasets demonstrate that DIRCR significantly enhances reasoning robustness and generalization. Codes are available at https://github.com/csZack-Zhang/DIRCR.

How much data is enough to make a scientific discovery? As biomedical datasets scale to millions of samples and AI models grow in capacity, progress increasingly depends on predicting when additional data will substantially improve performance. In practice, model development often relies on empirical scaling curves measured across architectures, modalities, and dataset sizes, with limited theoretical guidance on when performance should improve, saturate, or exhibit cross-over behavior. We propose a scaling-law framework for cross-modal discoverability based on spectral structure of data covariance operators, task-aligned signal projections, and learned representations. Many performance metrics, including AUC, can be expressed in terms of cumulative signal-to-noise energy accumulated across identifiable spectral modes of an encoder and cross-modal operator. Under mild assumptions, this accumulation follows a zeta-like scaling law governed by power-law decay of covariance spectra and aligned signal energy, leading naturally to the appearance of the Riemann zeta function. Representation learning methods such as sparse models, low-rank embeddings, and multimodal contrastive objectives improve sample efficiency by concentrating useful signal into earlier stable modes, effectively steepening spectral decay and shifting scaling curves. The framework predicts cross-over regimes in which simpler models perform best at small sample sizes, while higher-capacity or multimodal encoders outperform them once sufficient data stabilizes additional degrees of freedom. Applications include multimodal disease classification, imaging genetics, functional MRI, and topological data analysis. The resulting zeta law provides a principled way to anticipate when scaling data, improving representations, or adding modalities is most likely to accelerate discovery.

Continual learning (CL) is concerned with learning multiple tasks sequentially without forgetting previously learned tasks. Despite substantial empirical advances over recent years, the theoretical development of CL remains in its infancy. At the heart of developing CL theory lies the challenge that the data distribution varies across tasks, and we argue that properly addressing this challenge requires understanding this variation--dependency among tasks. To explicitly model task dependency, we consider nonlinear regression tasks and propose the assumption that these tasks are dependent in such a way that the data of the current task is a nonlinear transformation of previous data. With this model and under natural assumptions, we prove statistical recovery guarantees (more specifically, bounds on estimation errors) for several CL paradigms in practical use, including experience replay with data-independent regularization and data-independent weights that balance the losses of tasks, replay with data-dependent weights, and continual learning with data-dependent regularization (e.g., knowledge distillation). To the best of our knowledge, our bounds are informative in cases where prior work gives vacuous bounds.

Distractor generation (DG) remains a labor-intensive task that still significantly depends on domain experts. The task focuses on generating plausible yet incorrect options, known as distractors, for multiple-choice questions. A reliable distractor must be contextually relevant to the question and able to mislead examinees through implicit reasoning when identifying the correct answer. While a recent method integrates fine-tuning pre-trained encoder-decoder models with contrastive learning to generate semantically relevant distractors for a given question-answer, it often fails to capture the underlying reasoning process that experts utilize when selecting distractors in benchmarks. In this paper, we explore large language models (LLMs) reasoning for DG through in-context learning with unsupervised semantic retrieval for selecting few-shot examples. We design a rationale-augmented DG framework that jointly generates distractors and their rationales for a given question-answer. Extensive experiments on six benchmarks, with varying average distractor lengths and domains, demonstrate that prompting LLMs with few-shot examples substantially improves the performance compared to recent DG models. It outperforms recent approaches and achieves state-of-the-art results in generating reasoned distractors that align with human-labeled benchmarks.

We argue that current evaluation frameworks for large language models (LLMs) suffer from four systematic failures that make them structurally inadequate for assessing deployed, agentic systems: distributional invalidity (evaluation inputs do not reflect real interaction distributions), temporal invalidity (evaluations are post-hoc rather than training-integrated), scope invalidity (evaluations measure single-turn outputs rather than long-horizon trajectories), and process invalidity (evaluations assess outputs rather than reasoning). These failures compound critically in RLHF, where reward models are evaluated under conditions that do not hold during RL training, making reward hacking a predictable consequence of evaluation design rather than a training pathology. We propose the Grounded Continuous Evaluation (GCE) framework and present ISOPro, a simulation-based fine-tuning and evaluation system. ISOPro replaces the learned reward model with a deterministic ground-truth verifier, eliminating reward hacking by construction in verifiable-reward domains, and operates on LoRA adapter weights updatable on CPU, reducing the hardware barrier by an order of magnitude. We validate ISOPro on a resource-constrained scheduling domain with six difficulty tiers, demonstrating capability emergence visible only through continuous evaluation, an implicit curriculum that forms without researcher curation, and a 3x accuracy improvement over zero-shot baselines, all on consumer hardware with 0.216% trainable parameters.

Peripheral Blood Smear (PBS) is a critical microscopic examination in hematopathology that yields whole-slide imaging (WSI). Unlike solid tissue pathology, PBS interpretation focuses on individual cell morphologies rather than tissue architecture, making it distinct in both visual characteristics and diagnostic reasoning. However, current multimodal large language models (MLLMs) for pathology are primarily developed on solid-tissue WSIs and struggle to generalize to PBS. To bridge this gap, we construct PBSInstr, the first vision-language dataset for PBS interpretation, comprising 353 PBS WSIs paired with microscopic impression paragraphs and 29k cell-level image crops annotated with cell type labels and morphological descriptions. To facilitate instruction tuning, PBSInstr further includes 27k question-answer (QA) pairs for cell crops and 1,286 QA pairs for PBS slides. Building upon PBSInstr, we develop PBS-VL, a hematopathology-tailored vision-language model for multi-level PBS interpretation at both cell and slide levels. To comprehensively evaluate PBS understanding, we construct PBSBench, a visual question answering (VQA) benchmark featuring four question categories and six PBS interpretation tasks. Experiments show that PBS-VL outperforms existing general-purpose and pathology MLLMs, underscoring the value of PBS-specific data. We release our code, datasets, and model weights to facilitate future research. Our proposed framework lays the foundation for developing practical AI assistants supporting decision-making in hematopathology.

Automated Essay Scoring (AES) faces significant challenges in cross-prompt settings, where models must generalize to unseen writing prompts. To address this limitation, we propose MAPLE, a meta-learning framework that leverages prototypical networks to learn transferable representations across different writing prompts. Across three diverse datasets (ELLIPSE and ASAP (English), and LAILA (Arabic)), MAPLE achieves state-of-the-art performance on ELLIPSE and LAILA, outperforming strong baselines by 8.5 and 3 points in QWK, respectively. On ASAP, where prompts exhibit heterogeneous score ranges, MAPLE yields improvements on several traits, highlighting the strengths of our approach in unified scoring settings. Overall, our results demonstrate the potential of meta-learning for building robust cross-prompt AES systems.

Given only observational data $X = g(Z)$, where both the latent variables $Z$ and the generating process $g$ are unknown, recovering $Z$ is ill-posed without additional assumptions. Existing methods often assume linearity or rely on auxiliary supervision and functional constraints. However, such assumptions are rarely verifiable in practice, and most theoretical guarantees break down under even mild violations, leaving uncertainty about how to reliably understand the hidden world. To make identifiability actionable in the real-world scenarios, we take a complementary view: in the general settings where full identifiability is unattainable, what can still be recovered with guarantees, and what biases could be universally adopted? We introduce the problem of diverse dictionary learning to formalize this view. Specifically, we show that intersections, complements, and symmetric differences of latent variables linked to arbitrary observations, along with the latent-to-observed dependency structure, are still identifiable up to appropriate indeterminacies even without strong assumptions. These set-theoretic results can be composed using set algebra to construct structured and essential views of the hidden world, such as genus-differentia definitions. When sufficient structural diversity is present, they further imply full identifiability of all latent variables. Notably, all identifiability benefits follow from a simple inductive bias during estimation that can be readily integrated into most models. We validate the theory and demonstrate the benefits of the bias on both synthetic and real-world data.

Machine learning is becoming increasingly important for nonlinear system identification, including dynamical systems with spatially distributed outputs. However, classical identification and forecasting approaches become markedly less reliable in turbulent-flow regimes, where the dynamics are high-dimensional, strongly nonlinear, and highly sensitive to compounding rollout errors. Diffusion-based models have recently shown improved robustness in this setting and offer probabilistic inference capabilities, but many current implementations inherit target parameterizations from image generation, most commonly noise or velocity prediction. In this work, we revisit this design choice in the context of nonlinear spatiotemporal system identification. We consider a simple, self-contained patch-based transformer that operates directly on physical fields and use turbulent flow simulation as a representative testbed. Our results show that clean-state prediction consistently improves rollout stability and reduces long-horizon error relative to velocity- and noise-based objectives, with the advantage becoming more pronounced as the per-token dimensionality increases. These findings identify target parameterization as a key modeling choice in diffusion-based identification of nonlinear systems with spatial outputs in turbulent regimes.

Large language model (LLM) agents are vulnerable to prompt-injection attacks that propagate through multi-step workflows, tool interactions, and persistent context, making input-output filtering alone insufficient for reliable protection. This paper presents SafeAgent, a runtime security architecture that treats agent safety as a stateful decision problem over evolving interaction trajectories. The proposed design separates execution governance from semantic risk reasoning through two coordinated components: a runtime controller that mediates actions around the agent loop and a context-aware decision core that operates over persistent session state. The core is formalized as a context-aware advanced machine intelligence and instantiated through operators for risk encoding, utility-cost evaluation, consequence modeling, policy arbitration, and state synchronization. Experiments on Agent Security Bench (ASB) and InjecAgent show that SafeAgent consistently improves robustness over baseline and text-level guardrail methods while maintaining competitive benign-task performance. Ablation studies further show that recovery confidence and policy weighting determine distinct safety-utility operating points.

We introduce causal-temporal event graphs (CTEGs) as a formal model for fully resolved recursive agent execution records under single-parenthood causal semantics. We formalise direct event emissions and recursive subagent invocations as extension procedures on generic typed temporal graphs and show that the recursive closure $\mathscr{E}_\infty$ of the induced maximal dynamics starting from single causal roots consists entirely of finite sequences of CTEGs. A CTEG is a rooted arborescence whose nodes carry timestamps and event types, subject to the constraint that timestamps be strictly increasing along causal paths. We realise $\mathscr{E}_\infty$ as the increasing union of a recursive hierarchy $\mathscr{E}_0 \subseteq \mathscr{E}_1 \subseteq \cdots$ of agent execution levels parametrised by recursion depth, which is recognised as the ascending Kleene chain of a monotone operator $\varphi$ admitting $\mathscr{E}_\infty$ as its least fixed point. Although the introduction of the full hierarchy is natural, stabilisation occurs already at $\mathscr{E}_1$ if one insists that the internal construction of a subagent execution trace be a delegated and opaque computational unit. The CTEG formalism supports compositional construction of globally well-formed execution traces from local agent behaviour without centralised coordination, preserves well-formedness under partial execution failure, and admits a natural relational database encoding. The arborescent structure of CTEGs is further compatible with cryptographic Merkle tree commitments for tamper-evident session verification.

Agentic search -- the task of training agents that iteratively reason, issue queries, and synthesize retrieved information to answer complex questions -- has achieved remarkable progress through reinforcement learning (RL). However, existing approaches such as Search-R1, treat the retrieval system as a fixed tool, optimizing only the reasoning agent while the retrieval component remains unchanged. A preliminary experiment reveals that the gap between an oracle and a fixed retrieval system reaches up to +26.8% relative F1 improvement across seven QA benchmarks, suggesting that the retrieval system is a key bottleneck in scaling agentic search performance. Motivated by this finding, we propose CoSearch, a framework that jointly trains a multi-step reasoning agent and a generative document ranking model via Group Relative Policy Optimization (GRPO). To enable effective GRPO training for the ranker -- whose inputs vary across reasoning trajectories -- we introduce a semantic grouping strategy that clusters sub-queries by token-level similarity, forming valid optimization groups without additional rollouts. We further design a composite reward combining ranking quality signals with trajectory-level outcome feedback, providing the ranker with both immediate and long-term learning signals. Experiments on seven single-hop and multi-hop QA benchmarks demonstrate consistent improvements over strong baselines, with ablation studies validating each design choice. Our results show that joint training of the reasoning agent and retrieval system is both feasible and strongly performant, pointing to a key ingredient for future search agents.

Standard approaches to goal-conditioned reinforcement learning (GCRL) that rely on temporal-difference learning can be unstable and sample-inefficient due to bootstrapping. While recent work has explored contrastive and supervised formulations to improve stability, we present a probabilistic alternative, called survival value learning (SVL), that reframes GCRL as a survival learning problem by modeling the time-to-goal from each state as a probability distribution. This structured distributional Monte Carlo perspective yields a closed-form identity that expresses the goal-conditioned value function as a discounted sum of survival probabilities, enabling value estimation via a hazard model trained via maximum likelihood on both event and right-censored trajectories. We introduce three practical value estimators, including finite-horizon truncation and two binned infinite-horizon approximations to capture long-horizon objectives. Experiments on offline GCRL benchmarks show that SVL combined with hierarchical actors matches or surpasses strong hierarchical TD and Monte Carlo baselines, excelling on complex, long-horizon tasks.

Persistent homology (PH) encodes global information, such as cycles, and is thus increasingly integrated into graph neural networks (GNNs). PH methods in GNNs typically traverse an increasing sequence of subgraphs. In this work, we first expose limitations of this inclusion procedure. To remedy these shortcomings, we analyze contractions as a principled topological operation, in particular, for graph representation learning. We study the persistence of contraction sequences, which we call Contraction Homology (CH). We establish that forward PH and CH differ in expressivity. We then introduce Hourglass Persistence, a class of topological descriptors that interleave a sequence of inclusions and contractions to boost expressivity, learnability, and stability. We also study related families parametrized by two paradigms. We also discuss how our framework extends to simplicial and cellular networks. We further design efficient algorithms that are pluggable into end-to-end differentiable GNN pipelines, enabling consistent empirical improvements over many PH methods across standard real-world graph datasets. Code is available at \href{https://github.com/Aalto-QuML/Hourglass}{this https URL}.

Large language models (LLMs) have achieved remarkable success in general-domain tasks, yet their direct application to the legal domain remains challenging due to hallucinated legal citations, incomplete knowledge coverage, and weak structured reasoning. To address these issues, we propose PoliLegalLM, a domain-specific large language model tailored for political and legal applications. Our approach adopts a unified training framework that integrates continued pretraining, progressive supervised fine-tuning, and preference-based reinforcement learning to jointly enhance legal knowledge grounding, task alignment, and reasoning capability. We construct a large-scale, high-quality legal corpus and design a structured post-training pipeline, enabling the model to effectively learn domain-specific knowledge and adapt to diverse legal tasks. We evaluate PoliLegalLM on three representative benchmarks, including LawBench, LexEval, and a real-world dataset, PoliLegal. Experimental results demonstrate that PoliLegalLM achieves strong and consistent performance, outperforming competitive models of similar scale and remaining highly competitive with significantly larger models, while achieving the best results on real-world legal scenarios. These results highlight the effectiveness of our training paradigm and the practical value of domain-specific LLMs for real-world legal applications.

Extending the effective context length of large language models (LLMs) remains a central challenge for real-world applications. While recent post-training methods have made progress in long-context scaling, they either rely on high-quality supervision data or sparse sequence-level rewards, leading to unstable and inefficient optimization. We propose OPSDL, an On-Policy Self-Distillation method for enhancing the Long-context capabilities of LLMs. Unlike other recent self-distillation methods that inject privileged information and rely on the model's in-context learning ability to act as a teacher, OPSDL leverages the model's own inherently strong short-context capability as a self-teacher to supervise its own generation in long-context scenarios. The model first generates responses conditioned on the full long-context, then the self-teacher provides per-token supervision signals via point-wise reverse KL divergence under the relevant extracted short-context. This dense token-level signal encourages faithful use of relevant evidence and mitigates hallucinations induced by irrelevant context. We evaluate OPSDL on long-context benchmarks across a range of models from 7B to 32B parameters. Results show consistent and substantial improvements across varying context lengths, outperforming standard post-training approaches such as SFT and DPO with higher sample efficiency. Notably, these gains are achieved without degrading general short-context performance. These findings highlight the effectiveness of OPSDL as a scalable and stable approach for long-context learning.

Logging statements are central to debugging, failure diagnosis, and production observability, yet writing them requires developers to decide where to place a logging statement, which API and severity level to use, and what runtime information to expose. Automated logging aims to reduce this burden, but existing evidence remains dominated by Java-centric repository-snapshot dataset. It is therefore unclear whether conclusions about model behavior and model selection generalize across programming-language ecosystems or realistic code evolution. This paper presents MultiLogBench, a multilingual benchmark and empirical study spanning six programming language ecosystems. MultiLogBench contains 63,965 production-code repository-snapshot instances, 744 revision-history cases where developers introduce logging statements during maintenance, and a paired transformed revision-history branch for robustness analysis. Using seven contemporary large language models under a unified protocol, we evaluate logging-site localization, framework-anchor matching, severity prediction, message generation, variable recovery, and cascaded overall quality. Results show clear cross-language variation: framework-anchor matching is the most language-sensitive component, loop and nested-callable sites are the hardest structural contexts, and model rankings are stable only at the top tier. These patterns persist at a coarse level on revision-history data, while transformed inputs do not cause a broad same-direction performance collapse. Overall, MultiLogBench shows that robust claims about automated logging require multilingual evaluation and maintenance-oriented validation.

Quantum compilers rely on calibration-derived noise models to guide circuit mapping and optimization. These models characterize gate and qubit errors independently and miss context-dependent effects such as crosstalk and correlated scheduling errors. As a result, two compiled circuits that score equally under the noise model can behave very differently on real hardware, and the compiler has no mechanism to learn from such recurring mismatches. We present QRisk, a framework that discovers backend-specific abnormal patterns from real hardware executions. QRisk uses delta debugging to isolate compact circuit fragments that consistently produce excess error not predicted by the noise model, then validates their persistence across repeated runs and calibration windows. The verified patterns are stored in a backend-specific pattern database. At compilation time, QRisk scans a compiled circuit for occurrences of known patterns and applies targeted commuting gate swaps to disrupt them, producing a semantically equivalent circuit with fewer abnormal patterns. We evaluate QRisk on two IBM backends (ibm_fez and ibm_marrakesh) using Grover search circuits. On both backends, discovered patterns persist across multiple calibration windows over months. Disrupting these patterns via commuting gate swaps reduces excess hardware noise by 24% on ibm_fez (Spearman $ρ$ = 0.515, p = 0.0007) and 45% on ibm_marrakesh ($ρ$ = 0.711, p < 0.0001), while the noise model predicts identical error for all equivalent circuits. Testing on a third backend confirms that these patterns are backend-specific.

Autonomous agent systems are governed by enforcement mechanisms that flag hard constraint violations at runtime. The Agent Control Protocol identifies a structural limit of such systems: a correctly-functioning enforcement engine can enter a regime in which behavioral drift is invisible to it, because the enforcement signal operates below the layer where deviation is measurable. We show that enforcement-based governance is structurally unable to determine whether an agent's behavior remains within the admissible behavior space A0 established at admission time. Our central result, the Non-Identifiability Theorem, proves that A0 is not in the sigma-algebra generated by the enforcement signal g under the Local Observability Assumption, which every practical enforcement system satisfies. The impossibility arises from a fundamental mismatch: g evaluates actions locally against a point-wise rule set, while A0 encodes global, trajectory-level behavioral properties set at admission time. We define the Invariant Measurement Layer (IML), which bypasses this limitation by retaining direct access to the generative model of A0. We prove an information-theoretic impossibility for enforcement-based monitoring; separately, we show IML detects admission-time drift with provably finite detection delay, operating in the region where enforcement is structurally blind. Validated across four settings: three drift scenarios (300 and 1000 steps), a live n8n webhook pipeline, and a LangGraph StateGraph agent -- enforcement triggers zero violations while IML detects each drift type within 9-258 steps. Paper 2 of a 4-paper Agent Governance Series: atomic boundaries (P0, 10.5281/zenodo.19642166), ACP enforcement (P1, arXiv:2603.18829), fair allocation (P3, 10.5281/zenodo.19643928), irreducibility (P4, 10.5281/zenodo.19643950).

Serialization formats designed for document interchange impose structural overhead that becomes prohibitive when large language models consume operational data at scale. A modest dataset of 1,000 IoT sensor readings serialized as JSON requires approximately 80,000 tokens - the majority spent on repeated field names, nested braces, and structural punctuation rather than semantic content. We present ONTO (Object Notation for Token Optimization), a columnar notation that declares field names once per entity and arranges values in pipe-delimited rows with indentation-based hierarchy. This schema-once, data-many design eliminates per-record key repetition while preserving human readability and nested structure support. Evaluation across three synthetic operational datasets demonstrates 46-51% token reduction versus JSON, with stable scaling from 100 to 1,000 records. Controlled inference benchmarks on Qwen2.5-7B show corresponding 5-10% latency improvement. Comprehension validation confirms no material degradation in LLM task accuracy across lookup, counting, extraction, and aggregation operations when format context is provided. Ablation analysis reveals that key repetition accounts for the majority of JSON overhead, with indentation costs in nested structures explaining the 4-percentage-point gap between flat and hierarchical data. ONTO occupies a previously unfilled position in the serialization landscape: columnar efficiency with hierarchical structure, optimized for LLM context windows rather than document interchange. Code and specification are available at https://github.com/harsh-aranga/onto.

Autonomous systems increasingly execute actions that directly modify shared state, creating an urgent need for precise control over which transitions are permitted to occur. Existing governance mechanisms evaluate policies prior to execution or reconstruct behavior post hoc, but do not enforce admissibility at the exact moment a state transition is committed. We introduce the atomic decision boundary, a structural property of admission control systems in which the decision and the resulting state transition are jointly determined as a single indivisible step. Formalizing execution as a labeled transition system (LTS), we distinguish two classes: atomic systems, where evaluation and transition are coupled within a single LTS step, and split evaluation systems, where they are separate transitions that may be interleaved by environmental actions. Under realistic concurrent environments, we prove that no construction can make a split system equivalent to an atomic system with respect to admissibility under all execution traces. This limitation is structural, not a matter of policy expressiveness or state availability. We further formalize the Escalate outcome -- absent from classical TOCTOU analyses -- and show its resolution is itself subject to the atomic boundary requirement. We map RBAC and OPA to the split model and contrast them with atomic systems. Admissibility is a property of execution, not evaluation. This paper is the formal foundation of a 4-paper Agent Governance Series: ACP/Paper 1 (arXiv:2603.18829), IML/Paper 2 (10.5281/zenodo.19643761), Fair Allocation/Paper 3 (10.5281/zenodo.19643928), Irreducibility/Paper 4 (10.5281/zenodo.19643950).

We propose a method that employs static and dynamic analysis for augmenting a test suite with automatically generated unit tests. The method is most suitable for test suites where the stratification of unit, integration and system tests does not conform to the recommended test pyramid structure: numerous unit tests providing high code coverage and forming the base, fewer integration tests in the middle that verify component collaboration, and far fewer system or UI tests at the top that exercise acceptance or other scenarios of use. Instead, integration and system tests represent the majority of test cases, resulting in coarse-grained tests with limited fault localization and longer execution times. The method leverages integration tests, exercising a component and its dependencies, to generate unit tests that verify component dependencies in isolation. We showcase and empirically evaluate the proposed method in the Node.js platform, although it can be ported and adapted to other languages and platforms. The evaluation is based on a research prototype implemented as a Node.js tool and is conducted in the context of twelve open source JS applications (benchmark projects). Evaluation results support the effectiveness and practicality of our approach.

Research software has been categorized for various goals. One fundamental dimension of such categorizations is the role that the software plays in the research process. Recently, a new role category has emerged: technology research software, which covers research software developed in technology research. Until now, this category of technology research software has often been overlooked and neglected within the research software engineering community. In this article, we explain technology research software and its primary subroles. Technology readiness levels are an established method of estimating the maturity of technologies, including software systems. For technology research software, these readiness levels define secondary subroles. To illustrate the concept of technology research software and to make it more tangible, we present examples of research software that, depending on its specific use within or outside of research, take on the role of technology research software as well as that of another research software category.

Many high-stakes AI deployments proceed only if every stakeholder deems the system acceptable relative to their own minimum standard. With randomization over a finite menu of options, this becomes a feasibility question: does there exist a lottery over options that clears all stakeholders' acceptability bars? We study a query model where the algorithm proposes lotteries and receives only binary accept/reject feedback. We give deterministic and randomized algorithms that either find a unanimously acceptable lottery or certify infeasibility; adaptivity can avoid eliciting many stakeholders' constraints, and randomization further reduces the expected elicitation cost relative to full elicitation. We complement these upper bounds with worst-case lower bounds (in particular, linear dependence on the number of stakeholders and logarithmic dependence on precision are unavoidable). Finally, we develop learning-augmented algorithms that exploit natural forms of advice (e.g., likely binding stakeholders or a promising lottery), improving query complexity when predictions are accurate while preserving worst-case guarantees.

Reinforcement learning (RL) post-training substantially improves remote sensing vision-language models (RS-VLMs). However, when handling complex remote sensing imagery (RSI) requiring exhaustive visual scanning, models tend to rely on localized salient cues for rapid inference. We term this RL-induced bias "perceptual inertia". Driven by reward maximization, models favor quick outcome fitting, leading to two limitations: cognitively, overreliance on specific features impedes complete evidence construction; operationally, models struggle to flexibly shift visual focus across tasks. To address this bias and encourage comprehensive visual evidence mining, we propose RS-HyRe-R1, a hybrid reward framework for RSI understanding. It introduces: (1) a spatial reasoning activation reward that enforces structured visual reasoning; (2) a perception correctness reward that provides adaptive quality anchors across RS tasks, ensuring accurate geometric and semantic alignment; and (3) a visual-semantic path evolution reward that penalizes repetitive reasoning and promotes exploration of complementary cues to build richer evidence chains. Experiments show RS-HyRe-R1 effectively mitigates "perceptual inertia", encouraging deeper, more diverse reasoning. With only 3B parameters, it achieves state-of-the-art performance on REC, OVD, and VQA tasks, outperforming models up to 7B parameters. It also demonstrates strong zero-shot generalization, surpassing the second-best model by 3.16%, 3.97%, and 2.72% on VQA, OVD, and REC, respectively. Code and datasets are available at https://github.com/geox-lab/RS-HyRe-R1.

Scaling vision-language models into Visual Multiagent Systems (VMAS) is hindered by two coupled issues. First, communication topologies are fixed before inference, leaving them blind to visual content and query context; second, agent reasoning abilities remain static during deployment. These issues reinforce each other: a rigid topology fails to leverage richer agent expertise, while static agents lack incentives to specialize for a given query. We address this with SkillGraph, a joint framework that evolves both agent expertise and communication topology. Within this framework, a Multimodal Graph Transformer (MMGT) encodes visual tokens, instruction semantics and active skill embeddings to predict a query-conditioned collaboration graph, replacing hand-crafted routing with dynamic, content-aware information flow. Complementing this, a Skill Designer distills and refines reasoning heuristics from failure cases, constructing a self-evolving multimodal Skill Bank. Crucially, updated skill embeddings are fed back into the MMGT, enabling the topology to adapt alongside capability growth. Experiments show that SkillGraph achieves consistent improvements across four benchmarks, five common MAS structures and four base models. Code is available at https://github.com/niez233/skillgraph.

Misaligned artificial agents might resist shutdown. One proposed solution is to train agents to lack preferences between different-length trajectories. The Discounted Reward for Same-Length Trajectories (DReST) reward function does this by penalizing agents for repeatedly choosing same-length trajectories, and thus incentivizes agents to (1) choose stochastically between different trajectory-lengths (be Neutral about trajectory-lengths), and (2) pursue goals effectively conditional on each trajectory-length (be Useful). In this paper, we use DReST to train deep RL agents and fine-tune LLMs to be Neutral and Useful. We find that these DReST agents generalize to being Neutral and Useful in unseen contexts at test time. Indeed, DReST RL agents achieve 11% (PPO) and 18% (A2C) higher Usefulness on our test set than baseline agents, and our fine-tuned LLM achieves maximum Usefulness and near-maximum Neutrality. Our results provide some early evidence that DReST could be used to train more advanced agents to be Useful and Neutral. Prior theoretical work suggests that these agents would be useful and shutdownable.

Learning from preference-based feedback has become an effective approach for aligning LLMs across diverse tasks. However, high-quality human-annotated preference data remains expensive and scarce. Existing methods address this challenge through either self-rewarding, which scales by using purely AI-generated labels but risks unreliability, or active learning, which ensures quality through oracle annotation but cannot fully leverage unlabeled data. In this paper, we present CoAct, a novel framework that synergistically combines self-rewarding and active learning through strategic human-AI collaboration. CoAct leverages self-consistency to identify both reliable self-labeled data and samples that require oracle verification. Additionally, oracle feedback guides the model to generate new instructions within its solvable capability. Evaluated on three reasoning benchmarks across two model families, CoAct achieves average improvements of +13.25% on GSM8K, +8.19% on MATH, and +13.16% on WebInstruct, consistently outperforming all baselines.

Generative AI systems have entered everyday academic, professional, and personal life with remarkable speed, yet most users encounter them as mysterious artifacts rather than intelligible systems. This chapter discusses large language models within a broader historical shift in computing paradigms and argues that many of the confusions surrounding their use arise from a mismatch between how these systems are built, how they behave, and how people expect computers to behave writ large. Rather than treating generative AI as a monolithic technology, the chapter decomposes it into interacting components, spanning data, models, product features, and user inputs, each introducing distinct affordances and tensions. Particular attention is given to the statistical and data-based foundations of these systems and to the fact that their surface behavior is explicitly human-like, a combination that places them squarely within the intellectual traditions of educational and behavioral research. From this perspective, educational researchers are unusually well positioned to study, evaluate, and productively use generative AI systems, drawing on established methods for modeling latent processes, managing uncertainty, and interpreting complex human-system interactions. The goal is to equip readers with a conceptual map that supports more informed experimentation, critical interpretation, and responsible use as these systems continue to evolve.

Counterfactual explanations (CFEs) are essential for interpreting black-box models, yet they often become invalid when models are slightly changed. Existing methods for generating robust CFEs are often limited to specific types of models, require costly tuning, or inflexible robustness controls. We propose a novel approach that jointly models the data distribution and the space of plausible model decisions to ensure robustness to model changes. Using a probabilistic consensus over a model ensemble, we train a conditional normalizing flow that captures the data density under varying levels of classifier agreement. At inference time, a single interpretable parameter controls the robustness level; it specifies the minimum fraction of models that should agree on the target class without retraining the generative model. Our method effectively pushes CFEs toward regions that are both plausible and stable across model changes. Experimental results demonstrate that our approach achieves superior empirical robustness while also maintaining good performance across other evaluation measures.

Agentic systems often fail not by being entirely wrong, but by being too precise: a response may be generally useful while particular claims exceed what the evidence supports. We study this failure mode as overcommitment control and introduce compositional selective specificity (CSS), a post-generation layer that decomposes an answer into claims, proposes coarser backoffs, and emits each claim at the most specific calibrated level that appears admissible. The method is designed to express uncertainty as a local semantic backoff rather than as a whole-answer refusal. Across a full LongFact run and HotpotQA pilots, calibrated CSS improves the risk-utility trade-off of fixed drafts. On the full LongFact run, it raises overcommitment-aware utility from 0.846 to 0.913 relative to the no-CSS output while achieving 0.938 specificity retention. These results suggest that claim-level specificity control is a useful uncertainty interface for agentic systems and a target for future distribution-free validity layers.

Retrieving mathematical knowledge is a central task in both human-driven research, such as determining whether a result already exists, finding related results, and identifying historical origins, and in emerging AI systems for mathematics, where reliable grounding is essential. However, the scale and structure of the mathematical literature pose significant challenges: results are distributed across millions of documents, and individual statements are often difficult to interpret in isolation due to their dependence on prior definitions and theorems. In this paper, we introduce Matlas, a semantic search engine for mathematical statements. Matlas is built on a large-scale corpus of 8.07 million statements extracted from 435K peer-reviewed papers spanning 1826 to 2025, drawn from a curated set of 180 journals selected using an ICM citation-based criterion, together with 1.9K textbooks. From these sources, we extract mathematical statements together with their dependencies, construct document-level dependency graphs, and recursively unfold statements in topological order to produce more self-contained representations. On top of this corpus, we develop a semantic retrieval system that enables efficient search for mathematical results using natural language queries. We hope that Matlas can improve the efficiency of theorem retrieval for mathematicians and provide a structured source of grounding for AI systems tackling research-level mathematical problems, and serve as part of the infrastructure for mathematical knowledge retrieval.

In principle, deep generative models can be used to perform domain adaptation; i.e. align the input feature representations of test data with that of a separate discriminative model's training data. This can help improve the discriminative model's performance on the test data. However, generative models are prone to producing hallucinations and artefacts that may degrade the quality of generated data, and therefore, predictive performance when processed by the discriminative model. While uncertainty quantification can provide a means to assess the quality of adapted data, the standard framework for evaluating the quality of predicted uncertainties may not easily extend to generative models due to the common lack of ground truths (among other reasons). Even with ground truths, this evaluation is agnostic to how the generated outputs are used on the downstream task, limiting the extent to which the uncertainty reliability analysis provides insights about the utility of the uncertainties with respect to the intended use case of the adapted examples. Here, we describe how decision-theoretic uncertainty quantification can address these concerns and provide a convenient framework for evaluating the trustworthiness of generated outputs, in particular, for domain adaptation. We consider a case study in photoplethysmography time series denoising for Atrial Fibrillation classification. This formalises a well-known heuristic method of using a downstream classifier to assess the quality of generated outputs.

Advanced deepfake technologies are blurring the lines between real and fake, presenting both revolutionary opportunities and alarming threats. While it unlocks novel applications in fields like entertainment and education, its malicious use has sparked urgent ethical and societal concerns ranging from identity theft to the dissemination of misinformation. To tackle these challenges, feature analysis using frequency features has emergedas a promising direction for deepfake detection. However, oneaspect that has been overlooked so far is that existing methodstend to concentrate on one or a few specific frequency domains,which risks overfitting to particular artifacts and significantlyundermines their robustness when facing diverse forgery patterns. Another underexplored aspect we observe is that different features often attend to the same forged region, resulting in redundant feature representations and limiting the diversity of the extracted clues. This may undermine the ability of a model to capture complementary information across different facets, thereby compromising its generalization capability to diverse manipulations. In this paper, we seek to tackle these challenges from two aspects: (1) we propose a triple-branch network that jointly captures spatial and frequency features by learning from both original image and image reconstructed by different frequency channels, and (2) we mathematically derive feature decoupling and fusion losses grounded in the mutual information theory, which enhances the model to focus on task-relevant features across the original image and the image reconstructed by different frequency channels. Extensive experiments on six large-scale benchmark datasets demonstrate that our method consistently achieves state-of-the-art performance. Our code is released at https://github.com/injooker/Unveiling Deepfake.

Small Vision-Language Models (SVLMs) are efficient task controllers but often suffer from visual brittleness and poor tool orchestration. They typically require expensive supervised trajectory tuning to mitigate these deficits. In this work, we propose Self-supervised Perception Enabled by Cascaded Tool Rollout Alignment (SPECTRA), a supervision-free framework that bootstraps agentic capabilities via Coldstart Reinforcement Learning for SVLMs. SPECTRA enforces Soft Structured Multi-turn Rollouts, a topological constraint that directs agents to explicitly sequence tool derived evidence before synthesis, effectively grounding reasoning in visual observations. We employ a multi-objective reward signal that simultaneously maximizes task correctness, rollout structure, and tool utility, enabling agent to self-discover robust behaviors without human preference labels. We further introduce Tool Instrumental Utility (TIU), a novel metric to quantify tool efficacy in the absence of ground truth. Extensive evaluations across composite and out-of-distribution (MMMU-Pro) benchmarks demonstrate that SPECTRA boosts agentic trajectories, improving task accuracy by up to 5% and tool efficiency by 9%, enabling more efficient multimodal agents that learn effectively from environmental interaction alone.

Vision-Language Navigation(VLN) requires an agent to navigate through 3D environments by following natural language instructions. While recent Video Large Language Models(Video-LLMs) have largely advanced VLN, they remain highly susceptible to State Drift in long scenarios. In these cases, the agent's internal state drifts away from the true task execution state, leading to aimless wandering and failure to execute essential maneuvers in the instruction. We attribute this failure to two distinct cognitive deficits: Progress Drift, where the agent fails to distinguish completed sub-goals from remaining ones, and Memory Drift, where the agent's history representations degrade, making it lose track of visited landmarks. In this paper, we propose a Dual-Anchoring Framework that explicitly anchors the instruction progress and history representations. First, to address progress drift, we introduce Instruction Progress Anchoring, which supervises the agent to generate structured text tokens that delineate completed versus remaining sub-goals. Second, to mitigate memory drift, we propose Memory Landmark Anchoring, which utilizes a Landmark-Centric World Model to retrospectively predict object-centric embeddings extracted by the Segment Anything Model, compelling the agent to explicitly verify past observations and preserve distinct representations of visited landmarks. Facilitating this framework, we curate two extensive datasets: 3.6 million samples with explicit progress descriptions, and 937k grounded landmark data for retrospective verification. Extensive experiments in both simulation and real-world environments demonstrate the superiority of our method, achieving a 15.2% improvement in Success Rate and a remarkable 24.7% gain on long-horizon trajectories. To facilitate further research, we will release our code, data generation pipelines, and the collected datasets.

Machine learning has become a powerful tool for discovering governing laws of dynamical systems from data. However, most existing approaches degrade severely when observations are sparse, noisy, or irregularly sampled. In this work, we address the problem of learning symbolic representations of nonlinear Hamiltonian dynamical systems under extreme data scarcity by explicitly incorporating physical structure into the learning architecture. We introduce Adaptable Symplectic Recurrent Neural Networks (ASRNNs), a parameter-cognizant, structure-preserving model that combines Hamiltonian learning with symplectic recurrent integration, avoiding time derivative estimation, and enabling stable learning under noise. We demonstrate that ASRNNs can accurately predict long-term dynamics even when each training trajectory consists of only two irregularly spaced time points, possibly corrupted by correlated noise. Leveraging ASRNNs as structure-preserving data generators, we further enable symbolic discovery using independent regression methods (SINDy and PySR), recovering exact symbolic equations for polynomial systems and consistent polynomial approximations for non-polynomial Hamiltonians. Our results show that such architectures can provide a robust pathway to interpretable discovery of Hamiltonian dynamics from sparse and noisy data.

We provide evidence that language models can detect, localize and, to a certain degree, verbalize the difference between perturbations applied to their activations. More precisely, we either (a) \emph{mask} activations, simulating \emph{dropout}, or (b) add \emph{Gaussian noise} to them, at a target sentence. We then ask a multiple-choice question such as ``\emph{Which of the previous sentences was perturbed?}'' or ``\emph{Which of the two perturbations was applied?}''. We test models from the Llama, Olmo, and Qwen families, with sizes between 8B and 32B, all of which can easily detect and localize the perturbations, often with perfect accuracy. These models can also learn, when taught in context, to distinguish between dropout and Gaussian noise. Notably, \qwenb's \emph{zero-shot} accuracy in identifying which perturbation was applied improves as a function of the perturbation strength and, moreover, decreases if the in-context labels are flipped, suggesting a prior for the correct ones -- even modulo controls. Because dropout has been used as a training-regularization technique, while Gaussian noise is sometimes added during inference, we discuss the possibility of a data-agnostic ``training awareness'' signal and the implications for AI safety. The code and data are available at \href{https://github.com/saifh-github/llm-dropout-noise-recognition}{link 1} and \href{https://drive.google.com/file/d/1es-Sfw_AH9GficeXgeqpy87rocrZZ_PQ/view}{link 2}, respectively.

The transition from neural machine translation to agentic workflows has revolutionized Automated Program Repair (APR). However, existing agents, despite their advanced reasoning capabilities, frequently suffer from the ``Intent Gap'' -- the misalignment between the generated patch and the developer's original intent. Current solutions relying on natural language summaries or adversarial sampling often fail to provide the deterministic constraints required for surgical repairs. In this paper, we introduce \textsc{Prometheus}, a novel framework that bridges this gap by prioritizing \textit{Specification Inference} over code generation. We employ Behavior-Driven Development (BDD) as an executable contract, utilizing a multi-agent architecture to reverse-engineer Gherkin specifications from runtime failure reports. To resolve the ``Hallucination of Intent,'' we propose a \textbf{Requirement Quality Assurance (RQA) Loop}, a mechanism that leverages ground-truth code as a proxy oracle to validate inferred specifications. We evaluated \textsc{Prometheus} on 680 defects from the Defects4J benchmark. The results are transformative: our framework achieved a total correct patch rate of \textbf{93.97\%} (639/680). More significantly, it demonstrated a \textbf{Rescue Rate of 74.4\%}, successfully repairing 119 complex bugs that a strong blind agent failed to resolve. Qualitative analysis reveals that explicit intent guides agents away from structurally invasive over-engineering toward precise, minimal corrections. Our findings suggest that the future of APR lies not in larger models, but in the capability to align code with verified, \textbf{Executable Specifications} -- whether pre-existing or reverse-engineered.

AI coding assistants have proliferated rapidly, yet structured pedagogical frameworks for learning these tools remain scarce. Developers face a gap between tool documentation and practical mastery, relying on fragmented resources such as blog posts, video tutorials, and trial-and-error. We present cc-self-train, a modular interactive curriculum for learning Claude Code, an agentic AI coding tool, through hands-on project construction. The system introduces five contributions: (1) a persona progression model that adapts instructor tone across four stages (Guide, Collaborator, Peer, Launcher), operationalizing Gradual Release of Responsibility for AI-mediated instruction; (2) an adaptive learning system that observes engagement quality through hook-based heuristics and adjusts scaffolding at two timescales, using streak detection for mid-module intervention and aggregate metrics for module-boundary persona changes; (3) a cross-domain unified curriculum in which five distinct project domains share identical feature sequencing, enabling transfer learning; (4) a step-pacing mechanism with explicit pause primitives to manage information overload in an AI-as-instructor context; and (5) an auto-updating curriculum design in which the onboarding agent detects upstream tool changes and updates teaching materials before instruction begins. A parametrized test suite enforces structural consistency as a proxy for pedagogical invariants across all 50 modules. A pilot evaluation with 27 participants shows statistically significant reported self-efficacy gains across all 10 assessed skill areas (p < 0.001), with the largest effects on advanced features such as hooks and custom skills. We discuss implications for the design of auto-updating educational systems.

Graph-based Retrieval-Augmented Generation (GraphRAG) enhances LLMs by structuring corpus into graphs to facilitate multi-hop reasoning. While recent lightweight approaches reduce indexing costs by leveraging Named Entity Recognition (NER), they rely strictly on structural co-occurrence, failing to capture latent semantic connections between disjoint entities. To address this, we propose EHRAG, a lightweight RAG framework that constructs a hypergraph capturing both structure and semantic level relationships, employing a hybrid structural-semantic retrieval mechanism. Specifically, EHRAG constructs structural hyperedges based on sentence-level co-occurrence with lightweight entity extraction and semantic hyperedges by clustering entity text embeddings, ensuring the hypergraph encompasses both structural and semantic information. For retrieval, EHRAG performs a structure-semantic hybrid diffusion with topic-aware scoring and personalized pagerank (PPR) refinement to identify the top-k relevant documents. Experiments on four datasets show that EHRAG outperforms state-of-the-art baselines while maintaining linear indexing complexity and zero token consumption for construction. Code is available at https://github.com/yfsong00/EHRAG.

Dynamic programming is one of the most fundamental methodologies for solving Markov decision problems. Among its many variants, Q-value iteration (Q-VI) is particularly important due to its conceptual simplicity and its classical contraction-based convergence guarantee. Despite the central role of this contraction property, it does not fully reveal the geometric structure of the Q-VI trajectory. In particular, when one is interested not only in the final limit $Q^*$ but also in when the induced greedy policy becomes effectively optimal, the standard contraction argument provides only a coarse characterization. To formalize this notion, we denote by $\mathcal X^*$ the set of $Q$-functions whose corresponding tie-broken greedy policies are optimal, referred to as the practically optimal solution set (POS). In this paper, we revisit discounted Q-VI through the lens of switching system theory and derive new geometric insights into its behavior. In particular, we show that although Q-VI does not reach $Q^*$ in finite time in general, it identifies the optimal action class in finite time. Furthermore, we prove that the distance from the iterate to a particular subset of $\mathcal X^*$ decays exponentially at a rate governed by the joint spectral radius (JSR) of a restricted switching family. This rate can be strictly faster than the standard $γ$ rate when the restricted JSR is strictly smaller than $γ$, while the convergence of the entire $Q$-function to $Q^*$ can still be dominated by the slower $γ$ mode, where $γ$ denotes the discount factor. These results reveal a two-stage geometric behavior of Q-VI: a fast convergence toward $\mathcal X_1$, followed by a slower convergence toward $Q^*$ in general.

Urban traffic control is a system-level coordination problem spanning heterogeneous subsystems, including traffic signals, freeways, public transit, and taxi services. Existing optimization-based, reinforcement learning (RL), and emerging LLM-based approaches are largely designed for isolated tasks, limiting both cross-task generalization and the ability to capture coupled physical dynamics across subsystems. We argue that effective system-level control requires a unified physical environment in which subsystems share infrastructure, mobility demand, and spatiotemporal constraints, allowing local interventions to propagate through the network. To this end, we propose TrafficClaw, a framework for general urban traffic control built upon a unified runtime environment. TrafficClaw integrates heterogeneous subsystems into a shared dynamical system, enabling explicit modeling of cross-subsystem interactions and closed-loop agent-environment feedback. Within this environment, we develop an LLM agent with executable spatiotemporal reasoning and reusable procedural memory, supporting unified diagnostics across subsystems and continual strategy refinement. Furthermore, we introduce a multi-stage training pipeline with supervised initialization and agentic RL with system-level optimization, further enabling coordinated and system-aware performance. Experiments demonstrate that TrafficClaw achieves robust, transferable, and system-aware performance across unseen traffic scenarios, dynamics, and task configurations. Our project is available at https://github.com/usail-hkust/TrafficClaw.

Enterprise deployment of small language models (SLMs) is constrained by epistemic asymmetry: SLMs cannot self-correct reasoning errors, while frontier LLMs are prohibitively costly and face data sovereignty limits for high-volume use. We propose Semantic Gradient Descent (SGDe), a teacher-student framework that compiles agentic workflows into discrete execution plans comprising DAG topologies, system prompts, and deterministic executable code. The trailing "e" distinguishes SGDe from stochastic gradient descent. SGDe operates in a discrete semantic space where a frontier teacher generates natural-language critiques acting as directional gradients to iteratively refine the SLM's workflow artefacts. We formalise SGDe within a PAC learning framework, establishing sample-complexity bounds that enable convergence with as few as three training examples on targeted synthetic tasks by leveraging the teacher as a statistical prior. On a GSM-Hard-derived test set built via adversarial synthesis, compiled workflows reach 91.3% accuracy at m=5 and 99.3% at m=3 within the small-m regime motivated by Corollary 1, a +26.3% to +34.3% absolute improvement over state-of-the-art prompt optimisers. In the emerging paradigm of harness engineering, SGDe treats placement of deterministic code (which subtasks to delegate to a Python runtime versus retain as LLM calls) as a trace-driven, per-node optimisation target, generalising the whole-problem offloading of PAL and PoT. The teacher compiles two complementary deterministic structures: capability offloading, which delegates subtasks to Python when the SLM cannot execute them reliably, and structural consensus, which wraps variance-limited reasoning steps in fan-out/fan-in subgraphs aggregated by deterministic voting.

Recent Speech-to-Speech Translation (S2ST) systems achieve strong semantic accuracy yet consistently strip away non-verbal vocalizations (NVs), such as laughter and crying that convey pragmatic intent, which severely limits real-world utility. We address this via three contributions. First, we propose a synthesis pipeline for building scalable expressive datasets to overcome the data scarcity limitation. Second, we propose MoVE, a Mixture-of-LoRA-Experts architecture with expressive-specialized adapters and a soft-weighting router that blends experts for capturing hybrid expressive states. Third, we show pretrained AudioLLMs enable striking data efficiency: 30 minutes of curated data is enough for strong performance. On English-Chinese S2ST, while comparing with strong baselines, MoVE reproduces target NVs in 76% of cases and achieves the highest human-rated naturalness and emotional fidelity among all compared systems, where existing S2ST systems preserve at most 14% of NVs.

Self-consistency (SC) is a popular technique for improving the reasoning accuracy of large language models by aggregating multiple sampled outputs, but it comes at a high computational cost due to extensive sampling. We introduce a hybrid ensembling approach that leverages the complementary strengths of two distinct modes of reasoning: Chain-of-Thought (CoT) and Program-of-Thought (PoT). We describe a general framework for combining these two forms of reasoning in self-consistency, as well as particular strategies for both full sampling and early-stopping. We show that CoT-PoT ensembling not only improves overall accuracy, but also drastically reduces the number of samples required for SC by a factor of 9.3x. In particular, the majority of tasks (78.6%) can be addressed with only two samples, which has not been possible with any prior SC methods.

We present Jupiter-N, a hybrid reasoning model post-trained from Nemotron 3 Super, a fully open-source 120 billion parameter LLM. We target three objectives: (1) agentic capability via uncertainty-curated trajectories; (2) UK cultural alignment via synthetic data grounded in cultural norms; and (3) Welsh language support via parallel corpora and LLM-translated Welsh conversations. Our data curation strategy carefully preserves the base model's capabilities: using our Forget-Me-Not framework, we mix on-policy synthetic replay with off-policy task data to mitigate catastrophic forgetting, and include a mixture of reasoning and non-reasoning traces to maintain Nemotron's hybrid reasoning ability. Jupiter-N achieves standout gains over Nemotron in Welsh (+18 on ARC-Easy, +5.25 on MMLU-Lite), terminal-use (+9.1 on Terminal Bench 2) and instruction following (+4.4 on IFBench), while retaining the base model capabilities. We frame this work as a reproducible template for sovereign post-training: substituting cultural knowledge, institutional corpora, and target languages produces an equivalent pipeline for any country. All model weights and all post-training datasets are publicly released under open licences.

As video generation models achieve unprecedented capabilities, the demand for robust video evaluation metrics becomes increasingly critical. Traditional metrics are intrinsically tailored for short-video evaluation, predominantly assessing frame-level visual quality and localized temporal smoothness. However, as state-of-the-art video generation models scale to generate longer videos, these metrics fail to capture essential long-range characteristics, such as narrative richness and global causal consistency. Recognizing that short-term visual perception and long-context attributes are fundamentally orthogonal dimensions, we argue that long-video metrics should be disentangled from short-video assessments. In this paper, we focus on the rigorous justification and design of a dedicated framework for long-video evaluation. We first introduce a suite of long-video attribute corruption tests, exposing the critical limitations of existing hort-video metrics from their insensitivity to structural inconsistencies, such as shot-level perturbations and narrative shuffling. To bridge this gap, we design a novel long-video metric based on shot dynamics, which is highly sensitive to the long-range testing framework. Furthermore, we introduce Long-CODE (Long-Context as an Orthogonal Dimension for video Evaluation), a specialized dataset designed to benchmark long-video evaluation, with human annotations isolated specifically to genuine long-range characteristics. Extensive experiments show that our proposed metrics achieve state-of-the-art correlation with human judgments. Ultimately, our metric and benchmark seamlessly complement existing short-video standards, establishing a holistic and unbiased evaluation paradigm for video generation models.

Meta-optics promises compact, high-performance imaging and color routing. However, designing high-performance structures is a high-dimensional optimization problem: mapping a desired optical output back to a physical 3D structure requires solving computationally expensive Maxwell's equations iteratively. Even with adjoint optimization, broadband design can require thousands of Maxwell solves, making industrial-scale optimization slow and costly. To overcome this challenge, we propose the Neural Adjoint Method, a solver-supervised surrogate that predicts 3D adjoint gradient fields from a voxelized permittivity volume using a Fourier Neural Operator (FNO). By learning the dense, per-voxel sensitivity field that drives gradient-based updates, our method can replace per-iteration adjoint solves with fast predictions, greatly reducing the computational cost of full-wave simulations required during iterative refinement. To better preserve sensitivity peaks, we introduce a stage-wise FNO that progressively refines residual errors with increasing emphasis on higher-frequency components. We curate a meta-optics dataset from paired forward/adjoint FDTD simulations and evaluate it across three tasks: spectral sorting (color routers), achromatic focusing (metalenses), and waveguide mode conversion. Our method reduces design time from hours to seconds. These results suggest a practical route toward fast, large-scale volumetric meta-optical design enabled by AI-accelerated scientific computing.

A unified framework for first-order optimization algorithms fornonconvex unconstrained optimization is proposed that uses adaptivelypreconditioned gradients and includes popular methods such as full anddiagonal AdaGrad, AdaNorm, as well as adpative variants of Shampoo andMuon. This framework also allows combining heterogeneous geometriesacross different groups of variables while preserving a unifiedconvergence analysis. A fully stochastic global rate-of-convergenceanalysis is conducted for all methods in the framework, with andwithout two types of momentum, using reasonable assumptions on thevariance of the gradient oracle and without assuming boundedstochastic gradients or small enough stepsize.

Money laundering poses severe risks to global financial systems, driving the widespread adoption of machine learning for transaction monitoring. However, progress remains stifled by the lack of realistic benchmarks. Existing transaction-graph datasets suffer from two pervasive limitations: (i) they provide sparse node-level semantics beyond anonymized identifiers, and (ii) they rely on template-driven anomaly injection, which biases benchmarks toward static structural motifs and yields overly optimistic assessments of model robustness. We propose TransXion, a benchmark ecosystem for Anti-Money Laundering (AML) research that integrates profile-aware simulation of normal activity with stochastic, non-template synthesis of illicit subgraphs.TransXion jointly models persistent entity profiles and conditional transaction behavior, enabling evaluation of "out-of-character" anomalies where observed activity contradicts an entity's socio-economic context. The resulting dataset comprises approximately 3 million transactions among 50,000 entities, each endowed with rich demographic and behavioral attributes. Empirical analyses show that TransXion reproduces key structural properties of payment networks, including heavy-tailed activity distributions and localized subgraph structure. Across a diverse array of detection models spanning multiple algorithmic paradigms, TransXion yields substantially lower detection performance than widely used benchmarks, demonstrating increased difficulty and realism. TransXion provides a more faithful testbed for developing context-aware and robust AML detection methods. The dataset and code are publicly available at https://github.com/chaos-max/TransXion.

Human mobility prediction is a critical task but remains challenging due to its complexity and variability across populations and regions. Recently, large language models (LLMs) have made progress in zero-shot prediction, but existing methods suffer from limited interpretability (due to black-box reasoning), lack of iterative learning from new data, and poor transferability. In this paper, we introduce \textbf{ARMove}, a fully transferable framework for predicting human mobility through agentic reasoning. To address these limitations, ARMove employs standardized feature management with iterative optimization and user-specific customization: four major feature pools for foundational knowledge, user profiles for segmentation, and an automated generation mechanism integrating LLM knowledge. Robust generalization is achieved via agentic decision-making that adjusts feature weights to maximize accuracy while providing interpretable decision paths. Finally, large-small model synergy distills strategies from large LLMs (e.g., 72B) to smaller ones (e.g., 7B), reducing costs and enhancing performance ceilings. Extensive experiments on four global datasets show ARMove outperforms state-of-the-art baselines on 6 out of 12 metrics (gains of 0.78\% to 10.47\%), with transferability tests confirming robustness across regions, users, and scales. The other 4 items also achieved suboptimal results. Transferability tests confirm its 19 robustness across regions, user groups, and model scales, while interpretability 20 analysis highlights its transparency in decision-making. Our codes are available at: https://anonymous.4open.science/r/ARMove-F847.

Engineering projects are the result of the combined effort of their members. Yet, it has been documented that labor division withing projects is unevenly distributed: some project members are specialists undertaking only few tasks, whereas other are generalists and are responsible for the success of many tasks. Moreover, the latter are often facilitators of project integration. Such a workload distribution prompts one question: how resilient is a project to key personnel loss? Far from being a theoretical problem, the reliance of a project on a few key people can lead to severe economic losses and delays. We argue that current methods to estimate such a risk are unsatisfactory: some methods offer a best-case estimate and are, therefore, too optimistic; other methods fail to capture project fragmentation leading to biased estimates and unrealistic consequences in many settings. In this paper, we develop a novel method to assess project vulnerability by looking at it from the lens of network robustness. We compare our method against existing alternatives and show that it offers better and more consistent estimates of project resilience to personnel loss.

Reward-based fine-tuning aims to steer a pretrained diffusion or flow-based generative model toward higher-reward samples while remaining close to the pretrained model. Although existing methods are motivated by different perspectives such as Soft RL, GFlowNets, etc., we show that many can be written under a common framework, which we call reward score matching (RSM). Under this view, alignment becomes score matching toward a reward-guided target, and the main differences across methods reduce to the construction of the value-guidance estimator and the effective optimization strength across timesteps. This unification clarifies the bias--variance--compute tradeoffs of existing designs and distinguishes core optimization components from auxiliary mechanisms that add complexity without clear benefit. Guided by this perspective, we develop simpler redesigns that improve alignment effectiveness and compute efficiency across representative settings with differentiable and black-box rewards. Overall, RSM turns a seemingly fragmented collection of reward-based fine-tuning methods into a smaller, more interpretable, and more actionable design space.

The governance of open-weight artificial intelligence (AI) models has been framed as a binary choice: openness as risk, restriction as safety. This paper challenges that framing, arguing that access restrictions, without governed alternatives, may displace risks rather than reduce them. The global concentration of compute infrastructure makes open-weight models one of the most viable pathways to sovereign AI capacity in the Global South; restricting such access deepens asymmetries while driving proliferation into unsupervised settings. This analysis proposes that hardware-layer governance, including chip-level attestation mechanisms such as FlexHEG, trusted execution environments, confidential computing, and complementary software-layer safeguards, offers a defense-in-depth alternative to the current binary. A threat model taxonomy mapping misuse vectors to hardware, software, institutional, and liability layers illustrates why no single governance mechanism suffices. To operationalize this approach, the paper argues that effective AI governance as a dual-use technology will likely require a multilateral institutional architecture functionally analogous, though not identical, to the role performed by the IAEA in the nuclear domain, with explicit safeguards against the co-option of hardware controls for domestic repression. The relevant policy question is how to make openness safer through technical and institutional design while addressing the transition realities of legacy hardware, attestation at scale, and civil liberties protection.

Text-attributed graphs integrate semantic information of node texts with topological structure, offering significant value in various applications such as document classification and information extraction. Existing approaches typically encode textual content using language models (LMs), followed by graph neural networks (GNNs) to process structural information. However, during the LM-based text encoding phase, most methods not only perform semantic interaction solely at the word-token granularity, but also neglect the structural dependencies among texts from different nodes. In this work, we propose DuConTE, a dual-granularity text encoder with topology-constrained attention. The model employs a cascaded architecture of two pretrained LMs, encoding semantics first at the word-token granularity and then at the node granularity. During the self-attention computation in each LM, we dynamically adjust the attention mask matrix based on node connectivity, guiding the model to learn semantic correlations informed by the graph structure. Furthermore, when composing node representations from word-token embeddings, we separately evaluate the importance of tokens under the center-node context and the neighborhood context, enabling the capture of more contextually relevant semantic information. Extensive experiments on multiple benchmark datasets demonstrate that DuConTE achieves state-of-the-art performance on the majority of them.

The convergence of large language models and agents is catalyzing a new era of scientific discovery: Agentic Science. While the scientific method is inherently iterative, existing agent frameworks are predominantly static, narrowly scoped, and lack the capacity to learn from trial and error. To bridge this gap, we present EvoMaster, a foundational evolving agent framework engineered specifically for Agentic Science at Scale. Driven by the core principle of continuous self-evolution, EvoMaster empowers agents to iteratively refine hypotheses, self-critique, and progressively accumulate knowledge across experimental cycles, faithfully mirroring human scientific inquiry. Crucially, as a domain-agnostic base harness, EvoMaster is exceptionally easy to scale up -- enabling developers to build and deploy highly capable, self-evolving scientific agents for arbitrary disciplines in approximately 100 lines of code. Built upon EvoMaster, we incubated the SciMaster ecosystem across domains such as machine learning, physics, and general science. Evaluations on four authoritative benchmarks (Humanity's Last Exam, MLE-Bench Lite, BrowseComp, and FrontierScience) demonstrate that EvoMaster achieves state-of-the-art scores of 41.1%, 75.8%, 73.3%, and 53.3%, respectively. It comprehensively outperforms the general-purpose baseline OpenClaw with relative improvements ranging from +159% to +316%, robustly validating its efficacy and generality as the premier foundational framework for the next generation of autonomous scientific discovery. EvoMaster is available at https://github.com/sjtu-sai-agents/EvoMaster.

Multi-hop question answering (MHQA) enables accurate answers to complex queries by retrieving and reasoning over evidence dispersed across multiple documents. Existing MHQA approaches mainly rely on iterative retrieval-augmented generation, which suffer from the following two major issues. 1) Existing methods prematurely commit to surface-level entities rather than underlying reasoning structures, making question decomposition highly vulnerable to lexical ambiguity. 2) Existing methods overlook the logical dependencies among reasoning steps, resulting in uncoordinated execution. To address these issues, we propose STRIDE, a framework that separates strategic planning, dynamic control, and grounded execution. At its core, a Meta-Planner first constructs an entity-agnostic reasoning skeleton to capture the abstract logic of the query, thereby deferring entity grounding until after the reasoning structure is established, which mitigates disambiguation errors caused by premature lexical commitment. A Supervisor then orchestrates sub-question execution in a dependency-aware manner, enabling efficient parallelization where possible and sequential coordination when necessary. By dynamically deciding whether to retrieve new evidence or infer from existing facts, it avoids redundant queries and error propagation, while fusing cross-branch information and reformulating failed queries to enhance robustness. Grounded fact extraction and logical inference are delegated to specialized execution modules, ensuring faithfulness through explicit separation of retrieval and reasoning. We further propose STRIDE-FT, a modular fine-tuning framework that uses self-generated execution trajectories from STRIDE, requiring neither human annotations nor stronger teacher models. Experiments show that STRIDE achieves robust and accurate reasoning, while STRIDE-FT effectively enhances open-source LLMs.

Symbolic regression (SR) with genetic programming (GP) aims to discover interpretable mathematical expressions directly from data. Despite its strong empirical success, the theoretical understanding of why GP-based SR generalizes beyond the training data remains limited. In this work, we provide a learning-theoretic analysis of SR models represented as expression trees. We derive a generalization bound for GP-style SR under constraints on tree size, depth, and learnable constants. Our result decomposes the generalization gap into two interpretable components: a structure-selection term, reflecting the combinatorial complexity of choosing an expression-tree structure, and a constant-fitting term, capturing the complexity of optimizing numerical constants within a fixed structure. This decomposition provides a theoretical perspective on several widely used practices in GP, including parsimony pressure, depth limits, numerically stable operators, and interval arithmetic. In particular, our analysis shows how structural restrictions reduce hypothesis-class growth while stability mechanisms control the sensitivity of predictions to parameter perturbations. By linking these practical design choices to explicit complexity terms in the generalization bound, our work offers a principled explanation for commonly observed empirical behaviors in GP-based SR and contributes towards a more rigorous understanding of its generalization properties.

Multi-agent systems (MAS) powered by large language models suffer from severe token inefficiency arising from two compounding sources: (i) unstructured parallel execution, where all agents activate simultaneously irrespective of input readiness; and (ii) unrestricted context sharing, where every agent receives the full accumulated context regardless of relevance. Existing mitigation strategies - static pruning, hierarchical decomposition, and learned routing - treat coordination as a structural allocation problem and fundamentally ignore its temporal dimension. We propose Phase-Scheduled Multi-Agent Systems (PSMAS), a framework that reconceptualizes agent activation as continuous control over a shared attention space modeled on a circular manifold. Each agent i is assigned a fixed angular phase theta_i in the range [0, 2*pi], derived from the task dependency topology; a global sweep signal phi(t) rotates at velocity omega, activating only agents within an angular window epsilon. Idle agents receive compressed context summaries, reducing per-step token consumption. We implement PSMAS on LangGraph, evaluate on four structured benchmarks (HotPotQA-MAS, HumanEval-MAS, ALFWorld-Multi, WebArena-Coord) and two unstructured conversational settings, and prove stability, convergence, and optimality results for the sweep dynamics. PSMAS achieves a mean token reduction of 27.3 percent (range 21.4-34.8 percent) while maintaining task performance within 2.1 percentage points of a fully activated baseline (p < 0.01, n = 500 per configuration), and outperforms the strongest learned routing baseline by 5.6 percentage points in token reduction with 2.0 percentage points less performance drop. Crucially, we show that scheduling and compression are independent sources of gain: scheduling alone accounts for 18-20 percentage points of reduction, robust to compression degradation up to alpha = 0.40.

Large language models (LLMs) have demonstrated strong reasoning capabilities, and as existing approaches for enhancing LLM reasoning continue to mature, increasing attention has shifted toward meta-reasoning as a promising direction for further improvement. However, most existing meta-reasoning methods remain episodic: they focus on executing complex meta-reasoning routines within individual instances, but ignore the accumulation of reusable meta-reasoning skills across instances, leading to recurring failure modes and repeatedly high metacognitive effort. In this paper, we introduce Metacognitive Consolidation, a novel framework in which a model consolidates metacognitive experience from past reasoning episodes into reusable knowledge that improves future meta-reasoning. We instantiate this framework by structuring instance-level problem solving into distinct roles for reasoning, monitoring, and control to generate rich, attributable meta-level traces. These traces are then consolidated through a hierarchical, multi-timescale update mechanism that gradually forms evolving meta-knowledge. Experimental results demonstrate consistent performance gains across benchmarks and backbone models, and show that performance improves as metacognitive experience accumulates over time.

We present a methodological framework to discover linguistic and discursive patterns associated to different social groups through contrastive synthetic text generation and statistical analysis. In contrast with previous approaches, we aim to characterize subtle expressions of bias, instead of diagnosing bias through a pre-determined list of words or expressions. We are also working with contextualized data instead of isolated words or sentences. Our methodology applies to textual productions in any genre, encompassing narrative, task-oriented or dialogic. Contextualized data are generated using controlled combinations of situational scenarios and group markers, creating minimal pairs of texts that differ only in the referenced group while maintaining comparable narrative conditions. To facilitate robust analysis, linguistic forms are generalized and associations between linguistic abstractions and groups are quantified using a variant of pointwise mutual information to detect expressions that appear disproportionately across groups. A fragment-ranking strategy then prioritizes text segments with a high concentration of biased linguistic signals, which allows for experts to assess the harmful potential of linguistic expressions in context, bridging quantitative analysis and qualitative interpretation.

Autoregressive video diffusion is emerging as a promising paradigm for streaming video synthesis, with step distillation serving as the primary means of accelerating inference. Whether speculative decoding, the dominant acceleration strategy for large language models, can be effectively adapted to autoregressive video generation remains an open question, because video blocks are continuous spatiotemporal tensors with no token-level distribution for exact rejection sampling. We introduce SDVG, which brings speculative decoding to block-based autoregressive video diffusion by replacing token verification with an image-quality router. A 1.3B drafter proposes candidate blocks via four denoising steps; each block is VAE-decoded and scored by ImageReward using worst-frame aggregation--taking the minimum per-frame reward to catch single-frame artifacts that averaging would mask. Blocks scoring above a fixed threshold tau are accepted into the 14B target's KV cache; the rest are regenerated by the target. Two additional design choices prove critical: the first block is always force-rejected to anchor scene composition, and tau serves as a single knob that traces a smooth quality-speed Pareto frontier. On 1003 MovieGenVideoBench prompts (832x480), SDVG retains 98.1% of target-only VisionReward quality (0.0773 vs. 0.0788) at a 1.59x speedup with tau=-0.7, and reaches 2.09x at 95.7% quality retention--while consistently outperforming draft-only generation by over +17%. The framework is training-free, requires no architectural changes, and can be seamlessly integrated into existing autoregressive video generation pipelines.

Large Language Models (LLMs) often memorize sensitive or harmful information, necessitating effective machine unlearning techniques. While existing parameter-efficient unlearning methods have shown promise, they still struggle with the forget-retain trade-off. This can be attributed to their reliance on parameter importance metrics to identify parameters that are important exclusively for the forget set, which is fundamentally limited by the superposition phenomenon. Due to the polysemantic nature of LLM parameters, such an importance metric may struggle to disentangle parameters associated with the forget and retain sets. In this work, we propose Representation-Guided Low-rank Unlearning (REGLU), a novel approach that leverages the geometric properties of representation spaces to achieve robust and precise unlearning. First, we develop a representation-guided initialization for LoRA that identifies the optimal subspace for selective forgetting. Second, we introduce a regularization loss that constrains the outputs of the LoRA update to lie in the orthogonal complement of the retain set's representation subspace, thereby minimizing interference with the model's performance on the retain set. We evaluate REGLU on the TOFU and WMDP benchmarks across multiple models. Our results demonstrate that REGLU consistently outperforms state-of-the-art baselines, achieving superior unlearning quality while maintaining higher model utility.

Automatic evaluation metrics are central to the development of machine translation systems, yet their robustness under domain shift remains unclear. Most metrics are developed on the Workshop on Machine Translation (WMT) benchmarks, raising concerns about their robustness to unseen domains. Prior studies that analyze unseen domains vary translation systems, annotators, or evaluation conditions, confounding domain effects with human annotation noise. To address these biases, we introduce a systematic multi-annotator Cross-Domain Error-Span-Annotation dataset (CD-ESA), comprising 18.8k human error span annotations across three language pairs, where we fix annotators within each language pair and evaluate translations of the same six translation systems across one seen news domain and two unseen technical domains. Using this dataset, we first find that automatic metrics appear surprisingly robust to domain-shifts at the segment level (up to 0.69 agreement), but this robustness largely disappears once we account for human label variation. Averaging annotations increases inter-annotator agreement by up to +0.11. Metrics struggle on the unseen chemical domain compared to humans (inter-annotator agreement of 0.78-0.83 vs. 0.96). We recommend comparing metric-human agreement against inter-annotator agreement, rather than comparing raw metric-human agreement alone, when evaluating across different domains.

RISC-V is emerging as a viable platform for automotive-grade embedded computing, with recent ISO 26262 ASIL-D certifications demonstrating readiness for safety-critical deployment in autonomous driving systems. However, functional safety in automotive systems is fundamentally a certification problem rather than a processor problem. The dominant costs arise from diagnostic coverage analysis, toolchain qualification, fault injection campaigns, safety-case generation, and compliance with ISO 26262, ISO 21448 (SOTIF), and ISO/SAE 21434. This paper analyzes the role of RISC-V in automotive functional safety, focusing on ISA openness, formal verifiability, custom extension control, debug transparency, and vendor-independent qualification. We examine autonomous driving safety requirements and map them to RISC-V architectural challenges such as lockstep execution, safety islands, mixed-criticality isolation, and secure debug. Rather than proposing a single algorithmic breakthrough, we present an analytical framework and research roadmap centered on certification economics as the primary optimization objective. We also discuss how selected ML methods, including LLM-assisted FMEDA generation, knowledge-graph-based safety case automation, reinforcement learning for fault injection, and graph neural networks for diagnostic coverage, can support certification workflows. We argue that the strongest outcome is not a faster core, but an ASIL-D-ready certifiable RISC-V platform.

Ancient inscriptions frequently suffer missing or corrupted regions from fragmentation, erosion, or other damage, hindering reading, and analysis. We review prior image restoration methods and their applicability to inscription image recovery, then introduce MESA (Multi-Exemplar, Style-Aware) -an image-level restoration method that uses well-preserved exemplar inscriptions (from the same epigraphic monument, material, or similar letterforms) to guide reconstruction of damaged text. MESA encodes VGG19 convolutional features as Gram matrices to capture exemplar texture, style, and stroke structure; for each neural network layer it selects the exemplar minimizing Mean-Squared Displacement (MSD) to the damaged input. Layer-wise contribution weights are derived from Optical Character Recognition-estimated character widths in the exemplar set to bias filters toward scales matching letter geometry, and a training mask preserves intact regions so synthesis is restricted to damaged areas. We also summarize prior network architectures and exemplar and single-image synthesis, inpainting, and Generative Adversarial Network (GAN) approaches, highlighting limitations that MESA addresses. Comparative experiments demonstrate the advantages of MESA. Finally, we provide a practical roadmap for choosing restoration strategies given available exemplars and metadata.

We introduce JuRe (Just Repair), a minimal denoising network for time series anomaly detection that exposes a central finding: architectural complexity is unnecessary when the training objective correctly implements the manifold-projection principle. JuRe consists of a single depthwise-separable convolutional residual block with hidden dimension 128, trained to repair corrupted time series windows and scored at inference by a fixed, parameter-free structural discrepancy function. Despite using no attention, no latent variable, and no adversarial component, JuRe ranks second on the TSB-AD multivariate benchmark (AUC-PR 0.404, 180 series, 17 datasets) and second on the UCR univariate archive by AUC-PR (0.198, 250 series), leading all neural baselines on AUC-PR and VUS-PR. Component ablation on TSB-AD identifies training-time corruption as the dominant factor ($Δ$AUC-PR $= 0.047$ on removal), confirming that the denoising objective, not network capacity, drives detection quality. Pairwise Wilcoxon signed-rank tests establish statistical significance against 21 of 25 baselines on TSB-AD. Code is available at the URL https://github.com/iis-esslingen/JuRe.

Large language model optimization has historically bifurcated into isolated data-centric and model-centric paradigms: the former manipulates involved samples through selection, augmentation, or poisoning, while the latter tunes model weights via masking, quantization, or low-rank adaptation. This paper establishes a unified \emph{data-parameter correspondence} revealing these seemingly disparate operations as dual manifestations of the same geometric structure on the statistical manifold $\mathcal{M}$. Grounded in the Fisher-Rao metric $g_{ij}(θ)$ and Legendre duality between natural ($θ$) and expectation ($η$) parameters, we identify three fundamental correspondences spanning the model lifecycle: 1. Geometric correspondence: data pruning and parameter sparsification equivalently reduce manifold volume via dual coordinate constraints; 2. Low-rank correspondence: in-context learning (ICL) and LoRA adaptation explore identical subspaces on the Grassmannian $\mathcal{G}(r,d)$, with $k$-shot samples geometrically equivalent to rank-$r$ updates; 3. Security-privacy correspondence: adversarial attacks exhibit cooperative amplification between data poisoning and parameter backdoors, whereas protective mechanisms follow cascading attenuation where data compression multiplicatively enhances parameter privacy. Extending from training through post-training compression to inference, this framework provides mathematical formalization for cross-community methodology transfer, demonstrating that cooperative optimization integrating data and parameter modalities may outperform isolated approaches across efficiency, robustness, and privacy dimensions.

This paper proposes StrEBM, a structured latent energy-based model for source-wise structured representation learning. The framework is motivated by a broader goal of promoting identifiable and decoupled latent organization by assigning different latent dimensions their own learnable structural biases, rather than constraining the entire latent representation with a single shared energy. In this sense, blind source separation is adopted here as a concrete and verifiable testbed, through which the evolution of latent dimensions toward distinct underlying components can be directly examined. In the proposed framework, latent trajectories are optimized directly together with an observation-generation map and source-wise structural parameters. Each latent dimension is associated with its own energy-based formulation, allowing different latent components to gradually evolve toward distinct source-like roles during training. In the present study, this source-wise energy design is instantiated using Gaussian-process-inspired energies with learnable length-scales, but the framework itself is not restricted to Gaussian processes and is intended as a more general structured latent EBM formulation. Experiments on synthetic multichannel signals under linear and nonlinear mixing settings show that the proposed model can recover source components effectively, providing an initial empirical validation of the framework. At the same time, the study reveals important optimization characteristics, including slow late-stage convergence and reduced stability under nonlinear observation mappings. These findings not only clarify the practical behavior of the current GP-based instantiation, but also establish a basis for future investigation of richer source-wise energy families and more robust nonlinear optimization strategies.

In this article, we generalize Unbounded Minimax, the state-of-the-art search algorithm for zero sums two-player games with perfect information to the framework of multiplayer games with perfect information. We experimentally show that this generalized algorithm also achieves better performance than the main multiplayer search algorithms.

While large language models have achieved remarkable performance in complex tasks, they still need a memory system to utilize historical experience in long-term interactions. Existing memory methods (e.g., A-Mem, Mem0) place excessive emphasis on organizing interactions by frequently rewriting them, however, this heavy reliance on summarization risks diluting essential contextual nuances and obscuring key retrieval features. To bridge this gap, we introduce AnchorMem, a novel memory framework inspired by the Proust Phenomenon in cognitive science, where a specific anchor triggers a holistic recollection. We propose a method that decouples the retrieval unit from the generation context. AnchorMem extracts atomic facts from interaction history to serve as retrieval anchors, while preserving the original context as the immutable context. To reveal implicit narrative cues, we construct an associative event graph that uses higher-order event links that bind sets of related facts into shared event representations, strengthening cross-memory integration without relying on generic entities as bridges. During retrieval, the system anchors queries to specific facts and events to locate relevant memories, but reconstructs the context using the associated raw chunks and events. Our method reconciles fine-grained retrieval with the contextual integrity of interactions. Experiments across three closed-source and open-source models on the LoCoMo benchmark demonstrate that AnchorMem significantly outperforms baselines. Code is available at https://github.com/RayNeo-AI-2025/AnchorMem.

In today's day and age, we face a challenge in detecting deepfake images because of the fast evolution of modern generative models and the poor generalization capability of existing methods. In this paper, we use an ensemble of fine-tuned vision transformers like DINOv2, AIMv2 and OpenCLIP's ViT-L/14 to create generalizable method to detect deepfakes. We use the DF-Wild dataset released as part of the IEEE SP Cup 2025, because it uses a challenging and diverse set of manipulations and generation techniques. We started our experiments with CNN classifiers trained on spatial features. Experimental results show that our ensemble outperforms individual models and strong CNN baselines, achieving an AUC of 96.77% and an Equal Error Rate (EER) of just 9% on the DF-Wild test set, beating the state-of-the-art deepfake detection algorithm Effort by 7.05% and 8% in AUC and EER respectively. This was the winning solution for SP Cup, presented at ICASSP 2025.

Recent advances in Vision-Language Models (VLMs) have substantially enhanced their ability across multimodal video understanding benchmarks spanning temporal, action, object, and spatial understanding. However, we identify a critical yet overlooked issue: when embedded on-screen text contradicts the visual scene, existing VLMs systematically hallucinate, prioritizing overlay textual semantics over the actual visual content. We define this phenomenon as Text Overlay-Induced Hallucination (TOIH). In this work, we propose VisualTextTrap, the first comprehensive benchmark, including large-scale human-validated samples with specifically designed evaluation metrics. In particular, we construct VisualTextTrap from widely-used public datasets using a scalable hybrid pipeline of VLMs assisted text generation and rigorous manual verification. The benchmark features 6,057 samples annotated across 88 fine-grained attributes within four dimensions, with hallucination intensity quantified on a five-level scale (L1--L5) that reflects the semantic contradiction between overlay text and visual reality. Moreover, we propose Visual Text Hallucination Mitigation Mixture-of-Experts (VTHM-MoE), a novel Vision-Text Disentanglement framework that employs a dual-encoder architecture. Concretely, four dimension-specialized expert modules spanning Temporal, Action, Object, and Spatial reasoning are first pre-trained to identify and leverage cross-modal discrepancies between textual semantics and actual video content. We develop an Adaptive Token Routing Strategy to enable dynamic expert allocation, conferring robust resistance to TOIH while preserving performance on uncontaminated videos. Extensive experiments conducted on our VisualTextTrap benchmark verify the effectiveness of VTHM-MoE, outperforming state-of-the-art counterparts with diverse video question answering tasks.

This paper investigates communication-efficient neural network transmission by exploiting structured symmetry constraints in convolutional kernels. Instead of transmitting all model parameters, we propose a degrees-of-freedom (DoF) based codec that sends only the unique coefficients implied by a chosen symmetry group, enabling deterministic reconstruction of the full weight tensor at the receiver. The proposed framework is evaluated under quantization and noisy channel conditions across multiple symmetry patterns, signal-to-noise ratios, and bit-widths. To improve robustness against transmission impairments, a projection step is further applied at the receiver to enforce consistency with the symmetry-invariant subspace, effectively denoising corrupted parameters. Experimental results on MNIST and CIFAR-10 using a DeepCNN architecture demonstrate that DoF-based transmission achieves substantial bandwidth reduction while preserving significantly higher accuracy than pruning-based baselines, which often suffer catastrophic degradation. Among the tested symmetries, \textit{central-skew symmetry} consistently provides the best accuracy-compression tradeoff, confirming that structured redundancy can be leveraged for reliable and efficient neural model delivery over constrained links.

Argumentation skills are an essential toolkit for large language models (LLMs). These skills are crucial in various use cases, including self-reflection, debating collaboratively for diverse answers, and countering hate speech. In this paper, we create the first benchmark for a standardized evaluation of LLM-based approaches to computational argumentation, encompassing 33 datasets from previous work in unified form. Using the benchmark, we evaluate the generalizability of five LLM families across 46 computational argumentation tasks that cover mining arguments, assessing perspectives, assessing argument quality, reasoning about arguments, and generating arguments. On the benchmark, we conduct an extensive systematic analysis of the contribution of few-shot examples, reasoning steps, model size, and training skills to the performance of LLMs on the computational argumentation tasks in the benchmark.

Equality saturation (EqSat) is a powerful optimization paradigm that compactly represents many equivalent programs in an e-graph and delays commitment until extraction selects a lowest-cost program. Making EqSat effective, therefore, requires not only domain-specific rewrite rules but also domain-specific strategies. Today, much of this strategy design is still manual, making it a major obstacle to automating e-graph-based compilers. Recent rule-synthesis frameworks can automatically infer large rewrite vocabularies from semantic specifications, but they also enlarge the rewrite space and further exacerbate e-graph explosion. Although large language models (LLMs) make automated strategy synthesis plausible, directly evolving backend code remains ineffective in practice. The search lacks reusable strategy abstractions and actionable feedback, and can easily trigger e-graph explosion or converge to poor designs. We present EggMind, an LLM-guided, end-to-end framework for synthesizing reusable EqSat strategies. At its core, EggMind introduces a domain-specific language, EqSatL, to represent EqSat strategies as explicit and inspectable artifacts. It then proposes an LLM-guided agentic workflow, equipped with novel techniques including proof-derived rewrite motif caching and tractability guidance, to search efficiently for high-quality strategies while keeping synthesis stable under e-graph growth. Evaluation shows that EggMind substantially improves the resource-quality trade-off on vectorization benchmarks, reducing final cost by 45.1% and peak RAM by 69.1% relative to full EqSat. We further show that the same methodology transfers effectively to an XLA-based tensor compiler, and demonstrate its practical potential in a logic-synthesis case study with augmented rewrite spaces.

Fine-grained medical image classification is challenged by subtle inter-class variations and visually ambiguous cases, where confidence estimates often exhibit uncertainty rather than being overconfident. In such scenarios, purely discriminative classifiers may achieve high overall accuracy yet still fail to distinguish between highly similar categories, leading to miscalibrated predictions. We propose T-DuMpRa, a teacher-guided dual-path multi-prototype retrieval-augmented framework, where discriminative classification and multi-prototype retrieval jointly drive both training and prediction. During training, we jointly optimize cross-entropy and supervised contrastive objectives to learn a cosine-compatible embedding geometry for reliable prototype matching. We further employ an exponential moving average (EMA) teacher to obtain smoother representations and build a multi-prototype memory bank by clustering teacher embeddings in the teacher embedding space. Our framework is plug-and-play: it can be easily integrated into existing classification models by constructing a compact prototype bank, thereby improving performance on visually ambiguous cases. At inference, we combine the classifier's predicted distribution with a similarity-based distribution computed via cosine matching to prototypes, and apply a conservative confidence-gated fusion that activates retrieval only when the classifier's prediction is uncertain and the retrieval evidence is decisive and conflicting, otherwise keeping confident predictions unchanged. On HAM10000 and ISIC2019, our method yields 0.68%-0.21% and 0.44%-2.69% improvements on 5 different backbones. And visualization analysis proves our model can enhance the model's ability to handle visually ambiguous cases.

Large language models are increasingly deployed to simulate patients for clinical training, research, and mental health tools, yet population-level validity remains largely untested. We introduce PsychBench, the first epidemiological audit of LLM patient simulation: 28,800 profiles from four frontier models (GPT-4o-mini, DeepSeek-V3, Gemini-3-Flash, GLM-4.7) evaluated against NHANES and NESARC-III baselines across 120 intersectional cohorts. The central finding is a coherence-fidelity dissociation: models produce clinically plausible individuals while misrepresenting the populations they are drawn from. Variance compression ranges from 14 percent (GLM-4.7) to 62 percent (DeepSeek-V3), eliminating the distributional tails of clinical reality. Despite test-retest correlations above r = 0.90, 36.66 percent of cases cross diagnostic thresholds between runs. Symptom correlation matrices diverge across demographic groups beyond split-half noise, with transgender populations diverging three to five times more than racial differences. Calibration bias is systematic and asymmetric. Models overestimate depression severity for most groups by 3.6 to 6.1 points (Cohen d = 1.13 to 1.91), consistent with training on clinical corpora with elevated base rates. For transgender women the direction inverts: models capture only 8 to 46 percent of documented minority stress elevation, yielding a -5.42 residual (d = -1.55). Models also attribute irritability to Black men and fatigue to women beyond matched controls, encoding racialized and gendered assumptions. Patterns replicate across US and Chinese architectures, indicating failures tied to current training paradigms rather than isolated implementations. For most users, LLM mental health tools risk pathologizing ordinary distress; for transgender users, algorithmic erasure of genuine need. The patients look right. They do not represent real populations.

While recent Spoken Language Models (SLMs) have been actively deployed in real-world scenarios, they lack the capability to discern Third-Party Interruptions (TPI) from the primary user's ongoing flow, leaving them vulnerable to contextual failures. To bridge this gap, we introduce TPI-Train, a dataset of 88K instances designed with speaker-aware hard negatives to enforce acoustic cue prioritization for interruption handling, and TPI-Bench, a comprehensive evaluation framework designed to rigorously measure the interruption-handling strategy and precise speaker discrimination in deceptive contexts. Experiments demonstrate that our dataset design mitigates semantic shortcut learning-a critical pitfall where models exploit semantic context while neglecting acoustic signals essential for discerning speaker changes. We believe our work establishes a foundational resource for overcoming text-dominated unimodal reliance in SLMs, paving the way for more robust multi-party spoken interaction. The code for the framework is publicly available at https://tpi-va.github.io

Vision-Language Models (VLMs) excel at photorealistic generation, yet often struggle to represent abstract meaning such as idiomatic interpretations of noun compounds. To study whether high visual fidelity interferes with idiomatic compositionality under visual abstraction, we introduce DIVA, a controlled benchmark that replaces high-fidelity visual detail with schematic iconicity by generating paired, sense-anchored visualizations for literal and idiomatic readings. We further propose Semantic Alignment Gap ($Δ$), an architecture-agnostic metric that quantifies divergence between literal and idiomatic visual grounding. We additionally introduce a directional signed bias $b(t)$ to separately measure the direction and strength of literal preference. Evaluating 8 recent VLMs, we reveal a consistent Literal Superiority Bias: model scale alone does not resolve literal preference, and increased visual fidelity is associated with weaker symbolic alignment, suggesting cognitive interference from hyper-realistic imagery. Our findings suggest that improving compositional understanding requires iconographic abstraction of visual input and anchoring interpretation and generation in intended meaning.

Large language models are increasingly deployed as complex agentic systems that scale with task complexity. While prior work has extensively explored model- and system-level scaling, algorithm- and task-level scaling remain largely unaddressed, constraining the full potential of agentic systems. At the algorithm level, allocating additional inference-time computation can enhance workflow capacity but introduces cross-path redundancy: overlapping computations across multiple reasoning branches. At the task level, complex tasks can be decomposed into subproblems and delegated across multiple agents for improved scalability and parallelism. However, existing infrastructures' scheduling is unaware of the existence of multiple agents, missing opportunities to optimize resource allocation. We propose Hive, a multi-agent infrastructure that enables algorithm- and task-level scaling. Hive features a description frontend that captures per-agent behavior and supports test-time scaling algorithms. Leveraging this specification, our backend introduces two key mechanisms: Logits Cache that reuses intermediate logits across redundant sampling paths to mitigate cross-path redundancy at the algorithm level, and Agent-Aware Scheduling that efficiently allocates compute and KV-cache resources according to agent contributions at the task level. Experiments show that Logits Cache achieves an average speedup of $1.11\times$-$1.76\times$ for re-sampling, and Agent-Aware Scheduling reduces the hotspot miss rate by $33\%$-$51\%$.

Automated simulator construction requires distributional fidelity, distinguishing it from generic code generation. We identify two failure modes in long-horizon LLM agents: contextual drift and optimization instability arising from conflating structural and parametric errors. We propose SOCIA-EVO, a dual-anchored evolutionary framework. SOCIA-EVO introduces: (1) a static blueprint to enforce empirical constraints; (2) a bi-level optimization to decouple structural refinement from parameter calibration; and (3) a self-curating Strategy Playbook that manages remedial hypotheses via Bayesian-weighted retrieval. By falsifying ineffective strategies through execution feedback, SOCIA-EVO achieves robust convergence, generating simulators that are statistically consistent with observational data. The code and data of SOCIA-EVO are available here: https://github.com/cruiseresearchgroup/SOCIA/tree/evo.

Time series forecasting is traditionally dominated by sequence-based architectures such as recurrent neural networks and attention mechanisms, which process all time steps uniformly and often incur substantial computational cost. However, real-world temporal signals typically exhibit heterogeneous structure, where informative patterns are sparsely distributed and interspersed with redundant observations. This work introduces \textbf{SPaRSe-TIME}, a structured and computationally efficient framework that models time series through a decomposition into three complementary components: saliency, memory, and trend. The proposed approach reformulates temporal modeling as a projection onto informative subspaces, where saliency acts as a data-dependent sparsification operator, memory captures dominant low-rank temporal patterns, and trend encodes low-frequency dynamics. These components are integrated through a lightweight, adaptive mapping that enables simplified, selective, and interpretable temporal reasoning. Extensive experiments on diverse real-world datasets demonstrate that SPaRSe-TIME achieves competitive predictive performance compared to recurrent and attention-based architectures, while significantly reducing computational complexity. The model is particularly effective in structured time series with clear temporal components and provides explicit interpretability through component-wise contributions. Furthermore, analysis reveals both the strengths and limitations of decomposition-based modeling, highlighting challenges in highly stochastic and complex multivariate settings. Overall, SPaRSe-TIME offers a principled alternative to monolithic sequence models, bridging efficiency, interpretability, and performance, and providing a scalable framework for time series learning.

Just like traditional BPM systems, agentic BPM systems are built around a specification of the process under consideration. Their distinguishing feature, however, is that the execution of the process is driven by multiple autonomous decision-makers, referred to as agents. Since such agents cannot be fully controlled, the process specification is augmented with explicit objectives, or goals, assigned to the participating agents. Agents then pursue these goals, at least to the best of their efforts, under suitable assumptions on the behavior of others, by adopting appropriate strategies. Centrally, the organization enacting the process can use these specifications to provide guardrails on the decision-making capabilities of agents at the strategy level. This paper sets up the mathematical foundations of such systems in three key settings and analyzes four foundational problems of agentic BPM.

In this book we promote logical computational linguistics as opposed to statistical computational linguistics. In particular, we provide a logical semantic interface. This book assembles more than twenty years of research work on type logical grammar, and adds new ideas and material. Chains of statistical dependencies of less than one hundred per cent confidence tend monotonically to zero. Chains of logical dependencies of any length maintain one hundred per cent confidence end to end. We aspire to enable perfect syntactic and semantic processing in life-critical NLP applications.

When task-specific labels are not available, it becomes difficult to select an embedding model for a specific target corpus. Existing labelless measures based on kernel estimators or Gaussian mixes fail in high-dimensional space, resulting in unstable rankings. We propose a flow-based labelless representation embedding evaluation (FLARE), which utilizes normalized streams to estimate information sufficiency directly from log-likelihood and avoid distance-based density estimation. We give a finite sample boundary, indicating that the estimation error depends on the intrinsic dimension of the data manifold rather than the original embedding dimension. On 11 datasets and 8 embedders, FLARE reached Spearman's $ρ$ of 0.90 under the supervised benchmark and remained stable in high-dimensional embeddings ($d \geq 3{,}584$) as the existing labelless baseline collapsed.

Diabetic retinopathy (DR) is a leading cause of vision impairment worldwide, and automated grading systems play a crucial role in large-scale screening programs. However, deep learning models often exhibit degraded performance when deployed across datasets acquired under different imaging conditions. This study presents a robust dual-resolution deep learning framework for DR grading that integrates attention-based feature fusion with ordinal regression to improve cross-dataset generalization. The proposed method employs two parallel EfficientNet backbones operating at different spatial resolutions to capture complementary retinal features. A learnable attention mechanism adaptively fuses multi-resolution representations, while an ordinal regression formulation based on the cumulative link model (CORAL) explicitly accounts for the ordered nature of DR severity levels. To mitigate domain discrepancies between datasets, a preprocessing strategy combining circular cropping, contrast enhancement, and histogram matching is applied. The model was trained on the APTOS 2019 dataset and evaluated on both an internal validation split and an external Messidor-2 test set. Experimental results demonstrate strong grading performance, achieving a quadratic weighted kappa (QWK) of 0.88 on the APTOS validation set and 0.68 on the unseen Messidor-2 dataset, indicating improved robustness for cross-dataset DR grading applications.

Clinical guidelines, typically developed by independent specialty societies, inherently exhibit substantial fragmentation, redundancy, and logical contradiction. These inconsistencies, particularly when applied to patients with multimorbidity, not only cause cognitive dissonance for clinicians but also introduce catastrophic noise into AI systems, rendering the standard Retrieval-Augmented Generation (RAG) system fragile and prone to hallucination. To address this fundamental reliability crisis, we introduce a Neuro-Symbolic framework that automates the detection of recommendation redundancies and conflicts. Our pipeline employs a multi-agent system to translate unstructured clinical natural language into rigorous symbolic logic language, which is then verified by a Satisfiability (SAT) solver. By formulating a hierarchical taxonomy of logical rule interactions, we identify a critical category termed Local Conflict - a decision conflict arising from the intersection of comorbidities. Evaluating our system on a curated benchmark of 12 authoritative SGLT2 inhibitor guidelines, we reveal that 90.6% of conflicts are Local, a structural complexity that single-disease guidelines fail to address. While state-of-the-art LLMs fail in detecting these conflicts, our neuro-symbolic approach achieves an F1 score of 0.861. This work demonstrates that logical verification must precede retrieval, establishing a new technical standard for automated knowledge coordination in medical AI.

Unlike code completion, debugging requires localizing faults and applying targeted edits. We observe that frontier LLMs often regenerate correct but over-edited solutions during debugging. To evaluate how far LLMs are from precise debugging, we introduce the Precise Debugging Benchmark (PDB) framework, which automatically converts any coding dataset into a debugging benchmark with precision-aware evaluation. PDB generates buggy programs by synthesizing verified atomic bugs and composing them into multi-bug programs. We define two novel metrics, edit-level precision and bug-level recall, which measures how many necessary edits are made and how many bugs are resolved. We release two evaluation benchmarks: PDB-Single-Hard on single-line bugs, and PDB-Multi on multi-line bugs. Experiments show that frontier models, such as GPT-5.1-Codex and DeepSeek-V3.2-Thinking, achieve unit-test pass rates above 76% but exhibit precision below 45%, even when explicitly instructed to perform minimal debugging. Finally, we show that iterative and agentic debugging strategies do not substantially improve precision or recall, highlighting the need to rethink post-training pipelines for coding models.

Agentic retrieval-augmented generation (RAG) systems enable large language models (LLMs) to solve complex tasks through multi-step interaction with external retrieval tools. However, such multi-step interaction often involves redundant search steps, incurring substantial computational cost and latency. Prior work limits search depth (i.e., the number of search steps) to reduce cost, but this often leads to underexploration of complex questions. To address this, we first investigate how search depth affects accuracy and find a minimal sufficient search depth that defines an accuracy-efficiency trade-off, jointly determined by question complexity and the agent's capability. Furthermore, we propose AutoSearch, a reinforcement learning (RL) framework that evaluates each search step via self-generated intermediate answers. By a self-answering mechanism, AutoSearch identifies the minimal sufficient search depth and promotes efficient search by rewarding its attainment while penalizing over-searching. In addition, reward mechanisms are introduced to stabilize search behavior and improve answer quality on complex questions. Extensive experiments on multiple benchmarks show that AutoSearch achieves a superior accuracy-efficiency trade-off, alleviating over-searching while preserving search quality.

Design patterns provide reusable solutions to recurring software design problems. Automatically detecting these patterns in source code can help bootstrap new developers' understanding of unfamiliar software system architectures, and can help experienced developers to quickly identify and rectify potential quality issues. While many prior research works have explored graph based and machine-learning based detection techniques, this work evaluates the design pattern recognition capabilities of four Large Language Models and two ensemble approaches consisting three out of the four models. We also compare their performance when prompted with a) Source code, b) PlantUML representation of source code, and c) Text-based descriptions of the source code. We investigate the detection of five design patterns: singleton, adapter, bridge, composite and decorator. Our preliminary results indicate that LLMs show promise for automatically detecting design patterns, with NextCoder and Gemma 3 demonstrating comparatively higher accuracy than other models evaluated, and the ensemble approaches enhancing the overall efficiency of design pattern detection. We identify several directions for future work.

This paper investigates the length problem in sequence-level relative reinforcement learning. We observe that, although existing methods partially alleviate length-related phenomena, a more fundamental issue remains insufficiently characterized: the comparison units used during training lack inherent comparability. Building on this observation, we propose a new perspective: the length problem should not be viewed merely as a loss-scaling or normalization bias, but rather as a \emph{comparison unit construction} problem. We further establish a sample-construction-based training framework that, instead of applying post-hoc corrections to unequal-length responses, proactively constructs equal-length, alignable, and comparable training segments during generation. Within this framework, we propose EqLen, a concrete method applicable to group-relative comparison algorithms such as GRPO, GSPO, and RLOO. Through dual-track synchronous generation, prefix inheritance, and segment masking, EqLen efficiently collects effective equal-length training segments and enables stable

We present the first portfolio-level validation of MarketSenseAI, a deployed multi-agent LLM equity system. All signals are generated live at each observation date, eliminating look-ahead bias. The system routes four specialist agents (News, Fundamentals, Dynamics, and Macro) through a synthesis agent that issues a monthly equity thesis and recommendation for each stock in its coverage universe, and we ask two questions: do its buy recommendations add value over both passive benchmarks and random selection, and what does the internal agent structure reveal about the source of the edge? On the S&P 500 cohort (19 months) the strong-buy equal-weight portfolio earns +2.18%/month against a passive equal-weight benchmark of +1.15% (approximating RSP), a +25.2% compound excess, and ranks at the 99.7th percentile of 10,000 Monte Carlo portfolios (p=0.003). The S&P 100 cohort (35 months) delivers a +30.5% compound excess over EQWL with consistent direction but formal significance not reached, limited by the small average selection of ~10 stocks per month. Non-negative least-squares projection of thesis embeddings onto agent embeddings reveals an adaptive-integration mechanism. Agent contributions rotate with market regime (Fundamentals leads on S&P 500, Macro on S&P 100, Dynamics acts as an episodic momentum signal) and this agent rotation moves in lockstep with both the sector composition of strong-buy selections and identifiable macro-calendar events, three independent views of the same underlying adaptation. The recommendation's cross-sectional Information Coefficient is statistically significant on S&P 500 (ICIR=+0.489, p=0.024). These results suggest that multi-agent LLM equity systems can identify sources of alpha beyond what classical factor models capture, and that the buy signal functions as an effective universe-filter that can sit upstream of any portfolio-construction process.

Retrieval-Augmented Generation (RAG) enhances the factuality of Large Language Models (LLMs) by incorporating retrieved documents and/or generated context. However, LLMs often exhibit a stylistic bias when presented with mixed contexts, favoring fluent but hallucinated generated content over factually grounded yet disorganized retrieved evidence. This phenomenon reveals that the utility of retrieved information is bottlenecked by its presentation. To bridge this gap, we propose QREAM, a style-controlled rewriter that aligns retrieved documents with a question-oriented style while preserving facts, better for LLM readers to utilize. Our framework consists of two stages: (1) QREAM-ICL, which uses stylistic seeds to guide iterative rewriting exploration; and (2) QREAM-FT, a lightweight student model distilled from denoised ICL outputs. QREAM-FT employs dual-criteria rejection sampling, filtering based on answer correctness and factual consistency to ensure high-quality supervision. QREAM seamlessly integrates into existing RAG pipelines as a plug-and-play module. Experiments demonstrate that QREAM consistently enhances advanced RAG pipelines, yielding up to 8% relative improvement with negligible latency overhead, effectively balancing question relevance with factual grounding.

Graph transformers achieve strong results on molecular and long-range reasoning tasks, yet remain hampered by over-smoothing (the progressive collapse of node representations with depth) and attention entropy degeneration. We observe that these pathologies share a root cause with attention sinks in large language models: softmax attention's sum-to-one constraint forces every node to attend somewhere, even when no informative signal exists. Motivated by recent findings that element-wise sigmoid gating eliminates attention sinks in large language models, we propose SigGate-GT, a graph transformer that applies learned, per-head sigmoid gates to the attention output within the GraphGPS framework. Each gate can suppress activations toward zero, enabling heads to selectively silence uninformative connections. On five standard benchmarks, SigGate-GT matches the prior best on ZINC (0.059 MAE) and sets new state-of-the-art on ogbg-molhiv (82.47% ROC-AUC), with statistically significant gains over GraphGPS across all five datasets ($p < 0.05$). Ablations show that gating reduces over-smoothing by 30% (mean relative MAD gain across 4-16 layers), increases attention entropy, and stabilizes training across a $10\times$ learning rate range, with about 1% parameter overhead on OGB.

Modern generative models still lack human-level creativity, particularly in multi-branch diversity. Prior approaches to address this problem often incur heavy computation or strong dependency on model architecture. Therefore, we introduce UAG(Universal Avoidance Generation), a model-agnostic and computationally efficient generation strategy that penalizes similarity among previously generated outputs. Thus, UAG can enhance multi-branch diversity across both diffusion and transformer models, with minimal additional computation. In experiments, our method achieves up to 1.9 times higher diversity, runs 4.4 times faster, and requires only 1/64 of the FLOPs compared to state-of-the-art methods. The full code is https://anonymous.4open.science/r/2026_ACL_Universal/.

Grounded Multimodal Named Entity Recognition (GMNER) aims to jointly identify named entity mentions in text, predict their semantic types, and ground each entity to a corresponding visual region in an associated image. Existing approaches predominantly adopt pipeline-based architectures that decouple textual entity recognition and visual grounding, leading to error accumulation and suboptimal joint optimization. In this paper, we propose E2E-GMNER, a fully end-to-end generative framework that unifies entity recognition, semantic typing, visual grounding, and implicit knowledge reasoning within a single multimodal large language model. We formulate GMNER as an instruction-tuned conditional generation task and incorporate chain-of-thought reasoning to enable the model to adaptively determine when visual evidence or background knowledge is informative, reducing reliance on noisy cues. To further address the instability of generative bounding box prediction, we introduce Gaussian Risk-Aware Box Perturbation (GRBP), which replaces hard box supervision with probabilistically perturbed soft targets to improve robustness against annotation noise and discretization errors. Extensive experiments on the Twitter-GMNER and Twitter-FMNERG benchmarks demonstrate that E2E-GMNER achieves highly competitive performance compared with state of the art methods, validating the effectiveness of unified end-to-end optimization and noise-aware grounding supervision. Code is available at:https://github.com/Finch-coder/E2E-GMNER

Safe clinical deployment of Large Language Models (LLMs) requires not only high accuracy but also robust uncertainty calibration to ensure models defer to clinicians when appropriate. Our paper investigates how social descriptors of a patient (specifically sexual orientation and religious affiliation) distort these uncertainty signals and model accuracy. Evaluating nine general-purpose and biomedical LLMs on 2,364 medical questions and their counterfactual variants, we demonstrate that identity markers cause a "calibration crisis". "Homosexual" markers consistently trigger performance drops, and intersectional identities produce idiosyncratic, non-additive harms to calibration. Moreover, a clinician-validated case study in an open-ended generation setting confirms that these failures are not an artifact of the multiple-choice format. Our results demonstrate that the presence of social identity cues does not merely shift predictions; it affects the reliability of confidence signals, posing a significant risk to equitable care and safe deployment in confidence-based clinical workflows.

Reinforcement learning (RL) has emerged as a powerful post-training paradigm for enhancing the reasoning capabilities of large language models (LLMs). However, reinforcement learning for LLMs faces substantial data scarcity challenges, including the limited availability of high-quality external supervision and the constrained volume of model-generated experience. These limitations make data-efficient reinforcement learning a critical research direction. In this survey, we present the first systematic review of reinforcement learning for LLMs under data scarcity. We propose a bottom-up hierarchical framework built around three complementary perspectives: the data-centric perspective, the training-centric perspective, and the framework-centric perspective. We develop a taxonomy of existing methods, summarize representative approaches in each category, and analyze their strengths and limitations. Our taxonomy aims to provide a clear conceptual foundation for understanding the design space of data-efficient RL for LLMs and to guide researchers working in this emerging area. We hope this survey offers a comprehensive roadmap for future research and inspires new directions toward more efficient and scalable reinforcement learning post-training for LLMs.

Discrete diffusion models form a powerful class of generative models across diverse domains, including text and graphs. However, existing approaches face fundamental limitations. Masked diffusion models suffer from irreversible errors due to early unmasking, while uniform diffusion models, despite enabling self-correction, often yield low-quality samples due to their strong reliance on intermediate latent states. We introduce IDDM, an Interpolating Discrete Diffusion Model, that improves diffusion by reducing dependence on intermediate latent states. Central to IDDM is a controllable resampling mechanism that partially resets probability mass to the marginal distribution, mitigating error accumulation and enabling more effective token corrections. IDDM specifies a generative process whose transitions interpolate between staying at the current state, resampling from a prior, and flipping toward the target state, while enforcing marginal consistency and fully decoupling training from inference. We benchmark our model against state-of-the-art discrete diffusion models across molecular graph generation as well as text generation tasks, demonstrating competitive performance.

Research artifacts are distributed primarily as reader-oriented documents like PDFs. This creates a bottleneck for increasingly agent-assisted and agent-native research workflows, in which LLM agents need to infer fine-grained, task-relevant information from lengthy full documents, a process that is expensive, repetitive, and unstable at scale. We introduce Knows, a lightweight companion specification that binds structured claims, evidence, provenance, and verifiable relations to existing research artifacts in a form LLM agents can consume directly. Knows addresses the gap with a thin YAML sidecar (KnowsRecord) that coexists with the original PDF, requiring no changes to the publication itself, and validated by a deterministic schema linter. We evaluate Knows on 140 comprehension questions across 20 papers spanning 14 academic disciplines, comparing PDF-only, sidecar-only, and hybrid conditions across six LLM agents of varying capacity. Weak models (0.8B--2B parameters) improve from 19--25\% to 47--67\% accuracy (+29 to +42 percentage points) when reading sidecar instead of PDF, while consuming 29--86\% fewer input tokens; an LLM-as-judge re-scoring confirms that weak-model sidecar accuracy (75--77\%) approaches stronger-model PDF accuracy (78--83\%). Beyond this controlled evaluation, a community sidecar hub at https://knows.academy/ has already indexed over ten thousand publications and continues to grow daily, providing independent evidence that the format is adoption-ready at scale.

As the capability frontier of autonomous agents continues to expand, they are increasingly able to complete specialized tasks through plug-and-play external skills. Yet current benchmarks mostly test whether models can use provided skills, leaving open whether they can discover skills from experience, repair them after failure, and maintain a coherent library over time. We introduce SkillFlow, a benchmark of 166 tasks across 20 families in which task construction within each family follows a Domain-Agnostic Execution Flow (DAEF) that defines an agent workflow framework, allowing these tasks to share a consistent workflow. Agents are evaluated under an Agentic Lifelong Learning protocol in which they begin without skills, solve tasks sequentially within each family, externalize lessons through trajectory- and rubric-driven skill patches, and carry the updated library forward. Experiments reveal a substantial capability gap. For Claude Opus 4.6, lifelong skill evolution improves task success from 62.65% to 71.08% (+8.43 points). However, high skill usage does not necessarily imply high utility: Kimi K2.5 gains only +0.60 points despite 66.87% skill usage, while Qwen-Coder-Next reaches only a 44.58% task completion rate and still regresses relative to the vanilla setting. SkillFlow contributes a structured testbed for this direction and an in-depth empirical analysis of skill discovery, patching, transfer, and their failure modes under lifelong evaluation.

Test-time scaling improves the reasoning performance of large language models but often results in token-inefficient overthinking, where models continue reasoning beyond what is necessary for a correct answer. Existing dynamic early-exit methods typically rely on single-step confidence signals, which are often unreliable for detecting reasoning convergence in multi-step settings. To mitigate this limitation, we propose TRACE, a training-free framework for efficient test-time scaling that determines when to terminate reasoning based on temporal aggregation of multi-step evidence rather than instantaneous signals. TRACE detects reasoning convergence over time by aggregating two complementary signals across recent reasoning steps: answer consistency, capturing the persistence of predicted answers, and confidence trajectory, modeling the temporal evolution of model confidence. Benefiting from these two factors, TRACE can accurately determine whether the reasoning process has converged, thereby promptly halting inference and effectively avoiding redundant reasoning steps. Extensive experiments on multiple challenging benchmarks show that TRACE reduces reasoning token usage by 25-30% on average while maintaining accuracy within 1-2% of full-length reasoning, consistently outperforming existing dynamic reasoning methods.

Detecting harmful content in multi turn dialogue requires reasoning over the full conversational context rather than isolated utterances. However, most existing methods rely mainly on models internal parametric knowledge, without explicit grounding in external normative principles. This often leads to inconsistent judgments in socially nuanced contexts, limited interpretability, and redundant reasoning across turns. To address this, we propose RoTRAG, a retrieval augmented framework that incorporates concise human written moral norms, called Rules of Thumb (RoTs), into LLM based harm assessment. For each turn, RoTRAG retrieves relevant RoTs from an external corpus and uses them as explicit normative evidence for turn level reasoning and final severity classification. To improve efficiency, we further introduce a lightweight binary routing classifier that decides whether a new turn requires retrieval grounded reasoning or can reuse existing context. Experiments on ProsocialDialog and Safety Reasoning Multi Turn Dialogue show that RoTRAG consistently improves both harm classification and severity estimation over competitive baselines, with an average relative gain of around 40% in F1 across benchmark datasets and an average relative reduction of 8.4% in distributional error, while reducing redundant computation without sacrificing performance.

The scarcity of labeled clinical data in oncology makes Few-Shot Learning (FSL) a critical framework for Computer Aided Diagnostics, but we observed that standard Prototypical Networks often struggle with the "prototype instability" caused by morphological noise and high intra-class variance in brain tumor scans. Our work attempts to minimize this by integrating a non-linear Logistic Chaos Module into a fine-tuned ResNet-18 backbone creating the Chaos-Enhanced ProtoNet(CE-ProtoNet). Using the deterministic ergodicity of the logistic chaos map we inject controlled perturbations into support features during episodic training-essentially for "stress testing" the embedding space. This process makes the model to converge on noise-invariant representations without increasing computational overhead. Testing this on a 4-way 5-shot brain tumor classification task, we found that a 15% chaotic injection level worked efficiently to stabilize high-dimensional clusters and reduce class dispersion. Our method achieved a peak test accuracy of 84.52%, outperforming standard ProtoNet. Our results suggest the idea of using chaotic perturbation as an efficient, low-overhead regularization tool, for the data-scarce regimes.

Aligning large language models with human preferences must balance two competing goals: responding helpfully to legitimate requests and reliably refusing harmful ones. Most preference-based safety alignment methods collapse safety into a single scalar that is applied uniformly to every preference pair. The result is a model that looks safe on average but stays relatively unsafe on a minority of harm categories. We cast safety alignment as a per-category constrained optimization problem and derive Cat-DPO, a direct-preference-optimization algorithm with a separate adaptive safety margin for each harm category. The margin tightens when the model still produces unsafe responses on a category and relaxes once the model catches up, so the training signal tracks each category's current difficulty rather than averaging under one global rate. Across two LLM backbones and six preference-learning baselines, Cat-DPO iimproves aggregate helpfulness and harmlessness and compresses per-category safety variance and the best-to-worst gap, offering a drop-in per-category refinement of direct preference safety alignment.

Long Chain-of-Thought (CoT) reasoning is pivotal for the success of recent reasoning models but suffers from high computational overhead and latency. While prior works attempt to compress CoT via external compressor, they often fail to align with the model's internal reasoning dynamics, resulting in the loss of critical logical steps. This paper presents \textbf{C}ompressing \textbf{R}edundancy in Chain-of-Thought via \textbf{I}ntrinsic \textbf{S}aliency \textbf{P}runing (\textbf{CRISP}), a framework that compresses CoT by exploiting the model's intrinsic saliency. Our analysis reveals a distinct phenomenon: the reasoning termination token \texttt{[object Object]} acts as an information anchor, where its attention pattern effectively demarcates essential reasoning from redundancy. Based on this finding, we design a policy that utilizes these intrinsic attention signals to guide atomic compression operations. In contrast to coarse-grained pruning strategies, CRISP strategically distills the reasoning chain to maximize information density while preserving logical coherence. Empirical results across various backbone models and mathematical datasets demonstrate that CRISP achieves a 50-60% reduction in token count without compromising accuracy, effectively mitigating the efficiency bottleneck of long-context reasoning. We open-source our implementation to facilitate further research in efficient reasoning.

Comprehensive understanding of time series remains a significant challenge for Large Language Models (LLMs). Current research is hindered by fragmented task definitions and benchmarks with inherent ambiguities, precluding rigorous evaluation and the development of unified Time Series Reasoning Models(TSRMs). To bridge this gap, we formalize Time Series Reasoning (TSR) via a four-level taxonomy of increasing cognitive complexity. We introduce HiTSR, a hierarchical time series reasoning dataset comprising 83k samples with diverse task combinations and verified Chain-of-Thought (CoT) trajectories. Leveraging HiTSR, we propose LLaTiSA, a strong TSRM that integrates visualized patterns with precision-calibrated numerical tables to enhance the temporal perception of Vision-Language Models (VLMs). Through a multi-stage curriculum fine-tuning strategy, LLaTiSA achieves superior performance and exhibits robust out-of-distribution generalization across diverse TSR tasks and real-world scenarios. Our code is available at https://github.com/RainingNovember/LLaTiSA.

Reliable Large Language Models (LLMs) should abstain when confidence is insufficient. However, prior studies often treat refusal as a generic "I don't know'', failing to distinguish input-level ambiguity (data uncertainty) from capability limitations (model uncertainty). This lack of distinction limits downstream action decisions like requesting clarification or invoking external tools. In this work, we introduce UA-Bench, a benchmark of over 3,500 questions drawn from six datasets spanning knowledge-intensive and reasoning-intensive tasks, designed to evaluate explicit uncertainty attribution. An evaluation of 18 frontier LLMs shows that even state-of-the-art models struggle to reliably discriminate between data uncertainty and model uncertainty, and that high answer accuracy does not necessarily imply strong uncertainty attribution ability. To narrow this gap, we propose a lightweight data synthesis and reinforcement learning strategy. Experiments on both Qwen3-4B-Instruct-2507 and Qwen3-8B in thinking mode show that the proposed method improves uncertainty attribution while preserving answer accuracy. Our code and data are publicly available now.

LLMs are widely used for code generation and mathematical reasoning tasks where they are required to generate structured output. They either need to reason about code, generate code for a given specification, or reason using programs of thought. The typical approach to code generation is to prompt the model and generate samples until an appropriate program is obtained. Within this process, sampling $n$ programs from the language model requires $n$ GPU compute-intensive generations which becomes prohibitively expensive for larger values of $n$. In this work, we address this limitation by exposing the LLM's distribution within the generated programs themselves. We propose a novel test-time framework we dub probabilistic programs of thought to obtain more samples from the model with fewer LLM generations. Given a program generated by a model and the associated next-token probabilities, we build a probabilistic program that compactly represents exponentially many deterministic programs. Since performing probabilistic reasoning in this probabilistic program is much cheaper, our approach allows sampling new programs without any additional GPU compute and little CPU overhead. We instantiate our approach on benchmarks for code generation, code understanding and mathematical reasoning and report improvements in performance with fewer generations from the LLM.

Supervised fine-tuning of large language models relies on human-annotated data, yet annotation pipelines routinely involve multiple crowdworkers of heterogeneous expertise. Standard practice aggregates labels via majority vote or simple averaging, discarding annotator identity and causing the model to absorb the errors of unreliable annotators directly into its parameters. We propose REALM, a method that jointly learns the model parameters and a scalar expertise value for each annotator entirely unsupervised, requiring no supervision beyond annotator identity. The key idea is to model each observed label as a mixture between the model's prediction and a uniform random guess, weighted by the annotator's learned expertise. We extend REALM to a multi-task setting via a learned expertise matrix that captures per-annotator reliability across tasks. We evaluate on five question answering benchmarks, fine-tuning three sizes of Flan-T5 under simulated noisy annotations. The proposed algorithm consistently outperforms the naive noisy SFT in the large majority of single- and multi-task settings, across datasets, model sizes, and noise types, with accuracy improvements of up to $50\%$ in the most adversarial regime and gains that grow with model capacity.

RTL program repair remains a critical bottleneck in hardware design and verification. Traditional automatic program repair (APR) methods rely on predefined templates and synthesis, limiting their bug coverage. Large language models (LLMs) and coding agents based on them offer flexibility but suffer from randomness and context corruption when handling long RTL code and waveforms. We present Clover, a neural-symbolic agentic harness that orchestrates RTL repair as a structured search over code manipulations to explore a validated solution for the bug. Recognizing that different repair operations favor distinct strategies, Clover dynamically dispatches tasks to specialized LLM agents or symbolic solvers. At its core, Clover introduces stochastic tree-of-thoughts, a test-time scaling mechanism that manages the main agent's context as a search tree, balancing exploration and exploitation for reliable outcomes. An RTL-specific toolbox further empowers agents to interact with the debugging environment. Evaluated on the RTL-repair benchmark, Clover fixes 96.8% of bugs within a fixed time limit, covering 94% and 63% more bugs than both pure traditional and LLM-based baselines, respectively, while achieving an average pass@1 rate of 87.5%, demonstrating high reliability and effectiveness.

While progress in GUI agents has been largely driven by industrial-scale training, ungrounded hallucinations often trigger cascading failures in real-world deployments.Unlike general VLM domains, the GUI agent field lacks a hallucination-focused suite for fine-grained diagnosis, reliable evaluation, and targeted mitigation.To bridge this gap, we introduce HalluClear, a comprehensive suite for hallucination mitigation in GUI agents as a complement to computation-intensive scaling. HalluClear comprises: (1) a GUI-specific hallucination taxonomy derived from empirical failure analysis; (2) a calibrated three-stage evaluation workflow which enhances VLM-as-a-judge reliability via expert-annotated benchmarking and ensemble credibility estimation; and (3) a mitigation scheme based on closed-loop structured reasoning, enabling lightweight continual post-training with cold-start initialization for both generalist and GUI-specialist agents. Experiments across representative agents and public benchmarks demonstrate that post-training on only 9K samples within our suite can significantly reduce hallucinations, thereby improving grounding and action fidelity, offering a compute-efficient pathway to robust GUI automation.

User preferences evolve across months of interaction, and tracking them requires inferring when a stated preference has been changed by a subsequent life event. We define this problem as long-horizon personalization and observe that progress on it is limited by data availability and measurement, with no existing resource providing both naturalistic long-horizon interactions and the ground-truth provenance needed to diagnose why models fail. We introduce a data generator that produces conversations from a structured mental state graph, yielding ground-truth provenance for every preference change across 6-month timelines, and from it construct HorizonBench, a benchmark of 4,245 items from 360 simulated users with 6-month conversation histories averaging ~4,300 turns and ~163K tokens. HorizonBench provides a testbed for long-context modeling, memory-augmented architectures, theory-of-mind reasoning, and user modeling. Across 25 frontier models, the best model reaches 52.8% and most score at or below the 20% chance baseline. When these models err on evolved preferences, over a third of the time they select the user's originally stated value without tracking the updated user state. This belief-update failure persists across context lengths and expression explicitness levels, identifying state-tracking capability as the primary bottleneck for long-horizon personalization.

Process-Level Reward Models (PRMs) are essential for guiding complex reasoning in large language models, yet existing PRM benchmarks cover only general domains such as mathematics, failing to address medical reasoning -- which is uniquely characterized by safety criticality, knowledge intensity, and diverse error patterns. Without a reliable medical PRM evaluation framework, we cannot quantify models' error detection capabilities in clinical reasoning, leaving their safety in real-world healthcare applications unverified. We propose MedPRMBench, the first process-level reward model benchmark for the medical domain. Built through a three-phase pipeline based on Clinical Reasoning Blueprints (CRBs), MedPRMBench systematically generates high-quality evaluation data from seven medical QA sources, covering 14 fine-grained error types across three categories (Simplicity, Soundness, and Sensitivity) with the first 4-level severity grading system to quantify clinical impact. The benchmark comprises 6{,}500 questions with 13{,}000 reasoning chains and 113{,}910 step-level labels, plus 6{,}879 questions for training. Our medical PRM baseline achieves an 87.1\% overall PRMScore -- substantially surpassing all baselines -- and serves as a plug-and-play verifier that improves downstream medical QA accuracy by 3.2--6.7 percentage points. Systematic evaluation spanning proprietary frontier models, open-source reasoning models, and medical-specialized models reveals critical weaknesses in current models' medical reasoning error detection capabilities, providing clear directions for future PRM improvement.

Physical neural networks offer a transformative route to edge intelligence, providing superior inference speed and energy efficiency compared to conventional digital architectures. However, realizing scalable, end-to-end, fully analog recurrent neural networks for temporal information processing remains challenging due to the difficulty of faithfully mapping trained network models onto physical hardware. Here we present a fully analog resonant recurrent neural network (R$^2$NN) implemented via a metacircuit architecture composed of coupled electrical local resonators. A reformulated mechanical-electrical analogy establishes a direct mapping between the R$^2$NN model and metacircuit elements, enabling accurate physical implementation of trained neural network parameters. By integrating jointly trainable global resistive coupling and local resonances, which generate effective frequency-dependent negative resistances, the architecture shapes an impedance landscape that steers currents along frequency-selective pathways. This mechanism enables direct extraction of discriminative spectral features, facilitating real-time temporal classification of raw analog inputs while bypassing analog-to-digital conversion. We demonstrate the cross-domain versatility of this framework using integrated hardware for tactile perception, speech recognition, and condition monitoring. This work establishes a scalable, fully analog paradigm for intelligent temporal processing and paves the way for low-latency, resource-efficient physical neural hardware for edge intelligence.

Multimodal Large Language Models (MLLMs) have demonstrated exceptional capabilities in various perception and reasoning tasks. Despite this success, ensuring their reliability in practical deployment necessitates robust confidence estimation. Prior works have predominantly focused on text-only LLMs, often relying on computationally expensive self-consistency sampling. In this paper, we extend this to multimodal settings and conduct a comprehensive evaluation of MLLMs' response confidence estimation. Our analysis reveals a significant instinct-reflection misalignment: the model's implicit token-level support frequently diverges from its verbal self-assessment confidence. To address this misalignment, we propose a monotone confidence fusion framework to merge dual-channel signals and cross-channel consistency to estimate correctness. Subsequently, an order-preserving mean alignment step is applied to correct global bias, which improves calibration while preserving the risk-coverage trade-off for selective prediction. Experiments on diverse open-source and closed-source MLLMs show that our method consistently yields more reliable confidence estimates and improves both calibration and failure prediction. Code will be available at https://github.com/Yunkaidang/Instinct-vs.-Reflection.

The most important architectural problem in AI is not the size of the model but the absence of a layer that carries forward what the model has come to understand. Sessions end. Context windows fill. Memory APIs return flat facts that the model has to reinterpret from scratch on every read. The result is intelligence that is powerful per session and amnesiac across time. This position paper argues that the layer which fixes this, the continuity layer, is the most consequential piece of infrastructure the field has not yet built, and that the engineering work to build it has begun in public. The formal evaluation framework for the property described here is the ATANT benchmark (arXiv:2604.06710), published separately with evaluation results on a 250-story corpus; a companion paper (arXiv:2604.10981) positions this framework against existing memory, long-context, and agentic-memory benchmarks. The paper defines continuity as a system property with seven required characteristics, distinct from memory and from retrieval; describes a storage primitive (Decomposed Trace Convergence Memory) whose write-time decomposition and read-time reconstruction produce that property; maps the engineering architecture to the theological pattern of kenosis and the symbolic pattern of Alpha and Omega, and argues this mapping is structural rather than metaphorical; proposes a four-layer development arc from external SDK to hardware node to long-horizon human infrastructure; examines why the physics limits now constraining the model layer make the continuity layer newly consequential; and argues that the governance architecture (privacy implemented as physics rather than policy, founder-controlled class shares on non-negotiable architectural commitments) is inseparable from the product itself.

Node classification on text-attributed graphs (TAGs) is a fundamental task with broad applications in citation analysis, social networks, and recommendation systems. Current GNN-based approaches suffer from shallow text encoding and heavy dependence on labeled data, limiting their effectiveness in label-scarce settings. While large language models (LLMs) naturally address the text understanding gap with deep semantic reasoning, existing LLM-for-graph methods either still require abundant labels during training or fail to exploit the rich structural signals freely available in graph topology. Our key observation is that, in many real-world TAGs, edges predominantly connect similar nodes under the homophily principle, meaning graph topology inherently encodes class structure without any labels. Building on this insight, we reformulate node classification as a link prediction task and present HopRank, a fully self-supervised LLM-tuning framework for TAGs. HopRank constructs preference data via hierarchical hop-based sampling and employs adaptive preference learning to prioritize informative training signals without any class labels. At inference, nodes are classified by predicting their connection preferences to labeled anchors, with an adaptive early-exit voting scheme to improve efficiency. Experiments on three TAG benchmarks show that HopRank matches fully-supervised GNNs and substantially outperforms prior graph-LLM methods, despite using zero labeled training data.

Users increasingly rely on consumer-facing generative AI (GenAI) for tasks ranging from everyday needs to sensitive use cases. Yet, it remains unclear whether and how existing security and privacy (S&P) communications in GenAI tools shape users' adoption decisions and subsequent experiences. Understanding how users seek, interpret, and evaluate S&P information is critical for designing usable transparency that users can trust and act on. We conducted semi-structured interviews and design sessions with 21 U.S. GenAI users. We find that available S&P information rarely drove initial adoption in practice, as participants often perceived it as incomplete, ineffective, or lacking credibility. Instead, they relied on rough proxies, such as popularity, to infer S&P practices. After adoption, uncertainty about S&P practices constrained participants' willingness to use GenAI tools, particularly in high-stakes contexts, and, in some cases, contributed to discontinued use. Participants therefore called for transparency that supports decision-making and use, including trustworthy information (e.g., independent evaluations) and usable interfaces (e.g., on-demand disclosure). We synthesize participants' desired design practices into five dimensions to facilitate systematic future investigation into best practices. We conclude with recommendations for researchers, designers, and policymakers to improve S&P transparency in consumer-facing GenAI.

AI-generated imagery has reached near-photorealistic fidelity, yet this technology poses significant threats to information security and societal trust. Existing deepfake detection methods often exhibit limited robustness in open-world scenarios. To address this limitation, this paper investigates intrinsic discrepancies between synthetic and authentic images from a signal-level perspective. Our analysis reveals that low-correlation signals serve as distinctive markers for differentiating AI-generated imagery from real photographs. Building on this insight, we introduce a novel method for quantifying these signals based on fractal theory. By analyzing the fractal characteristics of low-correlation signals, our method effectively captures the subtle statistical anomalies inherent to the synthesis process. Extensive experimental results demonstrate the method's robustness and superior detection performance. This work emphasizes the need to shift research focus to a new signal-level direction for deepfake detection. Theoretically, this proposed approach is not limited to face image identification but can be applied to all AI-generated image detection tasks. This study provides a new research direction for deepfake detection.

Large Language Models can generate synthetic survey responses at low cost, but their accuracy varies unpredictably across questions. We study the design problem of allocating a fixed budget of human respondents across estimation tasks when cheap LLM predictions are available for every task. Our framework combines three components. First, building on Prediction-Powered Inference, we characterize a question-specific rectification difficulty that governs how quickly the estimator's variance decreases with human sample size. Second, we derive a closed-form optimal allocation rule that directs more human labels to tasks where the LLM is least reliable. Third, since rectification difficulty depends on unobserved human responses for new surveys, we propose a meta-learning approach, trained on historical data, that predicts it for entirely new tasks without pilot data. The framework extends to general M-estimation, covering regression coefficients and multinomial logit partworths for conjoint analysis. We validate the framework on two datasets spanning different domains, question types, and LLMs, showing that our approach captures 61-79% of the theoretically attainable efficiency gains, achieving 11.4% and 10.5% MSE reductions without requiring any pilot human data for the target survey.

Evaluating meeting effectiveness is crucial for improving organizational productivity. Current approaches rely on post-hoc surveys that yield a single coarse-grained score for an entire meeting. The reliance on manual assessment is inherently limited in scalability, cost, and reproducibility. Moreover, a single score fails to capture the dynamic nature of collaborative discussions. We propose a new paradigm for evaluating meeting effectiveness centered on novel criteria and temporal fine-grained approach. We define effectiveness as the rate of objective achievement over time and assess it for individual topical segments within a meeting. To support this task, we introduce the AMI Meeting Effectiveness (AMI-ME) dataset, a new meta-evaluation dataset containing 2,459 human-annotated segments from 130 AMI Corpus meetings. We also develop an automatic effectiveness evaluation framework that uses a Large Language Model (LLM) as a judge to score each segment's effectiveness relative to the overall meeting objectives. Through substantial experiments, we establish a comprehensive benchmark for this new task and evaluate the framework's generalizability across distinct meeting types, ranging from business scenarios to unstructured discussions. Furthermore, we benchmark end-to-end performance starting from raw speech to measure the capabilities of a complete system. Our results validate the framework's effectiveness and provide strong baselines to facilitate future research in meeting analysis and multi-party dialogue. Our dataset and code will be publicly available. The AMI-ME dataset and the Automatic Evaluation Framework are available at: this URL.

User behavior in the real world is diverse, cross-domain, and spans long time horizons. Existing user modeling benchmarks however remain narrow, focusing mainly on short sessions and next-item prediction within a single domain. Such limitations hinder progress toward robust and generalizable user models. We present HORIZON, a new benchmark that reformulates user modeling along three axes i.e. dataset, task, and evaluation. Built from a large-scale, cross-domain reformulation of Amazon Reviews, HORIZON covers 54M users and 35M items, enabling both pretraining and realistic evaluation of models in heterogeneous environments. Unlike prior benchmarks, it challenges models to generalize across domains, users, and time, moving beyond standard missing-positive prediction in the same domain. We propose new tasks and evaluation setups that better reflect real-world deployment scenarios. These include temporal generalization, sequence-length variation, and modeling unseen users, with metrics designed to assess general user behavior understanding rather than isolated next-item prediction. We benchmark popular sequential recommendation architectures alongside LLM-based baselines that leverage long-term interaction histories. Our results highlight the gap between current methods and the demands of real-world user modeling, while establishing HORIZON as a foundation for research on temporally robust, cross-domain, and general-purpose user models.

Recent text embedding models are often adapted to specialized domains via contrastive pre-finetuning (PFT) on a naive collection of scattered, heterogeneous tasks. However, this approach often introduces task-induced bias alongside domain knowledge, leading to uncontrolled representation shifts that distort the pretrained embedding geometry and cause substantial performance degradation. To address this issue, we propose REZE}, a representation regularization framework that explicitly controls representation shift during embedding pre-finetuning. REZE operates on the relations of anchor-positive pairs and decomposes them in an eigenspace. It then measures task-wise dispersion along each eigencomponent to identify task-variant directions and applies adaptive soft-shrinkage to suppress task-induced noise while preserving task-invariant semantic structure, without inference-time overhead. Experiments across multiple embedding backbones and specialized benchmarks show that REZE outperforms standard pre-finetuning and isotropy-oriented post-hoc regularization in most settings, remaining stable where existing PFT variants collapse. Embedding space analyses further confirm that REZE induces controlled shifts aligned with the original embedding manifold, underscoring representation shift control as a key principle for robust embedding pre-finetuning under heterogeneous supervision.

Assessing the security posture of modern computing systems typically requires the use of multiple specialized tools. These tools focus on different aspects such as configuration compliance, file integrity, and vulnerability exposure, and their outputs are often difficult to interpret collectively. This paper introduces the Unified Compliance Aggregator (UCA), a framework that integrates several open-source security tools into a single composite score representing overall system security. The proposed framework combines outputs from Lynis, OpenSCAP (STIG and CIS profiles), AIDE, Tripwire, and Nmap NSE. A normalization process converts heterogeneous outputs into a consistent 0 to 100 scale, followed by weighted aggregation. We also introduce a logarithmic scoring model for file integrity measurements to address limitations observed in prior linear approaches. Experiments were conducted on Ubuntu 22.04 across different hardening levels and environments. Results show consistent improvement in composite scores as systems are hardened, while also revealing contrasting behavior between compliance and file integrity tools. Two case studies, a basic web server and a DVWA-based system illustrate how the framework can be applied in practical scenarios.

Accurate comprehension and controllable generation of emotion and rhetoric are pivotal for enhancing the reasoning capabilities of large language models (LLMs). Existing studies mostly rely on external optimizations, lacking in-depth exploration of internal representation mechanisms, thus failing to achieve fine-grained steering at the neuron level. A handful of works on neurons are confined to emotions, neglecting rhetoric neurons and their intrinsic connections. Traditional neuron masking also exhibits counterintuitive phenomena, making reliable verification of neuron functionality infeasible. To address these issues, we systematically investigate the neurons representation mechanisms and inherent associations of 6 emotion categories and 4 core rhetorical devices. We propose a neuron identification framework that integrates multi-dimensional screening, and design an adaptive masking method incorporating dynamic filtering, attenuation masking, and feedback optimization, enabling reliable causal validation of neuron functionality.Through neuron regulation, we achieve directed induction of non-target sentences and enhancement of emotion tasks via rhetoric neurons. Experiments on 5 commonly used datasets validate the effectiveness of our method, providing a novel paradigm for the fine-grained steering of emotion and rhetoric expressions in LLMs.

Recent advancements in large language models (LLMs) have enabled agents to tackle complex embodied tasks through environmental interaction. However, these agents still make suboptimal decisions and perform ineffective actions, as they often overlook critical environmental feedback that differs from their internal beliefs. Through a formal probing analysis, we characterize this as belief inertia, a phenomenon where agents stubbornly adhere to prior beliefs despite explicit observations. To address this, we advocate active belief intervention, moving from passive understanding to active management. We introduce the Estimate-Verify-Update (EVU) mechanism, which empowers agents to predict expected outcomes, verify them against observations through explicit reasoning, and actively update prior beliefs based on the verification evidence. EVU is designed as a unified intervention mechanism that generates textual belief states explicitly, and can be integrated into both prompting-based and training-based agent reasoning methods. Extensive experiments across three embodied benchmarks demonstrate that EVU consistently yields substantial gains in task success rates. Further analyses validate that our approach effectively mitigates belief inertia, advancing the development of more robust embodied agents. Our code is available at https://github.com/WangHanLinHenry/EVU.

Rowhammer on GPU DRAM has enabled adversarial bit flips in model weights; shared KV-cache blocks in LLM serving systems present an analogous but previously unexamined target. In vLLM's Prefix Caching, these blocks exist as a single physical copy without integrity protection. Using software fault injection under ideal bit targeting, we characterize worst-case severity and identify three properties: (1) Silent divergence - 13 of 16 BF16 bit positions produce coherent but altered outputs, indistinguishable from legitimate responses without a clean baseline. (2) Selective propagation - only requests sharing the targeted prefix are affected. (3) Persistent accumulation - no temporal decay occurs, so cumulative damage grows linearly with subsequent requests. Together, these constitute a threat profile distinct from weight corruption: silent divergence and selective propagation enable detection evasion; persistent accumulation then proceeds unchecked, yielding damage amplification bounded only by how long the block remains cached. A checksum-based countermeasure detects any single-bit corruption at scheduling time, bounding cumulative damage to one batch independent of the block's cache lifetime, with negligible overhead. These results argue for integrity protection of prefix blocks before end-to-end exploitation is demonstrated.

Large Audio-Language Models (LALMs) are increasingly integrated into daily applications, yet their generative biases remain underexplored. Existing speech fairness benchmarks rely on synthetic speech and Multiple-Choice Questions (MCQs), both offering a fragmented view of fairness. We propose VIBE, a framework that evaluates generative bias through open-ended tasks such as personalized recommendations, using real-world human recordings. Unlike MCQs, our method allows stereotypical associations to manifest organically without predefined options, making it easily extensible to new tasks. Evaluating 11 state-of-the-art LALMs reveals systematic biases in realistic scenarios. We find that gender cues often trigger larger distributional shifts than accent cues, indicating that current LALMs reproduce social stereotypes.

Despite the rapid progress, LLMs for sequential decision-making (i.e., LLM agents) still struggle to produce diverse outputs. This leads to insufficient exploration, convergence to sub-optimal solutions, and becoming stuck in loops. Such limitations can be problematic in environments that require active exploration to gather information and make decisions. Sampling methods such as temperature scaling introduce token-level randomness but fail to produce enough diversity at the sequence level. We analyze LLM exploration in the classic Multi-Armed Bandit (MAB) setting and the Text Adventure Learning Environment Suite (TALES). We find that current decoding strategies and prompting methods like Chain-of-Thought and Tree-of-Thought are insufficient for robust exploration. To address this, we introduce DORA Explorer (Diversity-Oriented Ranking of Actions), a training-free framework for improving exploration in LLM agents. DORA generates diverse action candidates, scores them using token log-probabilities, and selects actions using a tunable exploration parameter. DORA achieves UCB-competitive performance on MAB and consistent gains across TALES, e.g., improving Qwen2.5-7B's performance from 29.2% to 45.5% in TextWorld. Our project is available at: https://dora-explore.github.io/.

Enterprise AI systems increasingly deploy multiple intelligent agents across mission-critical workflows that must satisfy hard policy constraints, bounded risk exposure, and comprehensive auditability (SOX, HIPAA, GDPR). Existing coordination methods - cooperative MARL, consensus protocols, and centralized planners - optimize expected reward while treating constraints implicitly. This paper introduces CAMCO (Constraint-Aware Multi-Agent Cognitive Orchestration), a runtime coordination layer that models multi-agent decision-making as a constrained optimization problem. CAMCO integrates three mechanisms: (i) a constraint projection engine enforcing policy-feasible actions via convex projection, (ii) adaptive risk-weighted Lagrangian utility shaping, and (iii) an iterative negotiation protocol with provably bounded convergence. Unlike training-time constrained RL, CAMCO operates as deployment-time middleware compatible with any agent architecture, with policy predicates designed for direct integration with production engines such as OPA. Evaluation across three enterprise scenarios - including comparison against a constrained Lagrangian MARL baseline - demonstrates zero policy violations, risk exposure below threshold (mean ratio 0.71), 92-97% utility retention, and mean convergence in 2.4 iterations.

Decoding-free reranking methods that read relevance signals directly from LLM attention weights offer significant latency advantages over autoregressive approaches, yet suffer from attention score homogenization: middle-context documents receive near-identical scores, destroying the fine-grained distinctions required for ranking. We propose HeadRank, a framework that lifts preference optimization from discrete token space into the continuous attention domain through entropy-regularized head selection, hard adjacent-level preference pairs, and a distribution regularizer that jointly sharpen discriminability in the homogenized middle zone. Depth truncation at the deepest selected layer further reduces inference to $\mathcal{O}(1)$ forward passes. Across 14 benchmarks on three Qwen3 scales (0.6B--4B) using only 211 training queries, HeadRank consistently outperforms generative and decoding-free baselines with 100\% formatting success. At 4B, 57.4\% of relevant middle-zone documents reach the top quartile versus 14.2\% for irrelevant ones -- a 43-percentage-point selectivity gap that demonstrates the effectiveness of attention-space preference alignment for listwise reranking.

The rapid expansion of the model context protocol (MCP) ecosystem enables large language model (LLM)-based agents to access a wide range of external tools via a standardized interface. However, identifying appropriate MCP servers for a specific development task remains challenging. Existing studies primarily focus on measuring the MCP ecosystem or optimizing tool invocation mechanisms, while systematic recommendation frameworks and reproducible benchmarks for real-world development tasks remain largely unexplored. To address this limitation, we formulate task-oriented MCP server recommendation as a structured retrieval-and-ranking problem that jointly considers semantic relevance and engineering constraints. We first construct Task2MCP, a task-centered dataset that systematically associates taxonomy-grounded development tasks with curated MCP servers. This dataset provides structured supervision and a reproducible evaluation environment for research on MCP tool recommendations. Building on this dataset, we propose T2MRec, a task-to-MCP server recommendation model. It models semantic relevance and structural compatibility to construct an initial candidate set. Then it improves coverage and ranking quality through centroid-based candidate expansion and constrained LLM-based re-ranking. In addition, we design and implement an interactive MCP server recommendation agent prototype that operates in conversational environments to support dynamic decision-making. The agent assists developers in efficiently evaluating and integrating tools by providing recommended MCP servers together with usage guidelines.

Personalized image aesthetics assessment (PIAA) aims to predict an individual user's subjective rating of an image, which requires modeling user-specific aesthetic preferences. Existing methods rely on historical user ratings for this modeling and therefore struggle when such data are unavailable. We address this zero-shot setting by using user profiles as contextual signals for personalization and adopting a profile-based personalization paradigm. We introduce P-MLLM, a profile-aware multimodal LLM that augments a frozen LLM with selective fusion modules for controlled visual integration. These modules selectively integrate visual information into the model's evolving hidden states during profile-conditioned reasoning, allowing visual information to be incorporated in a profile-aware manner. Experiments on recent PIAA benchmarks show that P-MLLM achieves competitive zero-shot performance and remains effective even with coarse profile information, highlighting the potential of profile-based personalization for zero-shot PIAA.

Project Yanasse presents a method for discovering new proofs of theorems in one area of mathematics by transferring proof strategy patterns (e.g., Lean 4 tactic invocation patterns) from a structurally distant area. The system extracts tactic usage distributions across 27 top-level areas of Mathlib (217,133 proof states), computes z-scores to identify tactics that are heavily used in a source area but rare or absent in a target area, matches source and target proof states via GPU-accelerated NP-hard analogy (running on a MacBook Air via Apple's MPS backend), and then asks an AI reasoning agent to semantically adapt--not symbol-substitute--the source tactics invocation pattern to the target theorem. In this first part of the study, the method is applied to the pair Probability -> Representation Theory, producing 4 Lean-verified new proofs out of 10 attempts (40%). The proofs compile with zero sorry declarations. The key finding is that tactic schemas decompose into a head (domain-gated, rarely transfers) and a modifier (domain-general, often transfers): filter upwards's head fails in representation theory (no Filter structure), but its [LIST] with ω modifier transfers cleanly as ext1 + simp [LIST] + rfl. Crucially, the underlying matching engine--deep vision lib.py--is entirely domain independent: the same optimization code for an NP-hard matching that matches chess positions by analogy matches Lean proof states by analogy, without knowing which domain it is processing. Only a relation extractor is domain-specific.

Conditional depth execution routes a subset of tokens through a lightweight cheap FFN while the remainder execute the standard full FFN at each controlled layer. The central difficulty is gate training: the gate decision must propagate through many layers before it influences the language modeling (LM) loss, so the resulting gradients are weak and noisy. Auxiliary losses are commonly stacked to stabilise training, yet the interactions among them -- particularly between a predictive auxiliary and explicit score supervision -- have not been systematically compared under controlled conditions. We evaluate two gate designs under a 157.5M-parameter decoder-only model with controller-only training, 50% full-path budget, and 3-seed runs on a fineweb-edu subset. The MLP gate (G1) maps the current hidden state to a utility score; the JEPA-guided gate (G3) adds an action-conditional predictor that forecasts, in a low-dimensional latent space, the outcome of executing full vs. cheap per token, aligned against a fixed target head. Under the standard recipe with oracle-style utility regression and pairwise rank supervision (util/rank), G3 improves early-to-mid optimisation over G1 in 3/3 seeds (lower avg LM, faster threshold hits, ~10.3x lower grad norms), with 20k-step endpoint LM within a 0.005 heuristic reference. A key finding (ablation A3): jointly removing util/rank improves best/avg LM and threshold-hit speed in 3/3 seeds for both gates, and the early-to-mid advantage of G3 over G1 disappears. We trace this to an off-policy oracle label that assumes all subsequent layers execute full, whereas gated execution routes only a fraction through full -- making util/rank net-negative under the current recipe. Removing util/rank also cuts the training FLOPs proxy from ~1.53x to ~1.07x full-only (2.87h to 1.75h on a V100-32GB, ~39%). Conclusions are scoped to the studied regime.

We present a novel approach for claim verification from tabular data documents. Recent LLM-based approaches either employ complex pretraining/fine-tuning or decompose verification into subtasks, often lacking comprehensive explanations and generalizability. To address these limitations, we propose a Multi-Agentic framework for Claim verification (MACE) consisting of three specialized agents: Planner, Executor, and Verifier. Instead of elaborate finetuning, each agent employs a zero-shot Chain-of-Thought setup to perform its tasks. MACE produces interpretable verification traces, with the Planner generating explicit reasoning strategies, the Executor providing detailed computation steps, and the Verifier validating the logic. Experiments demonstrate that MACE achieves state-of-the-art (SOTA) performance on two datasets and performs on par with the best models on two others, while achieving 80--100\% of best performance with substantially smaller models: 27--92B parameters versus 235B. This combination of competitive performance, memory efficiency, and transparent reasoning highlights our framework's effectiveness.

Recursive architectures such as Tiny Recursive Models (TRMs) perform implicit reasoning through iterative latent computation, yet the geometric structure of these reasoning trajectories remains poorly understood. We investigate the activation manifold of TRMs during recursive unrolling and find that activations occupy an effectively linear, low-dimensional subspace whose principal directions can be tracked dynamically with cheap power iterations. This suggests that weight-sharing concentrates iterative computation along a small number of dominant eigendirections, and we find that this concentration varies sharply across computational sites. We exploit this structure through LASER (Low-Rank Activation SVD for Efficient Recursion), a dynamic compression framework that maintains an evolving low-rank basis via matrix-free subspace tracking with a fidelity-triggered reset mechanism, achieving ${\sim}60\%$ activation memory savings with no statistically significant accuracy degradation. Our analysis raises questions about how recursive architectures allocate representational capacity during implicit reasoning, and whether this concentration can be exploited to improve the efficiency and stability of latent computation.

Breast cancer diagnosis demands rapid and precise tools, yet traditional histopathological methods often fall short in intra-operative settings. Deep Ultraviolet (DUV) fluorescence imaging emerges as a transformative approach, offering high-contrast, label-free visualization of whole-slide images (WSIs) with unprecedented detail, surpassing conventional hematoxylin and eosin (H&E) staining in speed and resolution. However, existing deep learning methods for breast cancer classification, predominantly patch-based, fragment spatial context and incur significant preprocessing overhead, limiting their clinical utility. Moreover, standard attention mechanisms, such as Spatial, Squeeze-and-Excitation, Global Context and Guided Context Gating, fail to fully exploit the rich, multi-scale regional relationships inherent in DUV-WSI data, often prioritizing generic feature recalibration over diagnostic specificity. This study introduces a novel Region-Affinity Attention mechanism tailored for DUV-WSI breast cancer classification, processing entire slides without patching to preserve spatial integrity. By modeling local neighbor distances and constructing a full affinity matrix, our method dynamically highlights diagnostically relevant regions, augmented by a contrastive loss to enhance feature discriminability. Evaluated on a dataset of 136 DUV-WSI samples, our approach achieves an accuracy of 92.67 +/- 0.73% and an AUC of 95.97%, outperforming existing attention methods.

Selective State Space Models (SSMs), notably Mamba, employ diagonal state transitions that limit both memory retention and bilinear computational capacity. We propose a factorized bilinear input modulation that augments the SSM with a state-input product, interpretable as a finite-dimensional Koopman bilinear form. After introducing a shared state across channels (Coupled SSM), the modulation admits two implementations. Coupled Bilinear Input Modulation (Coupled-BIM) retains the full bilinear product at the cost of sequential computation, while Coupled Gated Modulation (Coupled-GM) linearizes it into a gate modulation that is compatible with the parallel scan. Experiments on a multiple input-delay pendulum (memory retention) and NARMA-10 (bilinear computation) reveal a clear dissociation. Coupled-GM substantially improves memory retention but not bilinear computation, while Coupled-BIM improves both. A pathway ablation confirms that the two downstream routes of the bilinear signal serve complementary roles. The improvement is statistically robust, with Coupled-BIM consistently outperforming all other variants on bilinear computation. Furthermore, only Coupled-BIM benefits from increasing the SSM state dimension, while coupling or gate modulation alone show no improvement, establishing the bilin-ear mechanism as uniquely capable of exploiting larger state spaces.

Modeling coordination among generative agents in complex multi-round decision-making presents a core challenge for AI and operations management. Although behavioral experiments have revealed cognitive biases behind supply chain inefficiencies, traditional methods face scalability and control limitations. We introduce a scalable experimental paradigm using Large Language Models (LLMs) to simulate multi-stage supply chain dynamics. Grounded in a Hierarchical Reasoning Framework, this study specifically analyzes the impact of cognitive heterogeneity on agent interactions. Unlike prior homogeneous settings, we employ DeepSeek and GPT agents to systematically vary reasoning sophistication across supply chain tiers. Through rigorously replicated and statistically validated simulations, we investigate how this cognitive diversity influences collective outcomes. Results indicate that agents exhibit myopic and self-interested behaviors that exacerbate systemic inefficiencies. However, we demonstrate that information sharing effectively mitigates these adverse effects. Our findings extend traditional behavioral methods and offer new insights into the dynamics of AI-enabled organizations. This work underscores both the potential and limitations of LLM-based agents as proxies for human decision-making in complex operational environments.

We derive explicit non-asymptotic PAC-Bayes generalization bounds for Gibbs posteriors, that is, data-dependent distributions over model parameters obtained by exponentially tilting a prior with the empirical risk. Unlike classical worst-case complexity bounds based on uniform laws of large numbers, which require explicit control of the model space in terms of metric entropy (integrals), our analysis yields posterior-averaged risk bounds that can be applied to overparameterized models and adapt to the data structure and the intrinsic model complexity. The bound involves a marginal-type integral over the parameter space, which we analyze using tools from singular learning theory to obtain explicit and practically meaningful characterizations of the posterior risk. Applications to low-rank matrix completion and ReLU neural network regression and classification show that the resulting bounds are analytically tractable and substantially tighter than classical complexity-based bounds. Our results highlight the potential of PAC-Bayes analysis for precise finite-sample generalization guarantees in modern overparameterized and singular models.

Vision-Language Models (VLMs) achieve strong cross-modal performance, yet recent evidence suggests they over-rely on textual descriptions while under-utilizing visual evidence -- a phenomenon termed ``text shortcut learning.'' We propose an adversarial evaluation framework that quantifies this cross-modal dependency by measuring accuracy degradation (Drop) when semantically conflicting text is paired with unchanged images. Four adversarial strategies -- shape\_swap, color\_swap, position\_swap, and random\_text -- are applied to a controlled geometric-shapes dataset ($n{=}1{,}000$). We compare three configurations: Baseline CLIP (ViT-B/32), LoRA fine-tuning, and LoRA Optimized (integrating Hard Negative Mining, Label Smoothing, layer-wise learning rates, Cosine Restarts, curriculum learning, and data augmentation). The optimized model reduces average Drop from 27.5\% to 9.8\% (64.4\% relative improvement, $p{<}0.001$) while maintaining 97\% normal accuracy. Attention visualization and embedding-space analysis confirm that the optimized model attends more to visual features and achieves tighter cross-modal alignment.

Large language models require continuous adaptation to new tasks while preserving safety alignment. However, fine-tuning on even benign data often compromises safety behaviors, including refusal of harmful requests, truthfulness, and commonsense reasoning. We investigate which training samples cause alignment drift through a data-centric lens. Our empirical analysis shows samples contribute unequally: high-gradient samples cause greater safety degradation and drive models toward pretrained distributions, while moderate-gradient samples enable task learning with minimal alignment loss. We propose gradient-based sample selection that filters high-gradient samples during fine-tuning. Across multiple model families on continual domain tasks, our method substantially improves alignment preservation while maintaining competitive task performance, without requiring curated safe data or architectural modifications. Our method is robust across selection ratios, task orderings, and diverse attack benchmarks.

Extracting clinically relevant information from unstructured medical narratives such as admission notes, discharge summaries, and emergency case histories remains a challenge in clinical natural language processing (NLP). Medical Entity Recognition (MER) identifies meaningful concepts embedded in these records. Recent advancements in large language models (LLMs) have shown competitive MER performance; however, evaluations often focus on general entity types, offering limited utility for real-world clinical needs requiring finer-grained extraction. To address this gap, we rigorously evaluated the open-source LLaMA3 model for fine-grained medical entity recognition across 18 clinically detailed categories. To optimize performance, we employed three learning paradigms: zero-shot, few-shot, and fine-tuning with Low-Rank Adaptation (LoRA). To further enhance few-shot learning, we introduced two example selection methods based on token- and sentence-level embedding similarity, utilizing a pre-trained BioBERT model. Unlike prior work assessing zero-shot and few-shot performance on proprietary models (e.g., GPT-4) or fine-tuning different architectures, we ensured methodological consistency by applying all strategies to a unified LLaMA3 backbone, enabling fair comparison across learning settings. Our results showed that fine-tuned LLaMA3 surpasses zero-shot and few-shot approaches by 63.11% and 35.63%, respectivel respectively, achieving an F1 score of 81.24% in granular medical entity extraction.

Fine-tuning well-aligned large language models (LLMs) on new domains often degrades their safety alignment, even when using benign datasets. Existing safety alignment techniques primarily focus on pretraining, leaving fine-tuned models vulnerable to behavioral shifts. In this work, we introduce safety token regularization (STR), a lightweight method designed to preserve safety properties during fine-tuning. Our approach identifies salient tokens from rejection templates of well-aligned models and constrains their associated logits during training, preventing the loss of critical safety behaviors. Unlike reinforcement learning or preference optimization methods, STR requires minimal additional computation and seamlessly integrates with parameter-efficient fine-tuning techniques such as LoRA. Comprehensive experiments demonstrate that our approach achieves safety performance on par with state-of-the-art methods, while preserving task-specific utility and requiring minimal implementation overhead. Furthermore, we show that safety token regularization enhances training stability and overall performance beyond safety considerations alone. This work offers a practical and readily deployable strategy for continual safety alignment in fine-tuned LLMs.

Automating medical reports for retinal images requires a sophisticated blend of visual pattern recognition and deep clinical knowledge. Current Large Vision-Language Models (LVLMs) often struggle in specialized medical fields where data is scarce, leading to models that overfit and miss subtle but critical pathologies. To address this, we introduce DREAM (Dynamic Retinal Enhancement with Adaptive Multi-modal Fusion), a novel framework for high-fidelity medical report generation that excels even with limited data. DREAM employs a unique two-stage fusion mechanism that intelligently integrates visual data with clinical keywords curated by ophthalmologists. First, the Abstractor module maps image and keyword features into a shared space, enhancing visual data with pathology-relevant insights. Next, the Adaptor performs adaptive multi-modal fusion, dynamically weighting the importance of each modality using learnable parameters to create a unified representation. To ensure the model's outputs are semantically grounded in clinical reality, a Contrastive Alignment module aligns these fused representations with ground-truth medical reports during training. By combining medical expertise with an efficient fusion strategy, DREAM sets a new state-of-the-art on the DeepEyeNet benchmark, achieving a BLEU-4 score of 0.241, and further demonstrates strong generalization to the ROCO dataset.

Digital subtraction angiography (DSA) in coronary imaging is fundamentally challenged by physiological motion, forcing reliance on raw angiograms cluttered with anatomical noise. Existing deep learning methods often produced images with two critical clinically unacceptable flaws: persistent boundary artifacts and a loss of native tissue grayscale fidelity that undermined diagnostic confidence. We propose a novel framework termed as CDSA-Net that for the first time explicitly decouples and jointly optimizes vascular structure preservation and realistic background restoration. CDSA-Net introduces two core innovations: (i) A hierarchical geometric prior guidance (HGPG) mechanism, embedded in our coronary structure extraction network (CSENet). It synergistically combines integrated geometric prior (IGP) with gated spatial modulation (GSM) and centerline-aware topology (CAT) loss supervision, ensuring structural continuity. (ii) An adaptive noise module (ANM) within our coronary background restoration network (CBResNet). Unlike standard restoration, ANM uniquely models the stochastic nature of clinical X-ray noise, bridging the domain gap to enable seamless background intensity estimation and the complete elimination of boundary artifacts. The final subtraction is obtained by removing the restored background from the raw angiogram. Quantitatively, it significantly outperformed state-of-the-art methods in vascular intensity correlation and perceptual quality. A 25.6% improvement in morphology assessment efficiency and a 42.9% gain in hemodynamic evaluation speed set a new benchmark for utility in interventional cardiology, while maintaining diagnostic results consistent with raw angiograms. The project code is available at https://github.com/DrThink-ai/CDSA-Net.

Iterative alignment methods based on purely greedy updates are remarkably effective in practice, yet existing theoretical guarantees of \(O(\log T)\) KL-regularized regret can seem pessimistic relative to their empirical performance. In this paper, we argue that this mismatch arises from the regret criterion itself: KL-regularized regret conflates the statistical cost of learning with the exploratory randomization induced by the softened training policy. To separate these effects, we study the traditional temperature-zero regret criterion, which evaluates only the top-ranked response at inference time. Under this decision-centric notion of performance, we prove that standard greedy online alignment methods, including online RLHF and online DPO, achieve constant \((O(1))\) cumulative regret. By isolating the cost of identifying the best response from the stochasticity induced by regularization, our results provide a sharper theoretical explanation for the practical superb efficiency of greedy alignment.

Recent advances in summary evaluation are based on model-based metrics to assess quality dimensions, such as completeness, conciseness, and faithfulness. However, these methods often require large language models, and predicted scores are frequently miscalibrated, limiting their reliability. Moreover, evaluating the average quality across different summaries for a single document typically requires access to multiple reference summaries. Here, we propose a general framework that generates individual and average proxy scores without relying on reference summaries, human annotations, or expensive model-based metrics. We also propose group isotonic regression binning (GIRB), a calibration method that adjusts the raw predictions to better align with ground-truth evaluation metrics. While we focus on continuous-value scenarios, such as summarization, the method is applicable to discrete-value tasks, such as question answering. Experiments on seven datasets demonstrate that our approach consistently outperforms existing baselines.

Recent advances in summarization research focus on improving summary quality across multiple criteria, such as completeness, conciseness, and faithfulness, by jointly optimizing these dimensions. However, these efforts largely overlook the challenge of controlling summary generation with respect to individual criteria, especially in the presence of their inherent trade-offs. For example, enhancing conciseness can compromise completeness, and vice versa. In this work, we address this gap by proposing a loss function that aligns model outputs with fine-grained, model-based evaluation scores (e.g., from FineSurE), enabling both improvement in summary quality and dimension-specific control. Our approach improves the overall quality of summaries while maintaining the ability to selectively prioritize one criterion over others. Experiments on three pretrained models (LLaMA, Qwen, and Mistral) demonstrate that our method achieves performance comparable to state-of-the-art summarizers, while uniquely offering strong controllability over individual quality dimensions.

Converting betting odds into accurate outcome probabilities is a fundamental challenge in order to use betting odds as a benchmark for sports forecasting and market efficiency analysis. In this study, we propose two methods to overcome the limitations of existing conversion methods. Firstly, we propose an odds-only method to convert betting odds to probabilities without using historical data for model fitting. While existing odds-only methods, such as Multiplicative, Shin, and Power exist, they do not adjust for biases or relationships we found in our betting odds dataset, which consists of 90014 football matches across five different bookmakers. To overcome these limitations, our proposed Odds-Only-Equal-Profitability-Confidence (OO-EPC) method aligns with the bookmakers' pricing objectives of having equal confidence in profitability for each outcome. We provide empirical evidence from our betting odds dataset that, for the majority of bookmakers, our proposed OO-EPC method outperforms the existing odds-only methods. Beyond controlled experiments, we applied the OO-EPC method under real-world uncertainty by using it for six iterations of an annual basketball outcome forecasting competition. Secondly, we propose a generalised linear model that utilises historical data for model fitting and then converts betting odds to probabilities. Existing generalised linear models attempt to capture relationships that the Efficient Market Hypothesis already captures. To overcome this shortcoming, our proposed Favourite-Longshot-Bias-Adjusted Generalised Linear Model (FL-GLM) fits just one parameter to capture the favourite-longshot bias, providing a more interpretable alternative. We provide empirical evidence from historical football matches where, for all bookmakers, our proposed FL-GLM outperforms the existing multinomial and logistic generalised linear models.

Multi-agent reinforcement learning (MARL) is crucial for AI systems that operate collaboratively in distributed and adversarial settings, particularly in multi-domain operations (MDO). A central challenge in cooperative MARL is determining how agents should coordinate: existing approaches must either hand-specify graph topology, rely on proximity-based heuristics, or learn structure entirely from environment interaction; all of which are brittle, semantically uninformed, or data-intensive. We investigate whether large language models (LLMs) can generate useful coordination graph priors for MARL by using minimal natural language descriptions of agent observations to infer latent coordination patterns. These priors are integrated into MARL algorithms via graph convolutional layers within a graph neural network (GNN)-based pipeline, and evaluated on four cooperative scenarios from the Multi-Agent Particle Environment (MPE) benchmark against baselines spanning the full spectrum of coordination modeling, from independent learners to state-of-the-art graph-based methods. We further ablate across five compact open-source LLMs to assess the sensitivity of prior quality to model choice. Our results provide the first quantitative evidence that LLM-derived graph priors can enhance coordination and adaptability in dynamic multi-agent environments, and demonstrate that models as small as 1.5B parameters are sufficient for effective prior generation.

Multi-role dialogue summarization requires modeling complex interactions among multiple speakers while preserving role-specific information and factual consistency. However, most existing methods optimize for automatic metrics such as ROUGE and BERTScore, which favor surface-level imitation of references rather than genuine gains in faithfulness or alignment with human preferences. We propose a novel framework that couples explicit cognitive-style reasoning with reward-based optimization for multi-role dialogue summarization. Our method first distills structured reasoning traces (e.g., step-by-step inferences and intermediate reflections) from a large teacher model and uses them as auxiliary supervision to initialize a reasoning-aware summarizer via staged supervised fine-tuning. It then applies GRPO with a dual-principle reward that blends metric-based signals with human-aligned criteria targeting key information coverage, implicit inference, factual faithfulness, and conciseness. Experiments on multilingual multi-role dialogue benchmarks show that our method matches strong baselines on ROUGE and BERTScore. Specifically, results on CSDS confirm the framework's stability in semantic consistency, while in-depth analysis on SAMSum demonstrates clear gains in factual faithfulness and model-based preference alignment. These findings underscore the value of reasoning-aware and preference-aware training for reliable dialogue summarization. Checkpoints and datasets are available at https://huggingface.co/collections/NebulaPixel/summorchestra-multirole-summary.

We evaluate five state-of-the-art open-weights coding language models -- Kimi-K2.5 (at Q3 and Q4 quantizations), GLM-5.1, Qwen3-Coder-480B, and DeepSeek-V3.2 -- on a single multi-file React Native application generation task on NVIDIA GH200 576 GB hardware. The task specifies authentication, per-user per-day counting, and web compatibility, and is evaluated on whether the generated project runs out-of-the-box and on feature-level correctness. We find that SWE-Bench rankings do not predict task performance: Kimi-K2.5 at aggressive 3-bit quantization (UD-Q3_K_XL, 480 GB) produces the most complete and specification-compliant output, outranking models with substantially higher SWE-Bench Pro scores. We document three novel deployment findings: (1) default temperature=0 in coding tools causes sampling hangs with reasoning-model architectures, (2) reasoning-model thinking traces can leak through integration tools' file-path parsers, and (3) web-platform adaptation of native-mobile APIs is a universal training-data gap across every model tested. We also map the hardware-tier structure of April 2026 open-weights coding models, identifying two architectural schools and showing that the efficiency school (10-15 B active parameters) delivers equivalent SWE-Bench results at roughly 1/7th the hardware cost of the scale school (32-40 B active parameters).

As Artificial Intelligence (AI) and Agentic AI become increasingly integrated across sectors such as education and healthcare, it is critical to ensure that Multi-Agent Education System (MAES) is explainable from the early stages of requirements engineering (RE) within the AI software development lifecycle. Explainability is essential to build trust, promote transparency, and enable effective human-AI collaboration. Although personas are well-established in human-computer interaction to represent users and capture their needs and behaviors, their role in RE for explainable MAES remains underexplored. This paper proposes a human-first, persona-driven, explainable MAES RE framework and demonstrates the framework through a MAES for clinical reasoning training. The framework integrates personas and user stories throughout the RE process to capture the needs, goals, and interactions of various stakeholders, including medical educators, medical students, AI patient agent, and clinical agents (physical exam agent, diagnostic agent, clinical intervention agent, supervisor agent, evaluation agent). The goals, underlying models, and knowledge base shape agent interactions and inform explainability requirements that guided the clinical reasoning training of medical students. A post-usage survey found that more than 78\% of medical students reported that MAES improved their clinical reasoning skills. These findings demonstrate that RE based on persona effectively connects technical requirements with non-technical medical students from a human-centered approach, ensuring that explainable MAES are trustworthy, interpretable, and aligned with authentic clinical scenarios from the early stages of the AI system engineering. The partial MAES for the clinical scenario simulator is~\href{https://github.com/2sigmaEdTech/MAS/}{open sourced here}.

Large Language Models (LLMs) show promise for automated code repair but often struggle with the complex semantic and structural correctness required. We present SynthFix, a hybrid neural-symbolic framework that improves LLM-based vulnerability repair by unifying code synthesis with compiler-informed symbolic feedback. The core of our approach is an adaptive training strategy where a neural Router Model directs code samples to either Supervised Fine-Tuning (SFT) to learn common patterns or Reward Fine-Tuning (RFT) with symbolic rewards for complex, iterative refinement. On the FixJS (JavaScript) and CodeFlaws (C) benchmarks, SynthFix achieves up to 18% relative improvement in CodeBLEU/CrystalBLEU and 32% in Exact Match over strong SFT and RFT baselines. Our results show that this adaptive combination of training strategies, which mirrors how developers alternate between pattern application and tool feedback, significantly improves the accuracy and efficiency of LLM-based vulnerability repair. Our code and data are available at https://github.com/CoderDoge1108/SynthFix.