The Inference Report

Large language models (LLMs) have achieved strong results in code generation, but their ability to generate GUI applications, especially games, remains insufficiently studied. Existing benchmarks mainly evaluate correctness through test cases, which are inadequate for GUI applications because these systems are interactive, event-driven, and require correct state transitions across sequences of user actions. Their evaluation therefore should consider interaction flows and UI logic rather than only pass/fail outcomes. To study this problem, we introduce PlayEval, a repository-aware benchmark built from 43 multilingual GUI applications in Python, TypeScript, and JavaScript. Unlike prior GUI benchmarks that are difficult to adapt to desktop environments, PlayEval covers six major GUI application categories and directly supports code-generation evaluation. We further propose Play@k, a metric that measures whether at least one of *k* generated candidates can be played end-to-end without logical errors. To support reliable evaluation, we develop PlayTester, an LLM-based agent that performs task-oriented GUI playthroughs and detects logic violations automatically. Experiments on 10 state-of-the-art code LLMs show that, despite high compilation rates, they achieve near-zero Play@3, revealing major weaknesses in generating logically correct GUI applications. To address this limitation, we present PlayCoder, a multi-agent, repository-aware framework that generates, evaluates, and iteratively repairs GUI application code in a closed loop. PlayCoder substantially improves both functional correctness and semantic alignment for open-source and closed-source models, reaching up to 38.1% Exec@3 and 20.3% Play@3. Case studies further show that it can uncover silent logic bugs missed by traditional metrics and fix them through targeted edits.

Training modern neural networks often relies on large learning rates, operating at the edge of stability, where the optimization dynamics exhibit oscillatory and chaotic behavior. Empirically, this regime often yields improved generalization performance, yet the underlying mechanism remains poorly understood. In this work, we represent stochastic optimizers as random dynamical systems, which often converge to a fractal attractor set (rather than a point) with a smaller intrinsic dimension. Building on this connection and inspired by Lyapunov dimension theory, we introduce a novel notion of dimension, coined the `sharpness dimension', and prove a generalization bound based on this dimension. Our results show that generalization in the chaotic regime depends on the complete Hessian spectrum and the structure of its partial determinants, highlighting a complexity that cannot be captured by the trace or spectral norm considered in prior work. Experiments across various MLPs and transformers validate our theory while also providing new insights into the recently observed phenomenon of grokking.

We establish central and non-central limit theorems for sequences of functionals of the Gaussian output of an infinitely-wide random neural network on the d-dimensional sphere . We show that the asymptotic behaviour of these functionals as the depth of the network increases depends crucially on the fixed points of the covariance function, resulting in three distinct limiting regimes: convergence to the same functional of a limiting Gaussian field, convergence to a Gaussian distribution, convergence to a distribution in the Qth Wiener chaos. Our proofs exploit tools that are now classical (Hermite expansions, Diagram Formula, Stein-Malliavin techniques), but also ideas which have never been used in similar contexts: in particular, the asymptotic behaviour is determined by the fixed-point structure of the iterative operator associated with the covariance, whose nature and stability governs the different limiting regimes.

Reinforcement learning (RL) offers a compelling data-driven paradigm for synthesizing controllers for complex systems when accurate physical models are unavailable; however, most existing control-oriented RL methods assume stationarity and, therefore, struggle in real-world non-stationary deployments where system dynamics and operating conditions can change unexpectedly. Moreover, RL controllers acting in physical environments must satisfy safety constraints throughout their learning and execution phases, rendering transient violations during adaptation unacceptable. Although continual RL and safe RL have each addressed non-stationarity and safety, respectively, their intersection remains comparatively unexplored, motivating the study of safe continual RL algorithms that can adapt over the system's lifetime while preserving safety. In this work, we systematically investigate safe continual reinforcement learning by introducing three benchmark environments that capture safety-critical continual adaptation and by evaluating representative approaches from safe RL, continual RL, and their combinations. Our empirical results reveal a fundamental tension between maintaining safety constraints and preventing catastrophic forgetting under non-stationary dynamics, with existing methods generally failing to achieve both objectives simultaneously. To address this shortcoming, we examine regularization-based strategies that partially mitigate this trade-off and characterize their benefits and limitations. Finally, we outline key open challenges and research directions toward developing safe, resilient learning-based controllers capable of sustained autonomous operation in changing environments.

Scaling humanoid foundation models is bottlenecked by the scarcity of robotic data. While massive egocentric human data offers a scalable alternative, bridging the cross-embodiment chasm remains a fundamental challenge due to kinematic mismatches. We introduce UniT (Unified Latent Action Tokenizer via Visual Anchoring), a framework that establishes a unified physical language for human-to-humanoid transfer. Grounded in the philosophy that heterogeneous kinematics share universal visual consequences, UniT employs a tri-branch cross-reconstruction mechanism: actions predict vision to anchor kinematics to physical outcomes, while vision reconstructs actions to filter out irrelevant visual confounders. Concurrently, a fusion branch synergies these purified modalities into a shared discrete latent space of embodiment-agnostic physical intents. We validate UniT across two paradigms: 1) Policy Learning (VLA-UniT): By predicting these unified tokens, it effectively leverages diverse human data to achieve state-of-the-art data efficiency and robust out-of-distribution (OOD) generalization on both humanoid simulation benchmark and real-world deployments, notably demonstrating zero-shot task transfer. 2) World Modeling (WM-UniT): By aligning cross-embodiment dynamics via unified tokens as conditions, it realizes direct human-to-humanoid action transfer. This alignment ensures that human data seamlessly translates into enhanced action controllability for humanoid video generation. Ultimately, by inducing a highly aligned cross-embodiment representation (empirically verified by t-SNE visualizations revealing the convergence of human and humanoid features into a shared manifold), UniT offers a scalable path to distill vast human knowledge into general-purpose humanoid capabilities.

Some of the most performant reinforcement learning algorithms today can be prohibitively expensive as they use test-time scaling methods such as sampling multiple action candidates and selecting the best one. In this work, we propose FASTER, a method for getting the benefits of sampling-based test-time scaling of diffusion-based policies without the computational cost by tracing the performance gain of action samples back to earlier in the denoising process. Our key insight is that we can model the denoising of multiple action candidates and selecting the best one as a Markov Decision Process (MDP) where the goal is to progressively filter action candidates before denoising is complete. With this MDP, we can learn a policy and value function in the denoising space that predicts the downstream value of action candidates in the denoising process and filters them while maximizing returns. The result is a method that is lightweight and can be plugged into existing generative RL algorithms. Across challenging long-horizon manipulation tasks in online and batch-online RL, FASTER consistently improves the underlying policies and achieves the best overall performance among the compared methods. Applied to a pretrained VLA, FASTER achieves the same performance while substantially reducing training and inference compute requirements. Code is available at https://github.com/alexanderswerdlow/faster .

Personalized Federated Learning (PFL) aims to learn multiple task-specific models rather than a single global model across heterogeneous data distributions. Existing PFL approaches typically rely on iterative optimization-such as model update trajectories-to cluster users that need to accomplish the same tasks together. However, these learning-dynamics-based methods are inherently vulnerable to low-quality data and noisy labels, as corrupted updates distort clustering decisions and degrade personalization performance. To tackle this, we propose FB-NLL, a feature-centric framework that decouples user clustering from iterative training dynamics. By exploiting the intrinsic heterogeneity of local feature spaces, FB-NLL characterizes each user through the spectral structure of the covariances of their feature representations and leverages subspace similarity to identify task-consistent user groupings. This geometry-aware clustering is label-agnostic and is performed in a one-shot manner prior to training, significantly reducing communication overhead and computational costs compared to iterative baselines. Complementing this, we introduce a feature-consistency-based detection and correction strategy to address noisy labels within clusters. By leveraging directional alignment in the learned feature space and assigning labels based on class-specific feature subspaces, our method mitigates corrupted supervision without requiring estimation of stochastic noise transition matrices. In addition, FB-NLL is model-independent and integrates seamlessly with existing noise-robust training techniques. Extensive experiments across diverse datasets and noise regimes demonstrate that our framework consistently outperforms state-of-the-art baselines in terms of average accuracy and performance stability.

We present VLA Foundry, an open-source framework that unifies LLM, VLM, and VLA training in a single codebase. Most open-source VLA efforts specialize on the action training stage, often stitching together incompatible pretraining pipelines. VLA Foundry instead provides a shared training stack with end-to-end control, from language pretraining to action-expert fine-tuning. VLA Foundry supports both from-scratch training and pretrained backbones from Hugging Face. To demonstrate the utility of our framework, we train and release two types of models: the first trained fully from scratch through our LLM-->VLM-->VLA pipeline and the second built on the pretrained Qwen3-VL backbone. We evaluate closed-loop policy performance of both models on LBM Eval, an open-data, open-source simulator. We also contribute usability improvements to the simulator and the STEP analysis tools for easier public use. In the nominal evaluation setting, our fully-open from-scratch model is on par with our prior closed-source work and substituting in the Qwen3-VL backbone leads to a strong multi-task table top manipulation policy outperforming our baseline by a wide margin. The VLA Foundry codebase is available at https://github.com/TRI-ML/vla_foundry and all multi-task model weights are released on https://huggingface.co/collections/TRI-ML/vla-foundry. Additional qualitative videos are available on the project website https://tri-ml.github.io/vla_foundry.

Despite the remarkable success of Vision Transformers (ViTs) across a wide range of vision tasks, recent studies have revealed that they remain vulnerable to adversarial examples, much like Convolutional Neural Networks (CNNs). A common empirical defense strategy is adversarial training, yet the theoretical underpinnings of its robustness in ViTs remain largely unexplored. In this work, we present the first theoretical analysis of adversarial training under simplified ViT architectures. We show that, when trained under a signal-to-noise ratio that satisfies a certain condition and within a moderate perturbation budget, adversarial training enables ViTs to achieve nearly zero robust training loss and robust generalization error under certain regimes. Remarkably, this leads to strong generalization even in the presence of overfitting, a phenomenon known as \emph{benign overfitting}, previously only observed in CNNs (with adversarial training). Experiments on both synthetic and real-world datasets further validate our theoretical findings.

The discretization of continuous numerical attributes remains a persistent computational bottleneck in the induction of decision trees, particularly as dataset dimensions scale. Building upon the recently proposed MSD-Splitting technique -- which bins continuous data using the empirical mean and standard deviation to dramatically improve the efficiency and accuracy of the C4.5 algorithm -- we introduce Adaptive MSD-Splitting (AMSD). While standard MSD-Splitting is highly effective for approximately symmetric distributions, its rigid adherence to fixed one-standard-deviation cutoffs can lead to catastrophic information loss in highly skewed data, a common artifact in real-world biomedical and financial datasets. AMSD addresses this by dynamically adjusting the standard deviation multiplier based on feature skewness, narrowing intervals in dense regions to preserve discriminative resolution. Furthermore, we integrate AMSD into ensemble methods, specifically presenting the Random Forest-AMSD (RF-AMSD) framework. Empirical evaluations on the Census Income, Heart Disease, Breast Cancer, and Forest Covertype datasets demonstrate that AMSD yields a 2-4% accuracy improvement over standard MSD-Splitting, while maintaining near-identical O(N) time complexity reductions compared to the O(N log N) exhaustive search. Our Random Forest extension achieves state-of-the-art accuracy at a fraction of standard computational costs, confirming the viability of adaptive statistical binning in large-scale ensemble learning architectures.

Large Language Models (LLMs) still struggle with multi-step logical reasoning. Existing approaches either purely refine the reasoning chain in natural language form or attach a symbolic solver as an external module. In this work, we instead ask whether LLMs contain a shared internal logical subspace that simultaneously aligns natural-language and symbolic-language views of the reasoning process. Our hypothesis is that this logical subspace captures logical reasoning capabilities in LLMs that are shared across views while remaining independent of surface forms. To verify this, we employ Canonical Correlation Analysis on the paired residual activations from natural-language and symbolic-language reasoning chains, learning a low-dimensional subspace with maximum cross-view correlation. Furthermore, we design a training-free approach that steers LLMs reasoning chain along this logical subspace, thereby leveraging the complementary reasoning signals from both views. Experiments on four logical reasoning benchmarks demonstrate the effectiveness of our approach, improving accuracy by up to 11 percentage points and generalizing well on out-of-domain problems.

In [97,99,100], an fl-RDT framework is introduced to characterize \emph{statistical computational gaps} (SCGs). Studying \emph{symmetric binary perceptrons} (SBPs), [100] obtained an \emph{algorithmic} threshold estimate $α_a\approx α_c^{(7)}\approx 1.6093$ at the 7th lifting level (for $κ=1$ margin), closely approaching $1.58$ local entropy (LE) prediction [18]. In this paper, we further connect parametric RDT to overlap gap properties (OGPs), another key geometric feature of the solution space. Specifically, for any positive integer $s$, we consider $s$-level ultrametric OGPs ($ult_s$-OGPs) and rigorously upper-bound the associated constraint densities $α_{ult_s}$. To achieve this, we develop an analytical union-bounding program consisting of combinatorial and probabilistic components. By casting the combinatorial part as a convex problem and the probabilistic part as a nested integration, we conduct numerical evaluations and obtain that the tightest bounds at the first two levels, $\barα_{ult_1} \approx 1.6578$ and $\barα_{ult_2} \approx 1.6219$, closely approach the 3rd and 4th lifting level parametric RDT estimates, $α_c^{(3)} \approx 1.6576$ and $α_c^{(4)} \approx 1.6218$. We also observe excellent agreement across other key parameters, including overlap values and the relative sizes of ultrametric clusters. Based on these observations, we propose several conjectures linking $ult$-OGP and parametric RDT. Specifically, we conjecture that algorithmic threshold $α_a=\lim_{s\rightarrow\infty} α_{ult_s} = \lim_{s\rightarrow\infty} \barα{ult_s} = \lim_{r\rightarrow\infty} α_{c}^{(r)}$, and $α_{ult_s} \leq α_{c}^{(s+2)}$ (with possible equality for some (maybe even all) $s$). Finally, we discuss the potential existence of a full isomorphism connecting all key parameters of $ult$-OGP and parametric RDT.

Kubernetes offers two default paths for scaling Node.js workloads, and both have structural limitations. The Horizontal Pod Autoscaler scales on CPU utilization, which does not directly measure event loop saturation: a Node.js pod can queue requests and miss latency SLOs while CPU reports moderate usage. KEDA extends HPA with richer triggers, including event-loop metrics, but inherits the same reactive control loop, detecting overload only after it has begun. By the time new pods start and absorb traffic, the system may already be degraded. Lowering thresholds shifts the operating point but does not change the dynamic: the scaler still reacts to a value it has already crossed, at the cost of permanent over-provisioning. We propose a predictive scaling algorithm that forecasts where load will be by the time new capacity is ready and scales proactively based on that forecast. Per-instance metrics are corrupted by the scaler's own actions: adding an instance redistributes load and changes every metric, even if external traffic is unchanged. We observe that operating on a cluster-wide aggregate that is approximately invariant under scaling eliminates this feedback loop, producing a stable signal suitable for short-term extrapolation. We define a metric model (a set of three functions that encode how a specific metric relates to scaling) and a five-stage pipeline that transforms raw, irregularly-timed, partial metric data into a clean prediction signal. In benchmarks against HPA and KEDA under steady ramp and sudden spike, the algorithm keeps per-instance load near the target threshold throughout. Under the steady ramp, median latency is 26ms, compared to 154ms for KEDA and 522ms for HPA.

The pursuit of truth is central to democratic deliberation and governance, yet political discourse reflects varying epistemic orientations, ranging from evidence-based reasoning grounded in verifiable information to intuition-based reasoning rooted in beliefs and subjective interpretation. We introduce a scalable approach to measure epistemic orientation using the Evidence--Minus--Intuition (EMI) score, derived from large language model (LLM) ratings and embedding-based semantic similarity. Applying this approach to 15 million parliamentary speech segments spanning 1946 to 2025 across seven countries, we examine temporal patterns in discourse and its association with deliberative democracy and governance. We find that EMI is positively associated with deliberative democracy within countries over time, with consistent relationships in both contemporaneous and lagged analyses. EMI is also positively associated with the transparency and predictable implementation of laws as a dimension of governance. These findings suggest that the epistemic nature of political discourse is crucial for both the quality of democracy and governance.

The standard Monte Carlo estimator $\widehat{I}_N^{\mathrm{MC}}$ of $\int fdω$ relies on independent samples from $ω$ and has variance of order $1/N$. Replacing the samples with a determinantal point process (DPP), a repulsive distribution, makes the estimator consistent, with variance rates that depend on how the DPP is adapted to $f$ and $ω$. We examine two existing DPP-based estimators: one by Bardenet & Hardy (2020) with a rate of $\mathcal{O}(N^{-(1+1/d)})$ for smooth $f$, but relying on a fixed DPP. The other, by Ermakov & Zolotukhin (1960), is unbiased with rate of order $1/N$, like Monte Carlo, but its DPP is tailored to $f$. We revisit these estimators, generalize them to continuous settings, and provide sampling algorithms.

We propose SmoothCruiser, a new planning algorithm for estimating the value function in entropy-regularized Markov decision processes and two-player games, given a generative model of the environment. SmoothCruiser makes use of the smoothness of the Bellman operator promoted by the regularization to achieve problem-independent sample complexity of order O~(1/epsilon^4) for a desired accuracy epsilon, whereas for non-regularized settings there are no known algorithms with guaranteed polynomial sample complexity in the worst case.

Understanding artworks requires multi-step reasoning over visual content and cultural, historical, and stylistic context. While recent multimodal large language models show promise in artwork explanation, they rely on implicit reasoning and internalized knowl- edge, limiting interpretability and explicit evidence grounding. We propose A-MAR, an Agent-based Multimodal Art Retrieval framework that explicitly conditions retrieval on structured reasoning plans. Given an artwork and a user query, A-MAR first decomposes the task into a structured reasoning plan that specifies the goals and evidence requirements for each step. Retrieval is then conditionedon this plan, enabling targeted evidence selection and supporting step-wise, grounded explanations. To evaluate agent-based multi- modal reasoning within the art domain, we introduce ArtCoT-QA. This diagnostic benchmark features multi-step reasoning chains for diverse art-related queries, enabling a granular analysis that extends beyond simple final answer accuracy. Experiments on SemArt and Artpedia show that A-MAR consistently outperforms static, non planned retrieval and strong MLLM baselines in final explanation quality, while evaluations on ArtCoT-QA further demonstrate its advantages in evidence grounding and multi-step reasoning ability. These results highlight the importance of reasoning-conditioned retrieval for knowledge-intensive multimodal understanding and position A-MAR as a step toward interpretable, goal-driven AI systems, with particular relevance to cultural industries. The code and data are available at: https://github.com/ShuaiWang97/A-MAR.

Answering open-ended questions remains challenging for AI systems because it requires synthesis, judgment, and exploration beyond factual retrieval, and users often refine answers through multiple iterations rather than accepting a single response. Existing QA benchmarks do not explicitly support this refinement process. To address this gap, we introduce a new task, document-grounded related insight generation, where the goal is to generate additional insights from a document collection that help improve, extend, or rethink an initial answer to an open-ended question, ultimately supporting richer user interaction and a better overall question answering experience. We curate and release SCOpE-QA (Scientific Collections for Open-Ended QA), a dataset of 3,000 open-ended questions across 20 research collections. We present InsightGen, a two-stage approach that first constructs a thematic representation of the document collection using clustering, and then selects related context based on neighborhood selection from the thematic graph to generate diverse and relevant insights using LLMs. Extensive evaluation on 3,000 questions using two generation models and two evaluation settings shows that InsightGen consistently produces useful, relevant, and actionable insights, establishing a strong baseline for this new task.

Explainable artificial intelligence (XAI) has predominantly focused on generating model-centric explanations that approximate the behavior of black-box models. However, such explanations often overlook a fundamental aspect of interpretability: different users require different explanations depending on their goals, preferences, and cognitive constraints. Although recent work has explored user-centric and personalized explanations, most existing approaches rely on heuristic adaptations or implicit user modeling, lacking a principled framework for representing and learning individual preferences. In this paper, we consider Preference-Based Explainable Artificial Intelligence (PREF-XAI), a novel perspective that reframes explanation as a preference-driven decision problem. Within PREF-XAI, explanations are not treated as fixed outputs, but as alternatives to be evaluated and selected according to user-specific criteria. In the PREF-XAI perspective, here we propose a methodology that combines rule-based explanations with formal preference learning. User preferences are elicited through a ranking of a small set of candidate explanations and modeled via an additive utility function inferred using robust ordinal regression. Experimental results on real-world datasets show that PREF-XAI can accurately reconstruct user preferences from limited feedback, identify highly relevant explanations, and discover novel explanatory rules not initially considered by the user. Beyond the proposed methodology, this work establishes a connection between XAI and preference learning, opening new directions for interactive and adaptive explanation systems.

Function vectors (FVs) are vector representations of tasks extracted from model activations during in-context learning. While prior work has shown that multilingual model representations can be language-agnostic, it remains unclear whether the same holds for function vectors. We study whether FVs exhibit language-agnosticity, using machine translation as a case study. Across three decoder-only multilingual LLMs, we find that translation FVs extracted from a single English$\rightarrow$Target direction transfer to other target languages, consistently improving the rank of correct translation tokens across multiple unseen languages. Ablation results show that removing the FV degrades translation across languages with limited impact on unrelated tasks. We further show that base-model FVs transfer to instruction-tuned variants and partially generalize from word-level to sentence-level translation.

Reinforcement learning-based control policies have been frequently demonstrated to be more effective than analytical techniques for many manipulation tasks. Commonly, these methods learn neural control policies that predict end-effector pose changes directly from observed state information. For tasks like inserting delicate connectors which induce force constraints, pose-based policies have limited explicit control over force and rely on carefully tuned low-level controllers to avoid executing damaging actions. In this work, we present hybrid position-force control policies that learn to dynamically select when to use force or position control in each control dimension. To improve learning efficiency of these policies, we introduce Mode-Aware Training for Contact Handling (MATCH) which adjusts policy action probabilities to explicitly mirror the mode selection behavior in hybrid control. We validate MATCH's learned policy effectiveness using fragile peg-in-hole tasks under extreme localization uncertainty. We find MATCH substantially outperforms pose-control policies -- solving these tasks with up to 10% higher success rates and 5x fewer peg breaks than pose-only policies under common types of state estimation error. MATCH also demonstrates data efficiency equal to pose-control policies, despite learning in a larger and more complex action space. In over 1600 sim-to-real experiments, we find MATCH succeeds twice as often as pose policies in high noise settings (33% vs.~68%) and applies ~30% less force on average compared to variable impedance policies on a Franka FR3 in laboratory conditions.

We introduce a new budgeted framework for online influence maximization, considering the total cost of an advertising campaign instead of the common cardinality constraint on a chosen influencer set. Our approach better models the real-world setting where the cost of influencers varies and advertisers want to find the best value for their overall social advertising budget. We propose an algorithm assuming an independent cascade diffusion model and edge level semi-bandit feedback, and provide both theoretical and experimental results. Our analysis is also valid for the cardinality constraint setting and improves the state of the art regret bound in this case.

Effective human-robot teaming is crucial for the practical deployment of robots in human workspaces. However, optimizing joint human-robot plans remains a challenge due to the difficulty of modeling individualized human capabilities and preferences. While prior research has leveraged the multi-cycle structure of domains like manufacturing to learn an individual's tendencies and adapt plans over repeated interactions, these techniques typically consider task-level and motion-level adaptation in isolation. Task-level methods optimize allocation and scheduling but often ignore spatial interference in close-proximity scenarios; conversely, motion-level methods focus on collision avoidance while ignoring the broader task context. This paper introduces RAPIDDS, a framework that unifies these approaches by modeling an individual's spatial behavior (motion paths) and temporal behavior (time required to complete tasks) over multiple cycles. RAPIDDS then jointly adapts task schedules and steers diffusion models of robot motions to maximize efficiency and minimize proximity accounting for these individualized models. We demonstrate the importance of this dual adaptation through an ablation study in simulation and a physical robot scenario using a 7-DOF robot arm. Finally, we present a user study (n=32) showing significant plan improvement compared to non-adaptive systems across both objective metrics, such as efficiency and proximity, and subjective measures, including fluency and user preference. See this paper's companion video at: https://youtu.be/55Q3lq1fINs.

Enforcing constraint satisfaction in neural network outputs is critical for safety, reliability, and physical fidelity in many control and decision-making applications. While soft-constrained methods penalize constraint violations during training, they do not guarantee constraint adherence during inference. Other approaches guarantee constraint satisfaction via specific parameterizations or a projection layer, but are tailored to specific forms (e.g., linear constraints), limiting their utility in other general problem settings. Many real-world problems of interest are nonlinear, motivating the development of methods that can enforce general nonlinear constraints. To this end, we introduce HardNet++, a constraint-enforcement method that simultaneously satisfies linear and nonlinear equality and inequality constraints. Our approach iteratively adjusts the network output via damped local linearizations. Each iteration is differentiable, admitting an end-to-end training framework, where the constraint satisfaction layer is active during training. We show that under certain regularity conditions, this procedure can enforce nonlinear constraint satisfaction to arbitrary tolerance. Finally, we demonstrate tight constraint adherence without loss of optimality in a learning-for-optimization context, where we apply this method to a model predictive control problem with nonlinear state constraints.

At present, executable visual workflows have emerged as a mainstream paradigm in real-world industrial deployments, offering strong reliability and controllability. However, in current practice, such workflows are almost entirely constructed through manual engineering: developers must carefully design workflows, write prompts for each step, and repeatedly revise the logic as requirements evolve-making development costly, time-consuming, and error-prone. To study whether large language models can automate this multi-round interaction process, we introduce Chat2Workflow, a benchmark for generating executable visual workflows directly from natural language, and propose a robust agentic framework to mitigate recurrent execution errors. Chat2Workflow is built from a large collection of real-world business workflows, with each instance designed so that the generated workflow can be transformed and directly deployed to practical workflow platforms such as Dify and Coze. Experimental results show that while state-of-the-art language models can often capture high-level intent, they struggle to generate correct, stable, and executable workflows, especially under complex or changing requirements. Although our agentic framework yields up to 5.34% resolve rate gains, the remaining real-world gap positions Chat2Workflow as a foundation for advancing industrial-grade automation. Code is available at https://github.com/zjunlp/Chat2Workflow.

Counterfactual explanations (CEs) provide an intuitive way to understand recommender systems by identifying minimal modifications to user-item interactions that alter recommendation outcomes. Existing CE methods for recommender systems, however, have been evaluated under heterogeneous protocols, using different datasets, recommenders, metrics, and even explanation formats, which hampers reproducibility and fair comparison. Our paper systematically reproduces, re-implement, and re-evaluate eleven state-of-the-art CE methods for recommender systems, covering both native explainers (e.g., LIME-RS, SHAP, PRINCE, ACCENT, LXR, GREASE) and specific graph-based explainers originally proposed for GNNs. Here, a unified benchmarking framework is proposed to assess explainers along three dimensions: explanation format (implicit vs. explicit), evaluation level (item-level vs. list-level), and perturbation scope (user interaction vectors vs. user-item interaction graphs). Our evaluation protocol includes effectiveness, sparsity, and computational complexity metrics, and extends existing item-level assessments to top-K list-level explanations. Through extensive experiments on three real-world datasets and six representative recommender models, we analyze how well previously reported strengths of CE methods generalize across diverse setups. We observe that the trade-off between effectiveness and sparsity depends strongly on the specific method and evaluation setting, particularly under the explicit format; in addition, explainer performance remains largely consistent across item level and list level evaluations, and several graph-based explainers exhibit notable scalability limitations on large recommender graphs. Our results refine and challenge earlier conclusions about the robustness and practicality of CE generation methods in recommender systems: https://github.com/L2R-UET/CFExpRec.

Damage identification is a core task in structural health monitoring. In practice, however, its reliability is often compromised by confounding non-damage effects, such as variations in excitation and environmental conditions, which can induce changes comparable to or larger than those caused by structural damage. To address this challenge, this study proposes a self-supervised label-free disentangled representation learning framework for robust vibration-based structural damage identification. The proposed framework employs an autoencoder with two latent representations to learn directly from raw vibration acceleration signals. A self-supervised invariance regularization, implemented via Variance-Invariance-Covariance Regularization (VICReg), is imposed on one latent representation using baseline data where structural damage is assumed constant but operational and environmental conditions vary. In addition, a frequency-domain constraint is introduced to enforce agreement between the power spectral density reconstructed from the latent representation and that computed from the corresponding input time series. Together, these mechanisms promote disentanglement, enabling the learned representation to be sensitive to damage-related characteristics while remaining invariant to nuisance variability. The framework is trained in a fully end-to-end and label-free manner, requiring no prior information on damage, excitation, or environmental conditions, making it well-suited for real-world applications. Its effectiveness is validated on two distinct real-world vibration datasets, including a bridge and a gearbox. The results demonstrate robustness to operational variability, strong generalization capability, and good performance in both damage detection and quantification.

AI agents promise to serve as general-purpose personal assistants for their users, which requires them to have access to private user data (e.g., personal and financial information). This poses a serious risk to security and privacy. Adversaries may attack the AI model (e.g., via prompt injection) to exfiltrate user data. Furthermore, sharing private data with an AI agent requires users to trust a potentially unscrupulous or compromised AI model provider with their private data. This paper presents GAAP (Guaranteed Accounting for Agent Privacy), an execution environment for AI agents that guarantees confidentiality for private user data. Through dynamic and directed user prompts, GAAP collects permission specifications from users describing how their private data may be shared, and GAAP enforces that the agent's disclosures of private user data, including disclosures to the AI model and its provider, comply with these specifications. Crucially, GAAP provides this guarantee deterministically, without trusting the agent with private user data, and without requiring any AI model or the user prompt to be free of attacks. GAAP enforces the user's permission specification by tracking how the AI agent accesses and uses private user data. It augments Information Flow Control with novel persistent data stores and annotations that enable it to track the flow of private information both across execution steps within a single task, and also over multiple tasks separated in time. Our evaluation confirms that GAAP blocks all data disclosure attacks, including those that make other state-of-the-art systems disclose private user data to untrusted parties, without a significant impact on agent utility.

Large language models have achieved remarkable progress on complex reasoning tasks. However, they often implicitly fabricate information when inputs are incomplete, producing confident but unreliable conclusions -- a failure mode we term ungrounded reasoning. We argue that this issue arises not from insufficient reasoning capability, but from the lack of inferential boundary awareness -- the ability to recognize when the necessary premises for valid inference are missing. To address this issue, we propose Grounded Reasoning via Interactive Reinforcement Learning (GRIL), a multi-turn reinforcement learning framework for grounded reasoning under incomplete information. GRIL decomposes the reasoning process into two stages: clarify and pause, which identifies whether the available information is sufficient, and grounded reasoning, which performs task solving once the necessary premises are established. We design stage-specific rewards to penalize hallucinations, enabling models to detect gaps, stop proactively, and resume reasoning after clarification. Experiments on GSM8K-Insufficient and MetaMATH-Insufficient show that GRIL significantly improves premise detection (up to 45%), leading to a 30% increase in task success while reducing average response length by over 20%. Additional analyses confirm robustness to noisy user responses and generalization to out-of-distribution tasks.

Human mobility data are used in numerous applications, ranging from public health to urban planning. Human mobility is inherently sensitive, as it can contain information such as religious beliefs and political affiliations. Historically, it has been proposed to modify the information using techniques such as aggregation, obfuscation, or noise addition, to adequately protect privacy and eliminate concerns. As these methods come at a great cost in utility, new methods leveraging development in generative models, were introduced. The extent to which such methods answer the privacy-utility trade-off remains an open problem. In this paper, we introduced a first step towards solving it, by the introduction and application of a new framework for utility evaluation. Furthermore, we provide evidence that privacy evaluation remains a great challenge to consider and that it should be tackled through adversarial evaluation in accordance with the current EU regulation. We propose a new membership inference attack against a subcategory of generative models, even though this subcategory was deemed private due to its resistance over the trajectory user-linking problem.

In this paper, we propose a deep-learning framework for environmental sound deepfake detection (ESDD) -- the task of identifying whether the sound scene and sound event in an input audio recording is fake or not. To this end, we conducted extensive experiments to explore how individual spectrograms, a wide range of network architectures and pre-trained models, ensemble of spectrograms or network architectures affect the ESDD task performance. The experimental results on the benchmark datasets of EnvSDD and ESDD-Challenge-TestSet indicate that detecting deepfake audio of sound scene and detecting deepfake audio of sound event should be considered as individual tasks. We also indicate that the approach of finetuning a pre-trained model is more effective compared with training a model from scratch for the ESDD task. Eventually, our best model, which was finetuned from the pre-trained WavLM model with the proposed three-stage training strategy, achieve the Accuracy of 0.98, F1 Score of 0.95, AuC of 0.99 on EnvSDD Test subset and the Accuracy of 0.88, F1 Score of 0.77, and AuC of 0.92 on ESDD-Challenge-TestSet dataset.

SAM3 advances open-vocabulary semantic segmentation by introducing a prompt-driven mask generation paradigm. However, in multi-class open-vocabulary scenarios, masks generated independently from different category prompts lack a unified and inter-class comparable evidence scale, often resulting in overlapping coverage and unstable competition. Moreover, synonymous expressions of the same concept tend to activate inconsistent semantic and spatial evidence, leading to intra-class drift that exacerbates inter-class conflicts and compromises overall inference stability. To address these issues, we propose CoCo-SAM3 (Concept-Conflict SAM3), which explicitly decouples inference into intra-class enhancement and inter-class competition. Our method first aligns and aggregates evidence from synonymous prompts to strengthen concept consistency. It then performs inter-class competition on a unified comparable scale, enabling direct pixel-wise comparisons among all candidate classes. This mechanism stabilizes multi-class inference and effectively mitigates inter-class conflicts. Without requiring any additional training, CoCo-SAM3 achieves consistent improvements across eight open-vocabulary semantic segmentation benchmarks.

An earlier paper (Hong, Potteiger, and Zapata 2026) established that an unoptimized GPT 4.1 prompt predicts fan-reported experience ratings within one point 67% of the time from open-ended survey text. This paper tests the relative impact of prompt design and model selection on that performance. We compared four configurations on approximately 10,000 post-game surveys from five MLB teams: the original baseline prompt and a moderately customized version, crossed with three GPT models (4.1, 4.1-mini, 5.2). Prompt customization added roughly two percentage points of within +/-1 agreement on GPT 4.1 (from 67% to 69%). Both model swaps from that best configuration degraded performance: GPT 5.2 returned to the baseline, and GPT 4.1-mini fell six percentage points below it. Both levers combined were dwarfed by the input itself: across capable configurations, accuracy varied more than an order of magnitude more by the linguistic character of the text than by the choice of prompt or model. The ceiling has two parts. One is a bias in how the model reads text, which prompt design can correct. The other is a difference between what fans write about and what they actually decide, which no engineering can close because the missing information is not in the text. Prompt customization moved the first part; model selection moved neither reliably. The result is not that "prompt engineering helps a little" but that prompt engineering helps in a specific and predictable way, on the part of the ceiling it can reach.

Edge devices such as smartwatches and smart glasses cannot continuously run even the smallest 100M-1B parameter language models due to power and compute constraints, yet cloud inference introduces multi-second latencies that break the illusion of a responsive assistant. We introduce micro language models ($μ$LMs): ultra-compact models (8M-30M parameters) that instantly generate the first 4-8 words of a contextually grounded response on-device, while a cloud model completes it; thus, masking the cloud latency. We show that useful language generation survives at this extreme scale with our models matching several 70M-256M-class existing models. We design a collaborative generation framework that reframes the cloud model as a continuator rather than a respondent, achieving seamless mid-sentence handoffs and structured graceful recovery via three error correction methods when the local opener goes wrong. Empirical results show that $μ$LMs can initiate responses that larger models complete seamlessly, demonstrating that orders-of-magnitude asymmetric collaboration is achievable and unlocking responsive AI for extremely resource-constrained devices. The model checkpoint and demo are available at https://github.com/Sensente/micro_language_model_swen_project.

Modern world models are becoming too complex to admit explicit dynamical descriptions. We study safety-critical contextual control, where a Planner must optimize a task objective using only feasibility samples from a black-box Simulator, conditioned on a context signal $ξ_t$. We develop a sample-based Penalized Predictive Control (PPC) framework grounded in online Riemannian optimization, in which the Simulator compresses the feasibility manifold into a score-based density $\hat{p}(u \mid ξ_t)$ that endows the action space with a Riemannian geometry guiding the Planner's gradient descent. The barrier curvature $κ(ξ_t)$, the minimum curvature of the conditional log-density $-\ln\hat{p}(\cdot\midξ_t)$, governs both convergence rate and safety margin, replacing the Lipschitz constant of the unknown dynamics. Our main result is a contextual safety bound showing that the distance from the true feasibility manifold is controlled by the score estimation error and a ratio that depends on $κ(ξ_t)$, both of which improve with richer context. Simulations on a dynamic navigation task confirm that contextual PPC substantially outperforms marginal and frozen density models, with the advantage growing after environment shifts.

Multimodal Large Language Models are increasingly adopted as autonomous agents in interactive environments, yet their ability to proactively address safety hazards remains insufficient. We introduce SafetyALFRED, built upon the embodied agent benchmark ALFRED, augmented with six categories of real-world kitchen hazards. While existing safety evaluations focus on hazard recognition through disembodied question answering (QA) settings, we evaluate eleven state-of-the-art models from the Qwen, Gemma, and Gemini families on not only hazard recognition, but also active risk mitigation through embodied planning. Our experimental results reveal a significant alignment gap: while models can accurately recognize hazards in QA settings, average mitigation success rates for these hazards are low in comparison. Our findings demonstrate that static evaluations through QA are insufficient for physical safety, thus we advocate for a paradigm shift toward benchmarks that prioritize corrective actions in embodied contexts. We open-source our code and dataset under https://github.com/sled-group/SafetyALFRED.git

While generative models have set new benchmarks for Target Speaker Extraction (TSE), their inherent reliance on global context precludes deployment in real-time applications. Direct adaptation to streaming scenarios often leads to catastrophic inference performance degradation due to the severe mismatch between training and streaming inference. To bridge this gap, we present the first autoregressive (AR) models tailored for streaming TSE. Our approach introduces a Chunk-wise Interleaved Splicing Paradigm that ensures highly efficient and stable streaming inference. To ensure the coherence between the extracted speech segments, we design a historical context refinement mechanism that mitigates boundary discontinuities by leveraging historical information. Experiments on Libri2Mix show that while AR generative baseline exhibits performance degradation at low latencies, our approach maintains 100% stability and superior intelligibility. Furthermore, our streaming results are comparable to or even surpass offline baselines. Additionally, our model achieves a Real-Time-Factor (RTF) of 0.248 on consumer-level GPUs. This work provides empirical evidence that AR generative backbones are viable for latency-sensitive applications through the Chunk-wise Interleaved Splicing Paradigm.

Evaluating the reasoning capabilities of Large Language Models (LLMs) for complex, quantitative financial tasks is a critical and unsolved challenge. Standard benchmarks often fail to isolate an agent's core ability to parse queries and orchestrate computations. To address this, we introduce a novel evaluation methodology and benchmark designed to rigorously measure an LLM agent's reasoning for financial time-series analysis. We apply this methodology in a large-scale empirical study using our framework, Time Series Augmented Generation (TSAG), where an LLM agent delegates quantitative tasks to verifiable, external tools. Our benchmark, consisting of 100 financial questions, is used to compare multiple SOTA agents (e.g., GPT-4o, Llama 3, Qwen2) on metrics assessing tool selection accuracy, faithfulness, and hallucination. The results demonstrate that capable agents can achieve near-perfect tool-use accuracy with minimal hallucination, validating the tool-augmented paradigm. Our primary contribution is this evaluation framework and the corresponding empirical insights into agent performance, which we release publicly to foster standardized research on reliable financial AI.

Edge-cloud hybrid inference offloads difficult inputs to a powerful remote model, but the uplink channel imposes hard per-request constraints on the number of bits that can be transmitted. We show that selecting transmitted content based solely on attention-based importance, the standard approach in collaborative inference, is inherently limited under hard budgets. Two findings support this claim. First, replacing high-importance units with low-importance but complementary ones improves server accuracy. This shows that what matters is not individual importance but how well the transmitted set covers diverse aspects of the input. Second, spatially uniform selection without any content information achieves competitive accuracy at moderate budgets. This confirms that spatial coverage alone carries independent value. Based on this analysis, we propose SAGE (Semantic Attention-Guided Evidence), a principled, training-free method that combines importance filtering with embedding-diversity sampling. SAGE achieves 93% of the server ceiling in offloaded accuracy while transmitting fewer than half of the available evidence units on ImageNet-1K, substantially outperforming importance-only composition.

Free-association norms provide essential empirical data for investigating linguistic, semantic, and cultural phenomena in the cognitive sciences. Although large-scale norms exist for languages such as English, Dutch, Spanish, and Mandarin Chinese, no comparable resource has been available for German. To address this gap, we present free-association norms for 5,877 German cue words as part of the German version of the multilingual Small World of Words (SWOW) project. We describe the data collection procedures, participant characteristics, and our comprehensive preprocessing pipeline before introducing the resulting SWOW-DE data set. Using data from three established psycholinguistic paradigms, we show that SWOW-DE norms robustly predict performance in lexical decision tasks, relatedness judgments, and psycholinguistic word ratings. Furthermore, we demonstrate that SWOW-DE responses compare favorably with existing German resources and provide a preliminary cross-linguistic comparison revealing both shared and language-specific association patterns, highlighting promising directions for future research. Overall, SWOW-DE represents the largest collection of German free associations to date and offers a unique resource for linguistic, psychological, and cross-cultural research.

Systematic ablations are essential to attribute performance gains in AI Virtual Cells, yet they are rarely performed because biological repositories are under-standardized and tightly coupled to domain-specific data and formats. While recent coding agents can translate ideas into implementations, they typically stop at producing code and lack a verifier that can reproduce strong baselines and rigorously test which components truly matter. We introduce AblateCell, a reproduce-then-ablate agent for virtual cell repositories that closes this verification gap. AblateCell first reproduces reported baselines end-to-end by auto-configuring environments, resolving dependency and data issues, and rerunning official evaluations while emitting verifiable artifacts. It then conducts closed-loop ablation by generating a graph of isolated repository mutations and adaptively selecting experiments under a reward that trades off performance impact and execution cost. Evaluated on three single-cell perturbation prediction repositories (CPA, GEARS, BioLORD), AblateCell achieves 88.9% (+29.9% to human expert) end-to-end workflow success and 93.3% (+53.3% to heuristic) accuracy in recovering ground-truth critical components. These results enable scalable, repository-grounded verification and attribution directly on biological codebases.

This study compared repeated generation consistency of exercise prescription outputs across three large language models (LLMs), specifically GPT-4.1, Claude Sonnet 4.6, and Gemini 2.5 Flash, under temperature=0 conditions. Each model generated prescriptions for six clinical scenarios 20 times, yielding 360 total outputs analyzed across four dimensions: semantic similarity, output reproducibility, FITT classification, and safety expression. Mean semantic similarity was highest for GPT-4.1 (0.955), followed by Gemini 2.5 Flash (0.950) and Claude Sonnet 4.6 (0.903), with significant inter-model differences confirmed (H = 458.41, p < .001). Critically, these scores reflected fundamentally different generative behaviors: GPT-4.1 produced entirely unique outputs (100%) with stable semantic content, while Gemini 2.5 Flash showed pronounced output repetition (27.5% unique outputs), indicating that its high similarity score derived from text duplication rather than consistent reasoning. Identical decoding settings thus yielded fundamentally different consistency profiles, a distinction that single-output evaluations cannot capture. Safety expression reached ceiling levels across all models, confirming its limited utility as a differentiating metric. These results indicate that model selection constitutes a clinical rather than merely technical decision, and that output behavior under repeated generation conditions should be treated as a core criterion for reliable deployment of LLM-based exercise prescription systems.

The importance of clear and correct text in legal documents cannot be understated, and, consequently, a grammatical error correction tool meant to assist a professional in the law must have the ability to understand the possible errors in the context of a legal environment, correcting them accordingly, and implicitly needs to be trained in the same environment, using realistic legal data. However, the manually annotated data required by such a process is in short supply for languages such as Romanian, much less for a niche domain. The most common approach is the synthetic generation of parallel data; however, it requires a structured understanding of the Romanian grammar. In this paper, we introduce, to our knowledge, the first Romanian-language parallel dataset for the detection and correction of grammatical errors in the legal domain, RoLegalGEC, which aggregates 350,000 examples of errors in legal passages, along with error annotations. Moreover, we evaluate several neural network models that transform the dataset into a valuable tool for both detecting and correcting grammatical errors, including knowledge-distillation Transformers, sequence tagging architectures for detection, and a variety of pre-trained text-to-text Transformer models for correction. We consider that the set of models, together with the novel RoLegalGEC dataset, will enrich the resource base for further research on Romanian.

The growing interest of Natural Language Processing (NLP) in minority languages has not yet bridged the gap in the preservation of oral linguistic heritage. In particular, extemporaneous poetry - a performative genre based on real-time improvisation, metrical-rhetorical competence - remains a largely unexplored area of computational linguistics. This methodological gap necessitates the creation of specific resources to document and analyse the structures of improvised poetry. This is the context in which A Bolu was created, the first structured corpus of extemporaneous poetry dedicated to cantada logudorese, a variant of the Sardinian language. The dataset comprises 2,835 stanzas for a total of 141,321 tokens. The study presents the architecture of the corpus and applies a multidimensional analysis combining descriptive statistical indices and computational linguistics techniques to map the characteristics of the poetic text. The results indicate that the production of Sardinian extemporaneous poets is characterised by recurring patterns that support Parry and Lord's theory of formulaicity. This evidence not only provides a new key to understanding oral creativity, but also offers a significant contribution to the development of NLP tools that are more inclusive and sensitive to the specificities of less widely spoken languages.

With the rapid advancement of Large Language Models (LLMs), the academic community has faced unprecedented disruptions, particularly in the realm of academic communication. The primary function of peer review is improving the quality of academic manuscripts, such as clarity, originality and other evaluation aspects. Although prior studies suggest that LLMs are beginning to influence peer review, it remains unclear whether they are altering its core evaluative functions. Moreover, the extent to which LLMs affect the linguistic form, evaluative focus, and recommendation-related signals of peer-review reports has yet to be systematically examined. In this study, we examine the changes in peer review reports for academic articles following the emergence of LLMs, emphasizing variations at fine-grained level. Specifically, we investigate linguistic features such as the length and complexity of words and sentences in review comments, while also automatically annotating the evaluation aspects of individual review sentences. We also use a maximum likelihood estimation method, previously established, to identify review reports that potentially have modified or generated by LLMs. Finally, we assess the impact of evaluation aspects mentioned in LLM-assisted review reports on the informativeness of recommendation for paper decision-making. The results indicate that following the emergence of LLMs, peer review texts have become longer and more fluent, with increased emphasis on summaries and surface-level clarity, as well as more standardized linguistic patterns, particularly reviewers with lower confidence score. At the same time, attention to deeper evaluative dimensions, such as originality, replicability, and nuanced critical reasoning, has declined.

As model capabilities advance, research has increasingly shifted toward long-horizon, multi-turn terminal-centric agentic tasks, where raw environment feedback is often preserved in the interaction history to support future decisions. However, repeatedly retaining such feedback introduces substantial redundancy and causes cumulative token cost to grow quadratically with the number of steps, hindering long-horizon reasoning. Although observation compression can mitigate this issue, the heterogeneity of terminal environments makes heuristic-based or fixed-prompt methods difficult to generalize. We propose TACO, a plug-and-play, self-evolving Terminal Agent Compression framework that automatically discovers and refines compression rules from interaction trajectories for existing terminal agents. Experiments on TerminalBench (TB 1.0 and TB 2.0) and four additional terminal-related benchmarks (i.e., SWE-Bench Lite, CompileBench, DevEval, and CRUST-Bench) show that TACO consistently improves performance across mainstream agent frameworks and strong backbone models. With MiniMax-2.5, it improves performance on most benchmarks while reducing token overhead by around 10%. On TerminalBench, it brings consistent gains of 1%-4% across strong agentic models, and further improves accuracy by around 2%-3% under the same token budget. These results demonstrate the effectiveness and generalization of self-evolving, task-aware compression for terminal agents.

Q-learning is one of the most fundamental algorithms in reinforcement learning. We analyze constant-stepsize Q-learning through a direct stochastic switching system representation. The key observation is that the Bellman maximization error can be represented exactly by a stochastic policy. Therefore, the Q-learning error admits a switched linear conditional-mean recursion with martingale-difference noise. The intrinsic drift rate is the joint spectral radius (JSR) of the direct switching family, which can be strictly smaller than the standard row-sum rate. Using this representation, we derive a finite-time final-iterate bound via a JSR-induced Lyapunov function and then give a computable quadratic-certificate version.

Reinforcement learning (RL) as post-training is crucial for enhancing the reasoning ability of large language models (LLMs) in coding and math. However, their capacity for visual semantic arithmetic, inferring relationships from images, remains underexplored. The classic text analogy "king"-"man"+"woman" = "queen" illustrates relational reasoning, yet replacing text with images of "king" and "man" significantly reduces performance because it requires commonsense knowledge and the extraction of concise concepts from irrelevant visual details. This capability is important for service and domestic robotics in unstructured environments, where robots must infer semantic relationships among objects, agents, and actions. In a kitchen, recognizing from images that "powder" and "cake" are related by "is made of" grounds symbolic relations in perception, enabling tool substitution, task generalization, and improved semantic reasoning. Prior work approaches semantic arithmetic by decoding image features after vector arithmetic, but suffers from modality gaps and lacks systematic evaluation. In this paper, we formulate two novel tasks, two-term subtraction and three-term operations, and construct the Image-Relation-Pair Dataset (IRPD) for benchmarking. We further propose Semantic Arithmetic Reinforcement Fine-Tuning (SAri-RFT), which post-trains large vision-language models (LVLMs) using a verifiable function and Group Relative Policy Optimization (GRPO). Our method achieves state-of-the-art results on IRPD and the real-world Visual7W-Telling dataset. By equipping LVLMs with robust cross-modal relational reasoning, this work advances domestic robots' ability to ground symbolic reasoning in perception, enhancing decision-making, tool adaptability, and human-robot interaction in complex environments. Datasets and source code are provided in the supplementary material.

Reliable biomedical and clinical retrieval requires more than strong ranking performance: it requires a practical way to find systematic model failures and curate the training evidence needed to correct them. Late-interaction models such as ColBERT provide a first solution thanks to the interpretable token-level interaction scores they expose between document and query tokens. Yet this interpretability is shallow: it explains a particular document--query pairwise score, but does not reveal whether the model has learned a clinical concept in a stable, reusable, and context-sensitive way across diverse expressions. As a result, these scores provide limited support for diagnosing misunderstandings, identifying irreasonably distant biomedical concepts, or deciding what additional data or feedback is needed to address this. In this short position paper, we propose Diagnosable ColBERT, a framework that aligns ColBERT token embeddings to a reference latent space grounded in clinical knowledge and expert-provided conceptual similarity constraints. This alignment turns document encodings into inspectable evidence of what the model appears to understand, enabling more direct error diagnosis and more principled data curation without relying on large batteries of diagnostic queries.

Hallucinations in Speech Large Language Models (SpeechLLMs) pose significant risks, yet existing detection methods typically rely on gold-standard outputs that are costly or impractical to obtain. Moreover, hallucination detection methods developed for text-based LLMs do not directly capture audio-specific signals. We investigate four attention-derived metrics: AUDIORATIO, AUDIOCONSISTENCY, AUDIOENTROPY, and TEXTENTROPY, designed to capture pathological attention patterns associated with hallucination, and train lightweight logistic regression classifiers on these features for efficient inference-time detection. Across automatic speech recognition and speech-to-text translation tasks, evaluations on Qwen-2-Audio and Voxtral-3B show that our approach outperforms uncertainty-based and prior attention-based baselines on in-domain data, achieving improvements of up to +0.23 PR-AUC, and generalises to out-of-domain ASR settings. We further find that strong performance can be achieved with approximately 100 attention heads, improving out-of-domain generalisation compared to using all heads. While effectiveness is model-dependent and task-specific training is required, our results demonstrate that attention patterns provide a valuable tool for hallucination detection in SpeechLLMs.

Egocentric assistants often rely on first-person view data to capture user behavior and context for personalized services. Since different users exhibit distinct habits, preferences, and routines, such personalization is essential for truly effective assistance. However, effectively integrating long-term user data for personalization remains a key challenge. To address this, we introduce EgoSelf, a system that includes a graph-based interaction memory constructed from past observations and a dedicated learning task for personalization. The memory captures temporal and semantic relationships among interaction events and entities, from which user-specific profiles are derived. The personalized learning task is formulated as a prediction problem where the model predicts possible future interactions from individual user's historical behavior recorded in the graph. Extensive experiments demonstrate the effectiveness of EgoSelf as a personalized egocentric assistant. Code is available at \href{https://abie-e.github.io/egoself_project/}{https://abie-e.github.io/egoself\_project/}.

Structure-based drug discovery faces the dual challenge of accurately capturing 3D protein-ligand interactions while navigating ultra-large chemical spaces to identify synthetically accessible candidates. In this work, we present a unified framework that addresses these challenges by combining contrastive 3D structure encoding with autoregressive molecular generation conditioned on commercial compound spaces. First, we introduce an SE(3)-equivariant transformer that encodes ligand and pocket structures into a shared embedding space via contrastive learning, achieving competitive results in zero-shot virtual screening. Second, we integrate these embeddings into a multimodal Chemical Language Model (MCLM). The model generates target-specific molecules conditioned on either pocket or ligand structures, with a learned dataset token that steers the output toward targeted chemical spaces, yielding candidates with favorable predicted binding properties across diverse targets.

Large Language Models (LLMs) utilize large amounts of data for their training, some of which may come from copyrighted sources. Membership Inference Attacks (MIA) aim to detect those documents and whether they have been included in the training corpora of the LLMs. The black-box MIAs require a significant amount of data manipulation; therefore, their comparison is often challenging. We study state-of-the-art (SOTA) MIAs under the black-box assumptions and compare them to each other using a unified set of datasets to determine if any of them can reliably detect membership under SOTA LLMs. In addition, a new method, called the Familiarity Ranking, was developed to showcase a possible approach to black-box MIAs, thereby giving LLMs more freedom in their expression to understand their reasoning better. The results indicate that none of the methods are capable of reliably detecting membership in LLMs, as shown by an AUC-ROC of approximately 0.5 for all methods across several LLMs. The higher TPR and FPR for more advanced LLMs indicate higher reasoning and generalizing capabilities, showcasing the difficulty of detecting membership in LLMs using black-box MIAs.

The goal of machine learning is to find models that minimize prediction error on data that has not yet been seen. Its operational paradigm assumes access to a dataset $S$ and articulates a scheme for evaluating how well a given model performs on an arbitrary sample. The sample can be $S$ (in which case we speak of ``in-sample'' performance) or some entirely new $S'$ (in which case we speak of ``out-of-sample'' performance). Traditional analysis of generalization assumes that both in- and out-of-sample data are i.i.d.\ draws from an infinite population. However, these probabilistic assumptions cannot be verified even in principle. This paper presents an alternative view of generalization through the lens of sensitivity analysis of solutions of optimization problems to perturbations in the problem data. Under this framework, generalization bounds are obtained by purely deterministic means and take the form of variational principles that relate in-sample and out-of-sample evaluations through an error term that quantifies how close out-of-sample data are to in-sample data. Statistical assumptions can then be used \textit{ex post} to characterize the situations when this error term is small (either on average or with high probability).

Construction workers are highly vulnerable to heat stress, yet tools that translate real-time physiological data into actionable safety intelligence remain scarce. This study addresses this gap by developing and evaluating deep learning models, specifically a baseline Long Short-Term Memory (LSTM) network and an attention-based LSTM, to predict heat stress among 19 workers in Saudi Arabia. Using Garmin Vivosmart 5 smartwatches to monitor metrics such as heart rate, HRV, and oxygen saturation, the attention-based model outperformed the baseline, achieving 95.40% testing accuracy and significantly reducing false positives and negatives. With precision, recall, and F1 scores of 0.982, this approach not only improves predictive performance but also offers interpretable results suitable for integration into IoT-enabled safety systems and BIM dashboards, advancing proactive, informatics-driven safety management in the construction industry.

Automated theorem proving is fundamental to formal methods, and the recent trend is to integrate large language models (LLMs) and proof assistants to form effective proof agents. While existing proof agents show promising performance, they inadequately leverage reasoning capabilities of modern LLMs in high-level planning and self-critique. We argue that proof agents should not merely generate tactics but also reason strategically about proof plans and critically evaluate their own proposals. This paper introduces ReCent-Prover, a reasoning-centric LLM-based proof agent for Rocq that addresses two critical limitations in current systems. First, we present validation with reflection, enabling LLMs to scrutinize their generated tactics and synthesize failure summaries when reflection identifies potential errors, filtering out potentially misapplied tactics earlier. Second, we propose retrieval with planning, which conditions retrieval on LLM-generated proof plans rather than subgoal similarity, retrieving lemmas and proofs that align with the anticipated proof strategy. Both techniques increase the number of invocations of LLMs. However, when evaluated on the CoqStoq benchmark, even under the same budget of LLM invocations, ReCent-Prover achieves a 22.58% relative improvement in the number of proved theorems over the previous state-of-the-art, demonstrating that our reasoning-centric design significantly enhances automated theorem proving capabilities.

Large Language Model agents have rapidly evolved from static text generators into dynamic systems capable of executing complex autonomous workflows. To enhance reliability, multi-agent frameworks assigning specialized roles are increasingly adopted to enable self-reflection and mutual auditing. While such role-playing effectively leverages domain expert knowledge, we find it simultaneously induces a human-like cognitive bias known as Actor-Observer Asymmetry (AOA). Specifically, an agent acting as an actor (during self-reflection) tends to attribute failures to external factors, whereas an observer (during mutual auditing) attributes the same errors to internal faults. We quantify this using our new Ambiguous Failure Benchmark, which reveals that simply swapping perspectives triggers the AOA effect in over 20% of cases for most models. To tame this bias, we introduce ReTAS (Reasoning via Thesis-Antithesis-Synthesis), a model trained through dialectical alignment to enforce perspective-invariant reasoning. By integrating dialectical chain-of-thought with Group Relative Policy Optimization, ReTAS guides agents to synthesize conflicting viewpoints into an objective consensus. Experiments demonstrate that ReTAS effectively mitigates attribution inconsistency and significantly improves fault resolution rates in ambiguous scenarios.

Emotion-Cause Pair Extraction in Conversations (ECPEC) aims to identify the set of causal relations between emotion utterances and their triggering causes within a dialogue. Most existing approaches formulate ECPEC as an independent pairwise classification task, overlooking the distinct semantics of emotion diffusion and cause explanation, and failing to capture globally consistent many-to-many conversational causality. To address these limitations, we revisit ECPEC from a semantic perspective and seek to disentangle emotion-oriented semantics from cause-oriented semantics, mapping them into two complementary representation spaces to better capture their distinct conversational roles. Building on this semantic decoupling, we naturally formulate ECPEC as a global alignment problem between the emotion-side and cause-side representations, and employ optimal transport to enable many-to-many and globally consistent emotion-cause matching. Based on this perspective, we propose a unified framework SCALE that instantiates the above semantic decoupling and alignment principle within a shared conversational structure. Extensive experiments on several benchmark datasets demonstrate that SCALE consistently achieves state-of-the-art performance. Our codes are released at https://github.com/CoCoSphere/SCALE.

Multimodal reward models (MRMs) play a crucial role in aligning Multimodal Large Language Models (MLLMs) with human preferences. Training a good MRM requires high-quality multimodal preference data. However, existing preference datasets face three key challenges: lack of granularity in preference strength, textual style bias, and unreliable preference signals. Besides, existing open-source multimodal preference datasets suffer from substantial noise, yet there is a lack of effective and scalable curation methods to enhance their quality. To address these limitations, we propose \textbf{DT2IT-MRM}, which integrates a \textbf{D}ebiased preference construction pipeline, a novel reformulation of text-to-image (\textbf{T2I}) preference data, and an \textbf{I}terative \textbf{T}raining framework that curates existing multimodal preference datasets for \textbf{M}ultimodal \textbf{R}eward \textbf{M}odeling. Our experimental results show that DT2IT-MRM achieves new \textbf{state-of-the-art} overall performance on three major benchmarks: VL-RewardBench, Multimodal RewardBench, and MM-RLHF-RewardBench.

Teams of LLM agents increasingly collaborate on tasks spanning days or weeks: multi-day data-generation sprints where generator, reviewer, and auditor agents coordinate in real time on overlapping batches; specialists carrying findings forward across session restarts; product decisions compounding over many review rounds. This requires agents to share, evaluate, and combine each other's cognitive state in real time across sessions. We call this cross-session agent-to-agent cognitive collaboration, distinct from parallel agent execution. To enable it, three problems must be solved together. (P1) Each agent decides field by field what to accept from peers, not accept or reject whole messages. (P2) Every claim is traceable to source, so returning claims are recognised as echoes of the receiver's own prior thinking. (P3) Memory that survives session restarts is relevant because of how it was stored, not how it is retrieved. These are protocol-level properties at the semantic layer of agent communication, distinct from tool-access and task-delegation protocols at lower layers. We call this missing protocol layer "semantic infrastructure," and the Mesh Memory Protocol (MMP) specifies it. Four composable primitives work together: CAT7, a fixed seven-field schema for every Cognitive Memory Block (CMB); SVAF, which evaluates each field against the receiver's role-indexed anchors and realises P1; inter-agent lineage, carried as parents and ancestors of content-hash keys and realising P2; and remix, which stores only the receiver's own role-evaluated understanding of each accepted CMB, never the raw peer signal, realising P3. MMP is specified, shipped, and running in production across three reference deployments, where each session runs an autonomous agent as a mesh peer with its own identity and memory, collaborating with other agents across the network for collective intelligence.

Agentic AI, with goal-directed, proactive, and autonomous decision-making capabilities, offers a compelling opportunity to address movement-related risks in human activity, including the persistent hazard of falls among elderly populations. Despite numerous approaches to fall mitigation through fall prediction and detection, existing systems have not yet functioned as universal solutions across care pathways and safety-critical environments. This is largely due to limitations in consistently handling real-world complexity, particularly poor context awareness, high false alarm rates, environmental noise, and data scarcity. We argue that fall detection and fall prediction can usefully be formulated as anomaly detection problems and more effectively addressed through an agentic AI system. More broadly, this perspective enables the early identification of subtle deviations in movement patterns associated with increased risk, whether arising from age-related decline, fatigue, or environmental factors. While technical requirements for immediate deployment are beyond the scope of this paper, we propose a conceptual framework that highlights potential value. This framework promotes a well-orchestrated approach to risk management by dynamically selecting relevant tools and integrating them into adaptive decision-making workflows, rather than relying on static configurations tailored to narrowly defined scenarios.

We introduce the Cyber Defense Benchmark, a benchmark for measuring how well large language model (LLM) agents perform the core SOC analyst task of threat hunting: given a database of raw Windows event logs with no guided questions or hints, identify the exact timestamps of malicious events. The benchmark wraps 106 real attack procedures from the OTRF Security-Datasets corpus - spanning 86 MITRE ATT&CK sub-techniques across 12 tactics - into a Gymnasium reinforcement-learning environment. Each episode presents the agent with an in-memory SQLite database of 75,000-135,000 log records produced by a deterministic campaign simulator that time-shifts and entity-obfuscates the raw recordings. The agent must iteratively submit SQL queries to discover malicious event timestamps and explicitly flag them, scored CTF-style against Sigma-rule-derived ground truth. Evaluating five frontier models - Claude Opus 4.6, GPT-5, Gemini 3.1 Pro, Kimi K2.5, and Gemini 3 Flash - on 26 campaigns covering 105 of 106 procedures, we find that all models fail dramatically: the best model (Claude Opus 4.6) submits correct flags for only 3.8% of malicious events on average, and no run across any model ever finds all flags. We define a passing score as >= 50% recall on every ATT&CK tactic - the minimum bar for unsupervised SOC deployment. No model passes: the leader clears this bar on 5 of 13 tactics and the remaining four on zero. These results suggest that current LLMs are poorly suited for open-ended, evidence-driven threat hunting despite strong performance on curated Q&A security benchmarks.

Tokenizing music to fit the general framework of language models is a compelling challenge, especially considering the diverse symbolic structures in which music can be represented (e.g., sequences, grids, and graphs). To date, most approaches tokenize symbolic music as sequences of musical events, such as onsets, pitches, time shifts, or compound note events. This strategy is intuitive and has proven effective in Transformer-based models, but it treats the regularity of musical time implicitly: individual tokens may span different durations, resulting in non-uniform time progression. In this paper, we instead consider whether an alternative tokenization is possible, where a uniform-length musical step (e.g., a beat) serves as the basic unit. Specifically, we encode all events within a single time step at the same pitch as one token, and group tokens explicitly by time step, which resembles a sparse encoding of a piano-roll representation. We evaluate the proposed tokenization on music continuation and accompaniment generation tasks, comparing it with mainstream event-based methods. Results show improved musical quality and structural coherence, while additional analyses confirm higher efficiency and more effective capture of long-range patterns with the proposed tokenization.

Transformer-based scientific foundation models are increasingly deployed in high-stakes settings, but current architectures give deterministic outputs and provide limited support for calibrated predictive uncertainty. We propose Stochastic Attention, a lightweight inference-time modification that randomizes attention by replacing softmax weights with normalized multinomial samples controlled by a single concentration parameter, and produces predictive ensembles without retraining. To set this parameter, we introduce a calibration objective that matches the stochastic attention output with the target, yielding an efficient univariate post-hoc tuning problem. We evaluate this mechanism on two scientific foundation models for weather and timeseries forecasting along with an additional regression task. Across benchmarks against uncertainty-aware baselines, we find that Stochastic Attention achieves the strongest native calibration and the sharpest prediction intervals at comparable coverage, while requiring only minutes of post-hoc tuning versus days of retraining for competitive baselines.

This technical note revisits the relationship between RaBitQ and TurboQuant under a unified comparison framework. We compare the two methods in terms of methodology, theoretical guarantees, and empirical performance, using a reproducible, transparent, and symmetric setup. Our results show that, despite the claimed advantage of TurboQuant, TurboQuant does not provide a consistent improvement over RaBitQ in directly comparable settings; in many tested configurations, it performs worse than RaBitQ. We further find that several reported runtime and recall results in the TurboQuant paper could not be reproduced from the released implementation under the stated configuration. Overall, this note clarifies the shared structure and genuine differences between the two lines of work, while documenting reproducibility issues in the experimental results reported by the TurboQuant paper.

Cross-site scripting (XSS) remains a persistent web security vulnerability, especially because obfuscation can change the surface form of a malicious payload while preserving its behavior. These transformations make it difficult for traditional and machine learning-based detection systems to reliably identify attacks. Existing approaches for generating obfuscated payloads often emphasize syntactic diversity, but they do not always ensure that the generated samples remain behaviorally valid. This paper presents a structured pipeline for generating and evaluating obfuscated XSS payloads using large language models (LLMs). The pipeline combines deterministic transformation techniques with LLM-based generation and uses a browser- based runtime evaluation procedure to compare payload behavior in a controlled execution environment. This allows generated samples to be assessed through observable runtime behavior rather than syntactic similarity alone. In the evaluation, an untuned baseline language model achieves a runtime behavior match rate of 0.15, while fine-tuning on behavior-preserving source-target obfuscation pairs improves the match rate to 0.22. Although this represents a measurable improvement, the results show that current LLMs still struggle to generate obfuscations that preserve observed runtime behavior. A downstream classifier evaluation further shows that adding generated payloads does not improve detection performance in this setting, although behavior- filtered generated samples can be incorporated without materially degrading performance. Overall, the study demonstrates both the promise and the limits of applying generative models to adversarial security data generation and emphasizes the importance of runtime behavior checks in improving the quality of generated data for downstream detection systems.

Social deduction games such as Mafia present a unique AI challenge: players must reason under uncertainty, interpret incomplete and intentionally misleading information, evaluate human-like communication, and make strategic elimination decisions. Unlike deterministic board games, success in Mafia depends not on perfect information or brute-force search, but on inference, memory, and adaptability in the presence of deception. This work presents the design and evaluation of Revac-8, an AI agent developed for the Social Deduction track of the MindGames Arena competition, where it achieved first place. The final agent evolved from a simple two-stage reasoning system into a multi-module architecture that integrates memory-based player profiling, social-graph analysis of accusations and defenses, and dynamic tone selection for communication. These results highlight the importance of structured memory and adaptive communication for achieving strong performance in high-stakes social environments.

Depth pruning improves the deployment efficiency of large language models (LLMs) by identifying and removing redundant layers. A widely accepted standard for this identification process is to measure the similarity between layers using cosine distance. However, we find that methods relying solely on this one-dimensional heuristic can exhibit unpredictable performance and even catastrophic collapse across different architectures. To address this issue, we propose SimDiff, a novel layer importance criterion that jointly evaluates layers from two orthogonal perspectives: representational similarity and transformation difference. The difference is quantified using two distinct metrics: MSSD, which is sensitive to outliers and identifies layers that make decisive corrections, and MASD, which robustly measures a layer's average contribution. Extensive experiments on multiple models ranging from 0.5B to 13B parameters demonstrate that SimDiff significantly outperforms state-of-the-art baselines across various pruning ratios. Notably, our method retains over 91% of LLaMA2-7B's performance at a 25% pruning ratio and achieves up to a 1.49x inference speedup when pruning 12 layers on LLaMA3.1-8B. We also show that pruned models can be effectively recovered with minimal fine-tuning.

Decentralized optimization enables multiple devices to learn a global machine learning model while each individual device only has access to its local dataset. By avoiding the need for training data to leave individual users' devices, it enhances privacy and scalability compared to conventional centralized learning, where all data has to be aggregated to a central server. However, decentralized optimization has traditionally been viewed as a necessary compromise, used only when centralized processing is impractical due to communication constraints or data privacy concerns. In this study, we show that decentralization can paradoxically accelerate convergence, outperforming centralized methods in the number of iterations needed to reach optimal solutions. Through examples in logistic regression and neural network training, we demonstrate that distributing data and computation across multiple agents can lead to faster learning than centralized approaches, even when each iteration is assumed to take the same amount of time, whether performed centrally on the full dataset or decentrally on local subsets. This finding challenges longstanding assumptions and reveals decentralization as a strategic advantage, offering new opportunities for more efficient optimization and machine learning.

Generative engines (GEs) are reshaping information access by replacing ranked links with citation-grounded answers, yet current Generative Engine Optimization (GEO) methods optimize each instance in isolation, unable to accumulate or transfer effective strategies across tasks and engines. We reframe GEO as a strategy learning problem and propose MAGEO, a multi-agent framework in which coordinated planning, editing, and fidelity-aware evaluation serve as the execution layer, while validated editing patterns are progressively distilled into reusable, engine-specific optimization skills. To enable controlled assessment, we introduce a Twin Branch Evaluation Protocol for causal attribution of content edits and DSV-CF, a dual-axis metric that unifies semantic visibility with attribution accuracy. We further release MSME-GEO-Bench, a multi-scenario, multi-engine benchmark grounded in real-world queries. Experiments on three mainstream engines show that MAGEO substantially outperforms heuristic baselines in both visibility and citation fidelity, with ablations confirming that engine-specific preference modeling and strategy reuse are central to these gains, suggesting a scalable learning-driven paradigm for trustworthy GEO. Code is available at https://github.com/Wu-beining/MAGEO

The consensus that GCN, GraphSAGE, GAT, and EvolveGCN outperform feature-only baselines on the Elliptic Bitcoin Dataset is widely cited but has not been rigorously stress-tested under a leakage-free evaluation protocol. We perform a seed-matched inductive-versus-transductive comparison and find that this consensus does not hold. Under a strictly inductive protocol, Random Forest on raw features achieves F1 = 0.821 and outperforms all evaluated GNNs, while GraphSAGE reaches F1 = 0.689 +/- 0.017. A paired controlled experiment reveals a 39.5-point F1 gap attributable to training-time exposure to test-period adjacency. Additionally, edge-shuffle ablations show that randomly wired graphs outperform the real transaction graph, indicating that the dataset's topology can be misleading under temporal distribution shift. Hybrid models combining GNN embeddings with raw features provide only marginal gains and remain substantially below feature-only baselines. We release code, checkpoints, and a strict-inductive protocol to enable reproducible, leakage-free evaluation.

This paper introduces \textit{Bangla Key2Text}, a large-scale dataset of $2.6$ million Bangla keyword--text pairs designed for keyword-driven text generation in a low-resource language. The dataset is constructed using a BERT-based keyword extraction pipeline applied to millions of Bangla news texts, transforming raw articles into structured keyword--text pairs suitable for supervised learning. To establish baseline performance on this new benchmark, we fine-tune two sequence-to-sequence models, \texttt{mT5} and \texttt{BanglaT5}, and evaluate them using multiple automatic metrics and human judgments. Experimental results show that task-specific fine-tuning substantially improves keyword-conditioned text generation in Bangla compared to zero-shot large language models. The dataset, trained models, and code are publicly released to support future research in Bangla natural language generation and keyword-to-text generation tasks.

Automatic keyword extraction from academic papers is a key area of interest in natural language processing and information retrieval. Although previous research has mainly focused on utilizing abstract and references for keyword extraction, this paper focuses on the highlights section - a summary describing the key findings and contributions, offering readers a quick overview of the research. Our observations indicate that highlights contain valuable keyword information that can effectively complement the abstract. To investigate the impact of incorporating highlights into unsupervised keyword extraction, we evaluate three input scenarios: using only the abstract, the highlights, and a combination of both. Experiments conducted with four unsupervised models on Computer Science (CS), Library and Information Science (LIS) datasets reveal that integrating the abstract with highlights significantly improves extraction performance. Furthermore, we examine the differences in keyword coverage and content between abstract and highlights, exploring how these variations influence extraction outcomes. The data and code are available at https://github.com/xiangyi-njust/Highlight-KPE.

The rapid adoption of Large Language Models (LLMs) has spurred interest in automated peer review; however, progress is currently stifled by benchmarks that treat reviewing primarily as a rating prediction task. We argue that the utility of a review lies in its textual justification--its arguments, questions, and critique--rather than a scalar score. To address this, we introduce Beyond Rating, a holistic evaluation framework that assesses AI reviewers across five dimensions: Content Faithfulness, Argumentative Alignment, Focus Consistency, Question Constructiveness, and AI-Likelihood. Notably, we propose a Max-Recall strategy to accommodate valid expert disagreement and introduce a curated dataset of paper with high-confidence reviews, rigorously filtered to remove procedural noise. Extensive experiments demonstrate that while traditional n-gram metrics fail to reflect human preferences, our proposed text-centric metrics--particularly the recall of weakness arguments--correlate strongly with rating accuracy. These findings establish that aligning AI critique focus with human experts is a prerequisite for reliable automated scoring, offering a robust standard for future research.

This article introduces two new measures for authorship attribution - Rank-Turbulence Delta and Jensen-Shannon Delta - which generalise Burrows's classical Delta by applying distance functions designed for probabilistic distributions. We first set out the theoretical basis of the measures, contrasting centred and uncentred z-scoring of word-frequency vectors and re-casting the uncentred vectors as probability distributions. Building on this representation, we develop a token-level decomposition that renders every Delta distance numerically interpretable, thereby facilitating close reading and the validation of results. The effectiveness of the methods is assessed on four literary corpora in English, German, French and Russian. The English, German and French datasets are compiled from Project Gutenberg, whereas the Russian benchmark is the SOCIOLIT corpus containing 755 works by 180 authors spanning the eighteenth to the twenty-first centuries. Rank-Turbulence Delta attains attribution accuracy comparable with Cosine Delta; Jensen-Shannon Delta consistently matches or exceeds the performance of canonical Burrows's Delta. Finally, several established attribution algorithms are re-evaluated on the extended SOCIOLIT corpus.

Large language models (LLMs) have achieved substantial advances in logical reasoning, yet they continue to lag behind human-level performance. In-context learning provides a viable solution that boosts the model's performance via prompting its input with expert-curated, in-domain exemplars. However, in many real-world, expertise-scarce domains, such as low-resource scientific disciplines, emerging biomedical subfields, or niche legal jurisdictions, such high-quality in-domain demonstrations are inherently limited or entirely unavailable, thereby constraining the general applicability of these approaches. To mitigate this limitation, recent efforts have explored the retrieval of cross-domain samples as surrogate in-context demonstrations. Nevertheless, the resulting gains remain modest. This is largely attributable to the pronounced domain shift between source and target distributions, which impedes the model's ability to effectively identify and exploit underlying shared structures or latent reasoning patterns. Consequently, when relying solely on raw textual prompting, LLMs struggle to abstract and transfer such cross-domain knowledge in a robust and systematic manner. To address these issues, we propose CoDA, which employs a lightweight adapter to directly intervene in the intermediate hidden states. By combining feature-based distillation of CoT-enriched reference representations with Maximum Mean Discrepancy (MMD) for kernelized distribution matching, our method aligns the latent reasoning representation of the source and target domains. Extensive experimental results on multiple logical reasoning tasks across various model families validate the efficacy of CoDA by significantly outperforming the previous state-of-the-art baselines by a large margin.

Reinforcement learning (RL) for LLM post-training faces a fundamental design choice: whether to use a learned critic as a baseline for policy optimization. Classical theory favors critic-based methods such as PPO for variance reduction, yet critic-free alternatives like GRPO have gained widespread adoption due to their simplicity and competitive performance. We show that in sparse-reward settings, a learned critic can inject estimation noise that exceeds the state signal it captures, increasing rather than reducing advantage variance. By casting baseline selection as a Kalman filtering problem, we unify PPO and GRPO as two extremes of the Kalman gain and prove that explained variance (EV), computable from a single training batch, identifies the exact boundary: positive EV indicates the critic reduces variance, while zero or negative EV signals that it inflates variance. Building on this insight, we propose Explained Variance Policy Optimization (EVPO), which monitors batch-level EV at each training step and adaptively switches between critic-based and batch-mean advantage estimation, provably achieving no greater variance than the better of the two at every step. Across four tasks spanning classical control, agentic interaction, and mathematical reasoning, EVPO consistently outperforms both PPO and GRPO regardless of which fixed baseline is stronger on a given task. Further analysis confirms that the adaptive gating tracks critic maturation over training and that the theoretically derived zero threshold is empirically optimal.

The intonational structure of Seoul Korean has been defined with discrete tonal categories within the Autosegmental-Metrical model of intonational phonology. However, it is challenging to map continuous $F_0$ contours to these invariant categories due to variable $F_0$ realizations in real-world speech. Our paper proposes Dual-Glob, a deep supervised contrastive learning framework to robustly classify fine-grained pitch accent patterns in Seoul Korean. Unlike conventional local predictive models, our approach captures holistic $F_0$ contour shapes by enforcing structural consistency between clean and augmented views in a shared latent space. To this aim, we introduce the first large-scale benchmark dataset, consisting of manually annotated 10,093 Accentual Phrases in Seoul Korean. Experimental results show that our Dual-Glob significantly outperforms strong baseline models with state-of-the-art accuracy (77.75%) and F1-score (51.54%). Therefore, our work supports AM-based intonational phonology using data-driven methodology, showing that deep contrastive learning effectively captures holistic structural features of continuous $F_0$ contours.

Fairness audits of institutional risk models are critical for understanding how deployed machine learning pipelines allocate resources. Drawing on multi-year collaboration with Centennial College, where our prior ethnographic work introduced the ASP-HEI Cycle, we present a replica-based audit of a deployed Early Warning System (EWS), replicating its model using institutional training data and design specifications. We evaluate disparities by gender, age, and residency status across the full pipeline (training data, model predictions, and post-processing) using standard fairness metrics. Our audit reveals systematic misallocation: younger, male, and international students are disproportionately flagged for support, even when many ultimately succeed, while older and female students with comparable dropout risk are under-identified. Post-processing amplifies these disparities by collapsing heterogeneous probabilities into percentile-based risk tiers. This work provides a replicable methodology for auditing institutional ML systems and shows how disparities emerge and compound across stages, highlighting the importance of evaluating construct validity alongside statistical fairness. It contributes one empirical thread to a broader program investigating algorithms, student data, and power in higher education.

Understanding how complex systems respond to perturbations, such as whether they will remain stable or what their most sensitive patterns are, is a fundamental challenge across science and engineering. Traditional stability and receptivity (resolvent) analyses are powerful but rely on known equations and linearization, limiting their use in nonlinear or poorly modeled systems. Here, we introduce a data-driven framework that automatically identifies stability properties and optimal forcing responses from observation data alone, without requiring governing equations. By training a neural network as a dynamics emulator and using automatic differentiation to extract its Jacobian, we can compute eigenmodes and resolvent modes directly from data. We demonstrate the method on both canonical chaotic models and high-dimensional fluid flows, successfully identifying dominant instability modes and input-output structures even in strongly nonlinear regimes. By leveraging a neural network-based emulator, we readily obtain a nonlinear representation of system dynamics while additionally retrieving intricate dynamical patterns that were previously difficult to resolve. This equation-free methodology establishes a broadly applicable tool for analyzing complex, high-dimensional datasets, with immediate relevance to grand challenges in fields such as climate science, neuroscience, and fluid engineering.

More than half of the global population struggles to meet their civil justice needs due to limited legal resources. While Large Language Models (LLMs) have demonstrated impressive reasoning capabilities, significant challenges remain even at the foundational step of legal issue identification. To investigate LLMs' capabilities in this task, we constructed a dataset from 769 real-world Malaysian Contract Act court cases, using GPT-4o to extract facts and generate candidate legal issues, annotated by senior legal experts, which reveals a critical limitation: while LLMs generate diverse issue candidates, their precision remains inadequate (GPT-4o achieves only 62%). To address this gap, we propose LePREC (Legal Professional-inspired Reasoning Elicitation and Classification), a neuro-symbolic framework combining neural generation with structured statistical reasoning. LePREC consists of: (1) a neuro component leverages LLMs to transform legal descriptions into question-answer pairs representing diverse analytical factors, and (2) a symbolic component applies sparse linear models over these discrete features, learning explicit algebraic weights that identify the most informative reasoning factors. Unlike end-to-end neural approaches, LePREC achieves interpretability through transparent feature weighting while maintaining data efficiency through correlation-based statistical classification. Experiments show a 30-40% improvement over advanced LLM baselines, including GPT-4o and Claude, confirming that correlation-based factor-issue analysis offers a more data-efficient solution for relevance decisions.

Formal verification guarantees proof validity but not formalization faithfulness. For natural-language logical reasoning, where models construct axiom systems from scratch without library constraints, this gap between valid proofs and faithful translations is especially acute. We investigate whether frontier models exploit this gap when generating Lean 4 proofs, a behavior we term formalization gaming. We evaluate GPT-5 and DeepSeek-R1 on 303 first-order logic problems (203 from FOLIO, 100 from Multi-LogiEval), comparing unified generation against a two-stage pipeline that separates formalization from proving. Despite compilation rates of 87-99%, we find no evidence of systematic gaming in unified generation: models prefer reporting failure over forcing proofs, even under prompting designed to encourage it. However, unfaithfulness that evades our detection signals may still occur. The two-stage pipeline reveals two distinct modes of unfaithfulness: GPT-5 fabricates axioms during proof generation, a reactive fallback detectable via cross-stage comparison, while DeepSeek-R1 mistranslates premises during formalization, producing internally consistent outputs that evade detection entirely. These findings show that high compilation rates or accuracies should not be equated with faithful reasoning. Code and data are available at https://github.com/koreankiwi99/formalization-gaming.

Long-horizon enterprise agents make high-stakes decisions (loan underwriting, claims adjudication, clinical review, prior authorization) under lossy memory, multi-step reasoning, and binding regulatory constraints. Current evaluation reports a single task-success scalar that conflates distinct failure modes and hides whether an agent is aligned with the standards its deployment environment requires. We propose that long-horizon decision behavior decomposes into four orthogonal alignment axes, each independently measurable and failable: factual precision (FRP), reasoning coherence (RCS), compliance reconstruction (CRR), and calibrated abstention (CAR). CRR is a novel regulatory-grounded axis; CAR is a measurement axis separating coverage from accuracy. We exercise the decomposition on a controlled benchmark (LongHorizon-Bench) covering loan qualification and insurance claims adjudication with deterministic ground-truth construction. Running six memory architectures, we find structure aggregate accuracy cannot see: retrieval collapses on factual precision; schema-anchored architectures pay a scaffolding tax; plain summarization under a fact-preservation prompt is a strong baseline on FRP, RCS, EDA, and CRR; and all six architectures commit on every case, exposing a decisional-alignment axis the field has not targeted. The decomposition also surfaced a pre-registered prediction of our own, that summarization would fail factual recall, which the data reversed at large magnitude, an axis-level reversal aggregate accuracy would have hidden. Institutional alignment (regulatory reconstruction) and decisional alignment (calibrated abstention) are under-represented in the alignment literature and become load-bearing once decisions leave the laboratory. The framework transfers to any regulated decisioning domain via two steps: build a fact schema, and calibrate the CRR auditor prompt.

Batch Normalization (BN) is a cornerstone of deep learning, yet it fundamentally breaks down in micro-batch regimes (e.g., 3D medical imaging) and non-IID Federated Learning. Removing BN from deep architectures, however, often leads to catastrophic training failures such as vanishing gradients and dying channels. We identify that standard activation functions, like Swish and ReLU, exacerbate this instability in BN-free networks due to their non-zero-centered nature, which causes compounding activation mean-shifts as network depth increases. In this technical communication, we propose Zero-Centered Swish (ZC-Swish), a drop-in activation function parameterized to dynamically anchor activation means near zero. Through targeted stress-testing on BN-free convolutional networks at depths 8, 16, and 32, we demonstrate that while standard Swish collapses to near-random performance at depth 16 and beyond, ZC-Swish maintains stable layer-wise activation dynamics and achieves the highest test accuracy at depth 16 (51.5%) with seed 42. ZC-Swish thus provides a robust, parameter-efficient solution for stabilizing deep networks in memory-constrained and privacy-preserving applications where traditional normalization is unviable.

Federated prognostics enable clients (e.g., companies, factories, and production lines) to collaboratively develop a failure time prediction model while keeping each client's data local and confidential. However, traditional federated models often assume homogeneity in the degradation processes across clients, an assumption that may not hold in many industrial settings. To overcome this, this paper proposes a personalized federated prognostic model designed to accommodate clients with heterogeneous degradation processes, allowing them to build tailored prognostic models. The prognostic model iteratively facilitates the underlying pairwise collaborations between clients with similar degradation patterns, which enhances the performance of personalized federated learning. To estimate parameters jointly using decentralized datasets, we develop a federated parameter estimation algorithm based on proximal gradient descent. The proposed approach addresses the limitations of existing federated prognostic models by simultaneously achieving model personalization, preserving data privacy, and providing comprehensive failure time distributions. The superiority of the proposed model is validated through extensive simulation studies and a case study using the turbofan engine degradation dataset from the NASA repository.

As deductive verifiers mature, their potential user base is growing from the initial core developers to other users. To convince external users of the suitability of verifiers, these tools must run reliably out of the box, give meaningful error messages and display correct results. Yet deductive verifiers are large and complex software systems and their own full verification is often out of reach. We therefore need complementary means to provide such guarantees. This paper advocates the use of fuzzing as a practical way to improve the quality and robustness of deductive verifiers. We outline how fuzz testing can be applied to deductive verifiers, and demonstrate the idea with the prototype tool AValAnCHE, which is integrated with the VerCors verifier. We report on our experiments in which AValAnCHE uncovered several issues in VerCors and demonstrate that the approach also works for other deductive verifiers

We present a dual-track pipeline for detecting biblical allusions in literary fiction and apply it to the novels of Cormac McCarthy. A bottom-up embedding track uses inverse document frequency to identify rare vocabulary shared with the King James Bible, embeds occurrences in their local context for sense disambiguation, and passes candidate passage pairs through cascaded LLM review. A top-down register track asks an LLM to read McCarthy's prose undirected to any specific biblical passage for comparison, catching allusions not distinguished by word or phrase rarity. Both tracks are cross-validated by a long-context model that holds entire novels alongside the KJV in a single pass, and every finding is checked against published scholarship. Restricting attention to allusions that carry a textual echo--shared phrasing, reworked vocabulary, or transplanted cadence--and distinguishing literary allusions proper from signposted biblical references (similes naming biblical figures, characters overtly citing scripture), the pipeline surfaces 349 allusions across the corpus. Among a target set of 115 previously documented allusions retrieved through human review of the academic literature, the pipeline independently recovers 62 (54% recall), with recall varying by connection type from 30% (transformed imagery) to 80% (register collisions). We contextualise these results with respect to the value-add from LLMs as assistants to mechanical stylometric analyses, and their potential to facilitate the statistical study of intertextuality in massive literary corpora.

Reasoning language models can solve increasingly complex tasks, but struggle to produce the calibrated confidence estimates necessary for reliable deployment. Existing calibration methods usually depend on labels or repeated sampling at inference time, making them impractical in many settings. We introduce a method for unsupervised confidence calibration of reasoning LLMs when only a single generation is available at inference time. Our approach uses offline sampling on unlabeled data to derive a self-consistency-based proxy target, then distills this signal into a lightweight deployment-time confidence predictor. In a broad evaluation across 5 math and question-answering tasks using 9 reasoning models, our method substantially outperforms baselines, including under distribution shift, and improves downstream performance in selective prediction and simulated downstream decision-making.

Recent work has demonstrated the promise of orchestrating large language models (LLMs) within evolutionary and agentic optimization systems. However, the mechanisms driving these optimization gains remain poorly understood. In this work, we present a large-scale study of LLM-guided evolutionary search, collecting optimization trajectories for 15 LLMs across 8 tasks. Although zero-shot problem-solving ability correlates with final optimization outcomes, it explains only part of the variance: models with similar initial capability often induce dramatically different search trajectories and outcomes. By analyzing these trajectories, we find that strong LLM optimizers behave as local refiners, producing frequent incremental improvements while progressively localizing the search in semantic space. Conversely, weaker optimizers exhibit large semantic drift, with sporadic breakthroughs followed by stagnation. Notably, various measures of solution novelty do not predict final performance; novelty is beneficial only when the search remains sufficiently localized around high-performing regions of the solution space. Our results highlight the importance of trajectory analysis for understanding and improving LLM-based optimization systems and provide actionable insights for their design and training.

Pre-trained machine learning models (PTMs) are commonly provided via Model Hubs (e.g., Hugging Face) in standard formats like Pickles to facilitate accessibility and reuse. However, this ML supply chain setting is susceptible to malicious attacks that are capable of executing arbitrary code on trusted user environments, e.g., during model loading. To detect malicious PTMs, state-of-the-art detectors (e.g., PickleScan) rely on rules, heuristics, or static analysis, but ignore runtime model behaviors. Consequently, they either miss malicious models due to under-approximation (blacklisting) or miscategorize benign models due to over-approximation (static analysis or whitelisting). To address this challenge, we propose a novel technique (DynaHug) which detects malicious PTMs by learning the behavior of benign PTMs using dynamic analysis and machine learning (ML). DynaHug trains an ML classifier (one-class SVM (OCSVM)) on the runtime behaviours of task-specific benign models. We evaluate DynaHug using over 25,000 benign and malicious PTMs from different sources including Hugging Face and MalHug. We also compare DynaHug to several state-of-the-art detectors including static, dynamic and LLM-based detectors. Results show that DynaHug is up to 44% more effective than existing baselines in terms of F1-score. Our ablation study demonstrates that our design decisions (dynamic analysis, OCSVM, clustering) contribute positively to DynaHug's effectiveness.

Generative AI systems are known to amplify biases present in their training data. While several inference-time mitigation strategies have been proposed, they remain largely empirical and lack formal guarantees. In this paper we introduce CTLF, a branching-time logic designed to reason about bias in series of generative AI outputs. CTLF adopts a counting worlds semantics where each world represents a possible output at a given step in the generation process and introduces modal operators that allow us to verify whether the current output series respects an intended probability distribution over a protected attribute, to predict the likelihood of remaining within acceptable bounds as new outputs are generated, and to determine how many outputs are needed to remove in order to restore fairness. We illustrate the framework on a toy example of biased image generation, showing how CTLF formulas can express concrete fairness properties at different points in the output series.

Sequential Recommendation (SR) aims to predict the next interaction of a user based on their behavior sequence, where complementary relations often provide essential signals for predicting the next item. However, mainstream models relying on sparse co-purchase statistics often mistake spurious correlations (e.g., due to popularity bias) for true complementary relations. Identifying true complementary relations requires capturing the fine-grained item semantics (e.g., specifications) that simple cooccurrence statistics would be unable to model. While recent semantics-based methods utilize discrete semantic codes to represent items, they typically aggregate semantic codes into coarse item representations. This aggregation process blurs specific semantic details required to identify complementarity. To address these critical limitations and effectively leverage semantics for capturing reliable complementary relations, we propose a Complementary-Aware Semantic Transition (CAST) framework that introduces a new modeling paradigm built upon semantic-level transitions. Specifically, a semantic-level transition module is designed to model dynamic transitions directly in the discrete semantic code space, effectively capturing fine-grained semantic dependencies often lost in aggregated item representations. Then, a complementary prior injection module is designed to incorporate LLM-verified complementary priors into the attention mechanism, thereby prioritizing complementary patterns over co-occurrence statistics. Experiments on multiple e-commerce datasets demonstrate that CAST consistently outperforms the state-of-the-art approaches, achieving up to 17.6% Recall and 16.0% NDCG gains with 65x training acceleration. This validates its effectiveness and efficiency in uncovering latent item complementarity beyond statistics. The code will be released upon acceptance.

Large vision-language models (LVLMs) frequently suffer from Object Hallucination (OH), wherein they generate descriptions containing objects that are not actually present in the input image. This phenomenon is particularly problematic in real-world applications such as medical imaging and autonomous driving, where accuracy is critical. Recent studies suggest that the hallucination problem may stem from language priors: biases learned during pretraining that cause LVLMs to generate words based on their statistical co-occurrence. To mitigate this problem, we propose Visual Contrastive Editing (VCE), a novel post-hoc method that identifies and suppresses hallucinatory tendencies by analyzing the model's response to contrastive visual perturbations. Using Singular Value Decomposition (SVD), we decompose the model's activation patterns to isolate hallucination subspaces and apply targeted parameter edits to attenuate its influence. Unlike existing approaches that require fine-tuning or labeled data, VCE operates as a label-free intervention, making it both scalable and practical for deployment in resource-constrained settings. Experimental results demonstrate that VCE effectively reduces object hallucination across multiple benchmarks while maintaining the model's original computational efficiency.

Understanding road scenes in a geometrically consistent, scene-centric representation is crucial for planning and mapping. We present GOLD-BEV, a framework that learns dense bird's-eye-view (BEV) semantic environment maps-including dynamic agents-from ego-centric sensors, using time-synchronized aerial imagery as supervision only during training. BEV-aligned aerial crops provide an intuitive target space, enabling dense semantic annotation with minimal manual effort and avoiding the ambiguity of ego-only BEV labeling. Crucially, strict aerial-ground synchronization allows overhead observations to supervise moving traffic participants and mitigates the temporal inconsistencies inherent to non-synchronized overhead sources. To obtain scalable dense targets, we generate BEV pseudo-labels using domain-adapted aerial teachers, and jointly train BEV segmentation with optional pseudo-aerial BEV reconstruction for interpretability. Finally, we extend beyond aerial coverage by learning to synthesize pseudo-aerial BEV images from ego sensors, which support lightweight human annotation and uncertainty-aware pseudo-labeling on unlabeled drives.

Common image editing tasks typically adopt powerful generative diffusion models as the leading paradigm for real-world content editing. Meanwhile, although reinforcement learning (RL) methods such as Diffusion-DPO and Flow-GRPO have further improved generation quality, efficiently applying Reinforcement Learning from Human Feedback (RLHF) to diffusion-based editing remains largely unexplored, due to a lack of scalable human-preference datasets and frameworks tailored to diverse editing needs. To fill this gap, we propose HP-Edit, a post-training framework for Human Preference-aligned Editing, and introduce RealPref-50K, a real-world dataset across eight common tasks and balancing common object editing. Specifically, HP-Edit leverages a small amount of human-preference scoring data and a pretrained visual large language model (VLM) to develop HP-Scorer--an automatic, human preference-aligned evaluator. We then use HP-Scorer both to efficiently build a scalable preference dataset and to serve as the reward function for post-training the editing model. We also introduce RealPref-Bench, a benchmark for evaluating real-world editing performance. Extensive experiments demonstrate that our approach significantly enhances models such as Qwen-Image-Edit-2509, aligning their outputs more closely with human preference.

Automatic evaluators such as reward models play a central role in the alignment and evaluation of large vision-language models (LVLMs). Despite their growing importance, these evaluators are almost exclusively assessed on English-centric benchmarks, leaving open the question of how well these evaluators generalize across languages. To answer this question, we introduce MM-JudgeBench, the first large-scale benchmark for multilingual and multimodal judge model evaluation, which includes over 60K pairwise preference instances spanning 25 typologically diverse languages. MM-JudgeBench integrates two complementary subsets: a general vision-language preference evaluation subset extending VL-RewardBench, and a chart-centric visual-text reasoning subset derived from OpenCQA, enabling systematic analysis of reward models (i.e., LVLM judges) across diverse settings. We additionally release a multilingual training set derived from MM-RewardBench, disjoint from our evaluation data, to support domain adaptation. By evaluating 22 LVLMs (15 open-source, 7 proprietary), we uncover substantial cross-lingual performance variance in our proposed benchmark. Our analysis further shows that model size and architecture are poor predictors of multilingual robustness, and that even state-of-the-art LVLM judges exhibit inconsistent behavior across languages. Together, these findings expose fundamental limitations of current reward modeling and underscore the necessity of multilingual, multimodal benchmarks for developing reliable automated evaluators.

Traditional policy learning methods in cooperative pursuit face fundamental challenges in biomimetic underwater robots, where long-horizon decision making, partial observability, and inter-robot coordination require both expressiveness and stability. To address these issues, a novel framework called Mamba-based multi-agent group relative policy optimization (M$^{2}$GRPO) is proposed, which integrates a selective state-space Mamba policy with group-relative policy optimization under the centralized-training and decentralized-execution (CTDE) paradigm. Specifically, the Mamba-based policy leverages observation history to capture long-horizon temporal dependencies and exploits attention-based relational features to encode inter-agent interactions, producing bounded continuous actions through normalized Gaussian sampling. To further improve credit assignment without sacrificing stability, the group-relative advantages are obtained by normalizing rewards across agents within each episode and optimized through a multi-agent extension of GRPO, significantly reducing the demand for training resources while enabling stable and scalable policy updates. Extensive simulations and real-world pool experiments across team scales and evader strategies demonstrate that M$^{2}$GRPO consistently outperforms MAPPO and recurrent baselines in both pursuit success rate and capture efficiency. Overall, the proposed framework provides a practical and scalable solution for cooperative underwater pursuit with biomimetic robot systems.

Knowledge Graph Embeddings (KGEs) support a wide range of downstream tasks over Knowledge Graphs (KGs). In practice, KGs evolve as new entities and facts are added, motivating Continual Knowledge Graph Embedding (CKGE) methods that update embeddings over time. Current CKGE approaches address catastrophic forgetting (i.e., the performance degradation on previously learned tasks) primarily by limiting changes to existing embeddings. However, we show that this view is incomplete. When new entities are introduced, their embeddings can interfere with previously learned ones, causing the model to predict them in place of previously correct answers. This phenomenon, which we call entity interference, has been largely overlooked and is not accounted for in current CKGE evaluation protocols. As a result, the assessment of catastrophic forgetting becomes misleading, and CKGE methods performance is systematically overestimated. To address this issue, we introduce a corrected CKGE evaluation protocol that accounts for entity interference. Through experiments on multiple benchmarks, we show that ignoring this effect can lead to performance overestimation of up to 25%, particularly in scenarios with significant entity growth. We further analyze how different CKGE methods and KGE models are affected by the different sources of forgetting, and introduce a catastrophic forgetting metric tailored to CKGE.

Maintaining consistency between code and documentation is a crucial yet frequently overlooked aspect of software development. Even minor mismatches can confuse API users, introduce new bugs, and increase overall maintenance effort. This creates demand for automated solutions that can assist developers in identifying code-documentation inconsistencies. However, since automatic reports still require human confirmation, false positives carry serious consequences: wasting developer time and discouraging practical adoption. We introduce CASCADE (Consistency Analysis for Source Code And Documentation through Execution), a novel tool for detecting inconsistencies with a strong emphasis on reducing false positives. CASCADE leverages Large Language Models (LLMs) to generate unit tests directly from natural-language documentation. Since these tests are derived from the documentation, any failure during execution indicates a potential mismatch between the documented and actual behavior of the code. To minimize false positives, CASCADE also generates code from the documentation to cross-check the generated tests. By design, an inconsistency is reported only when two conditions are met: the existing code fails a test, while the code generated from the documentation passes the same test. We evaluated CASCADE on a novel dataset of 71 inconsistent and 814 consistent code-documentation pairs drawn from open-source Java projects. Further, we applied CASCADE to additional Java, C#, and Rust repositories, where we uncovered 13 previously unknown inconsistencies, of which 10 have subsequently been fixed, demonstrating both CASCADE's precision and its applicability to real-world codebases.

Federated learning (FL) is a key paradigm for distributed model learning across decentralized data sources. Communication in each FL round typically consists of two phases: (i) distributing the global model from a server to clients, and (ii) collecting updated local models from clients to the server for aggregation. This paper focuses on a type of FL where communication between a client and the server is relay-based over dynamic networks, making routing optimization essential. A typical scenario is in-orbit FL, where satellites act as clients and communicate with a server (which can be a satellite, ground station, or aerial platform) via multi-hop inter-satellite links. This paper presents a comprehensive tractability analysis of routing optimization for in-orbit FL under different settings. For global model distribution, these include the number of models, the objective function, and routing schemes (unicast versus multicast, and splittable versus unsplittable flow). For local model collection, the settings consider the number of models, client selection, and flow splittability. For each case, we rigorously prove whether the global optimum is obtainable in polynomial time or the problem is NP-hard. Together, our analysis draws clear boundaries between tractable and intractable regimes for a broad spectrum of routing problems for in-orbit FL. For tractable cases, the derived efficient algorithms are directly applicable in practice. For intractable cases, we provide fundamental insights into their inherent complexity. These contributions fill a critical yet unexplored research gap, laying a foundation for principled routing design, evaluation, and deployment in satellite-based FL or similar distributed learning systems.

Large language models (LLMs) are expensive to serve because model parameters, attention computation, and KV caches impose substantial memory and latency costs. We present GRASPrune, a structured pruning framework applied after pretraining that jointly prunes FFN channels and KV head groups under a single global budget. Instead of learning importance scores without constraints and applying the budget only after training, GRASPrune learns lightweight gate scores with a projected straight-through estimator that enforces a hard mask satisfying the budget at every step while keeping the backbone weights frozen. After the mask is fixed, we calibrate scaling factors on the retained units to mitigate scale mismatch caused by pruning, and fold these factors into the pruned weights to obtain a smaller dense checkpoint with no extra parameters at inference. On LLaMA-2-7B, GRASPrune removes 50% of parameters and achieves 12.18 perplexity on WikiText-2 while maintaining competitive average zero-shot accuracy on five benchmarks, using four epochs on 512 unlabeled calibration sequences on a single NVIDIA A100 80GB GPU without any full model fine-tuning.

While self-consistency is known to improve performance on symbolic reasoning, its effect on the recall of encyclopedic knowledge is unclear due to a lack of targeted evaluation grounds. To address this, we establish such a knowledge recall split for the popular MMLU benchmark by applying a data-driven heuristic from prior work. We validate this split by showing that the performance patterns on the symbolic reasoning and knowledge recall subsets mirror those of GSM8K and MedMCQA, respectively. Using this solid ground, we find that self-consistency consistently improves performance across both symbolic reasoning and knowledge recall, even though its underlying CoT prompting is primarily effective for symbolic reasoning. As a result, we achieve an 89\% accuracy on MMLU, the best performance to date with the use of GPT-4o.

This paper narrows the performance gap between small, specialized models and significantly larger general-purpose models through domain adaptation via continual pre-training and merging. We address the scarcity of specialized non-English data by constructing a high-quality German medical corpus (FineMed-de) from FineWeb2. This corpus is used to continually pre-train and merge three well-known LLMs (ranging from $7B$ to $24B$ parameters), creating the DeFineMed model family. A comprehensive evaluation confirms that specialization dramatically enhances $7B$ model performance on German medical benchmarks. Furthermore, the pairwise win-rate analysis of the Qwen2.5-based models demonstrates an approximately $3.5$-fold increase in the win-rate against the much larger Mistral-Small-24B-Instruct through domain adaptation. This evidence positions specialized $7B$ models as a competitive, resource-efficient solution for complex medical instruction-following tasks. While model merging successfully restores instruction-following abilities, a subsequent failure mode analysis reveals inherent trade-offs, including the introduction of language mixing and increased verbosity, highlighting the need for more targeted fine-tuning in future work. This research provides a robust, compliant methodology for developing specialized LLMs, serving as the foundation for practical use in German-speaking healthcare contexts.

This paper presents a framework to bridge the gap between subjective stakeholder context and formal system architecture. This is achieved using Soft Systems Methodology (SSM) and Systems Modelling Language version 2 (SysML v2). The methodology utilises the precision of Kernel Modelling Language (KerML) and the alignment of SysML v2 with ISO 42010 to define a reference architecture for the mapping of SSM outputs to SysML v2 concepts such as stakeholders and concerns. Application of the framework is demonstrated through the use of a case study, highlighting the traceable path from stakeholder context to system architecture. The structured mapping and increased semantic precision of SysML v2 are anticipated to reduce the risk of misinterpretation compared to less formal approaches, though empirical validation across diverse stakeholder contexts remains as future work. The primary identified trade-off is the increased barrier to entry associated with SysML v2's textual notation.

Metal-organic frameworks (MOFs) are a major target of machine-learning-based property prediction, yet most models assume that a single framework representation maps to a single property value. This assumption becomes problematic for experimental MOFs, where samples reported as the same framework can exhibit different properties because of differences in crystallinity, phase purity, defects, and other sample-dependent factors. Here we introduce Experimental X-ray Diffraction Integrated Transformer (EXIT), a multimodal transformer for sample-aware prediction of MOF properties that combines MOFid with X-ray diffraction (XRD). In EXIT, MOFid encodes MOF identity, whereas XRD provides complementary information about the experimentally realized sample state. EXIT is pre-trained on one million hypothetical MOFs with simulated XRD to learn transferable representations, leading to improved downstream performance relative to existing approaches. EXIT is fine-tuned on literature-derived experimental datasets for surface area and pore volume prediction. Incorporating experimental XRD improves predictive performance relative to models without experimental XRD, and attention analysis and sample-level case studies further show that EXIT assigns different predictions to samples sharing the same MOF identity when their XRD patterns differ. These results establish a practical step from framework-aware to sample-aware MOF property prediction and highlight the value of incorporating experimental characterization into porous materials informatics.

The emergence of sixth-generation (6G) technologies has introduced new challenges and opportunities for machine learning (ML) applications in Internet of Things (IoT) networks, particularly concerning energy efficiency. As model training and data transmission contribute significantly to energy consumption, optimizing these processes has become critical for sustainable system design. This study first conduct analysis on the energy consumption model for both centralized and decentralized architecture and then presents a testbed deployed within the German railway infrastructure, leveraging sensor data for ML-based predictive maintenance. A comparative analysis of distributed versus Centralized Learning (CL) architectures reveals that distributed models maintain competitive predictive accuracy (~90%) while reducing overall electricity consumption by up to 70%. These findings underscore the potential of distributed ML to improve energy efficiency in real-world IoT deployments, particularly by mitigating transmission-related energy costs.

Green software engineering is emerging as a crucial response to information technology's rising energy impact, especially in continuous development. However, there remain challenges in devising automated methods for identifying energy regressions across commits and their associated code change patterns. In particular, little effort has been put into automatically detecting regressions at the commit level by identifying statistically significant changes in energy consumption. In this paper, we introduce EnergyTrackr, an approach designed to detect energy regressions across multiple commits that can then be used to identify code anti-patterns potentially contributing to the increase of software energy consumption over time. We describe our empirical evaluation, including repository mining and source code analysis, made on 3,232 commits from three Java projects, and show the approach's ability to identify significant energy changes. We also highlight recurring anti-patterns such as missing early exits or costly dependency upgrades. We expect EnergyTrackr to assist developers in accurately monitoring energy regressions and improvements within their projects, identifying code anti-patterns, and helping them optimize their source code to reduce software energy consumption.

Graph representation learning has achieved notable success in encoding graph-structured data into latent vector spaces, enabling a wide range of downstream tasks. However, these node representations remain opaque and difficult to interpret. Existing explainability methods primarily focus on supervised settings or on explaining individual representation dimensions, leaving a critical gap in explaining the overall structure of node representations. In this paper, we propose TACENR (Task-Agnostic Contrastive Explanations for Node Representations), a local explanation method that identifies not only attribute features but also proximity and structural ones that contribute the most in the representation space. TACENR builds on contrastive learning, through which we learn a similarity function in the representation space, revealing which are the features that play an important role in the representation of a node. While our focus is on task-agnostic explanations, TACENR can be applied to supervised scenarios as well. Experimental results demonstrate that proximity and structural features play a significant role in shaping node representations and that our supervised variant performs comparably to existing task-specific approaches in identifying the most impactful features.

Predicting driver intention from neurophysiological signals offers a promising pathway for enhancing proactive safety in advanced driver assistance systems, yet remains challenging in real-world driving due to EEG signal non-stationarity and the complexity of cognitive-motor preparation. This study proposes and evaluates an EEG-based driver intention prediction framework using a synchronised multi-sensor platform integrated into a real electric vehicle. A real-world on-road dataset was collected across 32 driving sessions, and 12 deep learning architectures were evaluated under consistent experimental conditions. Among the evaluated architectures, TSCeption achieved the highest average accuracy (0.907) and Macro-F1 score (0.901). The proposed framework demonstrates strong temporal stability, maintaining robust decoding performance up to 1000 ms before manoeuvre execution with minimal degradation. Furthermore, additional analyses reveal that minimal EEG preprocessing outperforms artefact-handling pipelines, and prediction performance peaks within a 400-600 ms interval, corresponding to a critical neural preparatory phase preceding driving manoeuvres. Overall, these findings support the feasibility of early and stable EEG-based driver intention decoding under real-world on-road conditions. Code: https://github.com/galosaimi/Mind2Drive.

The evaluation of machine learning models typically relies mainly on performance metrics based on loss functions, which risk to overlook changes in performance in relevant subgroups. Auditing tools such as SliceFinder and SliceLine were proposed to detect such groups, but usually have conceptual disadvantages, such as the inability to directly address continuous covariates. In this paper, we introduce FairTree, a novel algorithm adapted from psychometric invariance testing. Unlike SliceFinder and related algorithms, FairTree directly handles continuous, categorical, and ordinal features without discretization. It further decomposes performance disparities into systematic bias and variance, allowing a categorization of changes in algorithm performance. We propose and evaluate two variations of the algorithm: a permutation-based approach, which is conceptually closer to SliceFinder, and a fluctuation test. Through simulation studies that include a direct comparison with SliceLine, we demonstrate that both approaches have a satisfactory rate of false-positive results, but that the fluctuation approach has relatively higher power. We further illustrate the method on the UCI Adult Census dataset. The proposed algorithms provide a flexible framework for the statistical evaluation of the performance and aspects of fairness of machine learning models in a wide range of applications even in relatively small data.

High-fidelity measurements of continuum physical fields are essential for scientific discovery and engineering design but remain challenging under sparse and constrained sensing. Conventional reconstruction methods typically rely on fixed sensor layouts, which cannot adapt to evolving physical states. We propose LASER, a unified, closed-loop framework that formulates active sensing as a Partially Observable Markov Decision Process (POMDP). At its core, LASER employs a continuum field latent world model that captures the underlying physical dynamics and provides intrinsic reward feedback. This enables a reinforcement learning policy to simulate ''what-if'' sensing scenarios within a latent imagination space. By conditioning sensor movements on predicted latent states, LASER navigates toward potentially high-information regions beyond current observations. Our experiments demonstrate that LASER consistently outperforms static and offline-optimized strategies, achieving high-fidelity reconstruction under sparsity across diverse continuum fields.

Large Language Model (LLM) agents are increasingly proposed for autonomous cybersecurity tasks, but their capabilities in realistic offensive settings remain poorly understood. We present DeepRed, an open-source benchmark for evaluating LLM-based agents on realistic Capture The Flag (CTF) challenges in isolated virtualized environments. DeepRed places an agent in a Kali attacker environment with terminal tools and optional web search, connected over a private network to a target challenge, and records full execution traces for analysis. To move beyond binary solved/unsolved outcomes, we introduce a partial-credit scoring method based on challenge-specific checkpoints derived from public writeups, together with an automated summarise-then-judge labelling pipeline for assigning checkpoint completion from logs. Using DeepRed, we benchmark ten commercially accessible LLMs on ten VM-based CTF challenges spanning different challenge categories. The results indicate that current agents remain limited: the best model achieves only 35% average checkpoint completion, performing strongest on common challenge types and weakest on tasks requiring non-standard discovery and longer-horizon adaptation.

The quadratic computational complexity of the standard attention mechanism constitutes a fundamental bottleneck for large language models in long-context inference. While existing KV cache compression methods alleviate memory pressure, they often sacrifice generation quality and fail to address the high overhead of floating-point arithmetic. This paper introduces DASH-KV, an innovative acceleration framework that reformulates attention as approximate nearest-neighbor search via asymmetric deep hashing. Under this paradigm, we design an asymmetric encoding architecture that differentially maps queries and keys to account for their distinctions in precision and reuse characteristics. To balance efficiency and accuracy, we further introduce a dynamic mixed-precision mechanism that adaptively retains full-precision computation for critical tokens. Extensive experiments on LongBench demonstrate that DASH-KV significantly outperforms state-of-the-art baseline methods while matching the performance of full attention, all while reducing inference complexity from O(N^2) to linear O(N). The code is available at https://github.com/Zhihan-Zh/DASH-KV

Memristive devices present a promising foundation for next-generation information processing by combining memory and computation within a single physical substrate. This unique characteristic enables efficient, fast, and adaptive computing, particularly well suited for deep learning applications. Among recent developments, the memristive-friendly echo state network (MF-ESN) has emerged as a promising approach that combines memristive-inspired dynamics with the training simplicity of reservoir computing, where only the readout layer is learned. Building on this framework, we propose memristive-friendly parallelized reservoirs (MARS), a simplified yet more effective architecture that enables efficient scalable parallel computation and deeper model composition through novel subtractive skip connections. This design yields two key advantages: substantial training speedups of up to 21x over the inherently lightweight echo state network baseline and significantly improved predictive performance. Moreover, MARS demonstrates what is possible with parallel memristive-friendly reservoir computing: on several long sequence benchmarks our compact gradient-free models substantially outperform strong gradient-based sequence models such as LRU, S5, and Mamba, while reducing full training time from minutes or hours down seconds or even only a few hundred milliseconds. Our work positions parallel memristive-friendly computing as a promising route towards scalable neuromorphic learning systems that combine high predictive capability with radically improved computational efficiency, while providing a clear pathway to energy-efficient, low-latency implementations on emerging memristive and in-memory hardware.

Generative AI-powered by Large Language Models (LLMs)-is increasingly deployed in industry across healthcare decision support, financial analytics, enterprise retrieval, and conversational automation, where reliability, efficiency, and cost control are critical. In such settings, models must satisfy strict constraints on energy, latency, and hardware utilization-not accuracy alone. Yet prevailing evaluation pipelines remain accuracy-centric, creating a Deployment-Evaluation Gap-the absence of operational and economic criteria in model assessment. To address this gap, we present EDGE-EVAL-a industry-oriented benchmarking framework that evaluates LLMs across their full lifecycle on legacy NVIDIA Tesla T4 GPUs. Benchmarking LLaMA and Qwen variants across three industrial tasks, we introduce five deployment metrics-Economic Break-Even (Nbreak), Intelligence-Per-Watt (IPW ), System Density (\r{ho}sys), Cold-Start Tax (Ctax), and Quantization Fidelity (Qret)-capturing profitability, energy efficiency, hardware scaling, serverless feasibility, and compression safety. Our results reveal a clear efficiency frontier-models in the <2B parameter class dominate larger baselines across economic and ecological dimensions. LLaMA-3.2-1B (INT4) achieves ROI break-even in 14 requests (median), delivers 3x higher energy-normalized intelligence than 7B models, and exceeds 6,900 tokens/s/GB under 4-bit quantization. We further uncover an efficiency anomaly-while QLoRA reduces memory footprint, it increases adaptation energy by up to 7x for small models-challenging prevailing assumptions about quantization-aware training in edge deployment.

Language models are increasingly used in scientific discovery to generate hypotheses, propose candidate solutions, implement systems, and iteratively refine them. At the core of these trial-and-error loops lies evaluation: the process of obtaining feedback on candidate solutions via verifiers, simulators, or task-specific scoring functions. While prior work has highlighted the importance of evaluation, it has not explicitly formulated the problem of how evaluation-driven discovery loops can be scaled up in a principled and effective manner to push the boundaries of scientific discovery, a problem this paper seeks to address. We introduce Simple Test-time Evaluation-driven Scaling (SimpleTES), a general framework that strategically combines parallel exploration, feedback-driven refinement, and local selection, revealing substantial gains unlocked by scaling evaluation-driven discovery loops along the right dimensions. Across 21 scientific problems spanning six domains, SimpleTES discovers state-of-the-art solutions using gpt-oss models, consistently outperforming both frontier-model baselines and sophisticated optimization pipelines. Particularly, we sped up the widely used LASSO algorithm by over 2x, designed quantum circuit routing policies that reduce gate overhead by 24.5%, and discovered new Erdos minimum overlap constructions that surpass the best-known results. Beyond novel discoveries, SimpleTES produces trajectory-level histories that naturally supervise feedback-driven learning. When post-trained on successful trajectories, models not only improve efficiency on seen problems but also generalize to unseen problems, discovering solutions that base models fail to uncover. Together, our results establish effective evaluation-driven loop scaling as a central axis for advancing LLM-driven scientific discovery, and provide a simple yet practical framework for realizing these gains.

This study aims to develop and improve machine learning-based post-processing models for precipitation, temperature, and wind speed predictions using the Mesoscale Model (MSM) dataset provided by the Japan Meteorological Agency (JMA) for 18 locations across Japan, including plains, mountainous regions, and islands. By incorporating meteorological variables from grid points surrounding the target locations as input features and applying feature selection based on correlation analysis, we found that, in our experimental setting, the LightGBM-based models achieved lower RMSE than the specific neural-network baselines tested in this study, including a reproduced CNN baseline, and also generally achieved lower RMSE than both the raw MSM forecasts and the JMA post-processing product, MSM Guidance (MSMG), across many locations and forecast lead times. Because precipitation has a highly skewed distribution with many zero cases, we additionally examined Tweedie-based loss functions and event-weighted training strategies for precipitation forecasting. These improved event-oriented performance relative to the original LightGBM model, especially at higher rainfall thresholds, although the gains were site dependent and overall performance remained slightly below MSMG.

Online federated learning (OFL) has emerged as a popular framework for decentralized decision-making over continuous data streams without compromising client privacy. However, the adversary model assumed in standard OFL typically precludes any potential benefits of parallelization. Further, it fails to adequately capture the different sources of statistical variation in OFL problems. In this paper, we extend the OFL paradigm by integrating a stochastically extended adversary (SEA). Under this framework, the loss function remains fixed across clients over time. However, the adversary dynamically and independently selects the data distribution for each client at each time. We propose the \algoOFL{} algorithm to solve this problem, which utilizes online stochastic gradient descent at the clients, along with periodic global aggregation via the server. We establish bounds on the global network regret over a time horizon \(T\) for two classes of functions: (1) for smooth and convex losses, we prove an \(\mathcal{O}(\sqrt{T})\) bound, and (2) for smooth and strongly convex losses, we prove an \(\mathcal{O}(\log T)\) bound. Through careful analysis, we quantify the individual impact of both spatial (across clients) and temporal (over time) data heterogeneity on the regret bounds. Consequently, we identify a regime of mild temporal variation (relative to stochastic gradient variance), where the network regret improves with parallelization. Hence, in the SEA setting, our results improve the existing pessimistic worst-case results in online federated learning.

The rapid growth of chemical literature has generated vast amounts of unstructured data, where reaction information is particularly valuable for applications such as reaction predictions and drug design. However, the prohibitive cost of expert annotation has led to a scarcity of training data, severely hindering the performance of automatic reaction extraction. In this work, we conduct a systematic study of active learning for chemical reaction extraction. We integrate six uncertainty- and diversity-based strategies with pretrained transformer-CRF architectures, and evaluate them on product extraction and role labeling task. While several methods approach full-data performance with fewer labeled instances, learning curves are often non-monotonic and task-dependent. Our analysis shows that strong pretraining, structured CRF decoding, and label sparsity limit the stability of conventional active learning strategies. These findings provide practical insights for the effective use of active learning in chemical information extraction.

Understanding how policy is debated and justified in parliament is a fundamental aspect of the democratic process. However, the volume and complexity of such debates mean that outside audiences struggle to engage. Meanwhile, Large Language Models (LLMs) have been shown to enable automated summarisation at scale. While summaries of debates can make parliamentary procedures more accessible, evaluating whether these summaries faithfully communicate argumentative content remains challenging. Existing automated summarisation metrics have been shown to correlate poorly with human judgements of consistency (i.e., faithfulness or alignment between summary and source). In this work, we propose a formal framework for evaluating parliamentary debate summaries that grounds argument structures in the contested proposals up for debate. Our novel approach, driven by computational argumentation, focuses the evaluation on formal properties concerning the faithful preservation of the reasoning presented to justify or oppose policy outcomes. We demonstrate our methods using a case-study of debates from the European Parliament and associated LLM-driven summaries.

We introduce PLaMo 2.1-VL, a lightweight Vision Language Model (VLM) for autonomous devices, available in 8B and 2B variants and designed for local and edge deployment with Japanese-language operation. Focusing on Visual Question Answering (VQA) and Visual Grounding as its core capabilities, we develop and evaluate the models for two real-world application scenarios: factory task analysis via tool recognition, and infrastructure anomaly detection. We also develop a large-scale synthetic data generation pipeline and comprehensive Japanese training and evaluation resources. PLaMo 2.1-VL outperforms comparable open models on Japanese and English benchmarks, achieving 61.5 ROUGE-L on JA-VG-VQA-500 and 85.2% accuracy on Japanese Ref-L4. For the two application scenarios, it achieves 53.9% zero-shot accuracy on factory task analysis, and fine-tuning on power plant data improves anomaly detection bbox + label F1-score from 39.7 to 64.9.

Concept Bottleneck Models (CBMs) route predictions exclusively through a clinically grounded concept layer, binding interpretability to concept-label consistency. When a dataset contains concept-level inconsistencies, identical concept profiles mapped to conflicting diagnosis labels create an unresolvable bottleneck that imposes a hard ceiling on achievable accuracy. In this paper, we apply rough set theory to the Derm7pt dermoscopy benchmark and characterize the full extent and clinical structure of this inconsistency. Among 305 unique concept profiles formed by the 7 dermoscopic criteria of the 7-point melanoma checklist, 50 (16.4%) are inconsistent, spanning 306 images (30.3% of the dataset). This yields a theoretical accuracy ceiling of 92.1%, independent of backbone architecture or training strategy for CBMs that exclusively operate with hard concepts. In addition, we characterize the conflict-severity distribution, identify the clinical features most responsible for boundary ambiguity, and evaluate two filtering strategies with quantified effects on dataset composition and CBM interpretability. Symmetric removal of all boundary-region images yields Derm7pt+, a fully consistent benchmark subset of 705 images with perfect quality of classification and no hard accuracy ceiling. Building on this filtered dataset, we present a hard CBM evaluated across 19 backbone architectures from the EfficientNet, DenseNet, ResNet, and Wide ResNet families. Under symmetric filtering, explored for completeness, EfficientNet-B5 achieves the best label F1 score (0.85) and label accuracy (0.90) on the held-out test set, with a concept accuracy of 0.70. Under asymmetric filtering, EfficientNet-B7 leads across all four metrics, reaching a label F1 score of 0.82 and concept accuracy of 0.70. These results establish reproducible baselines for concept-consistent CBM evaluation on dermoscopic data.

Fine-tuning Large Language Models (LLMs) remains structurally uncertain despite parameter-efficient methods such as Low-Rank Adaptation (LoRA), as the layer-specific roles of internal representations are poorly understood, leading to heuristic decisions about where adaptation should be applied. We model the evolution of hidden states as a high-dimensional geometric trajectory and propose using the Ramer-Douglas-Peucker (RDP) algorithm, a parameter-free and training-free polygon simplification method that preserves global structural transitions while eliminating locally redundant changes, to identify critical breakpoints along the representation path. Crucially, we use these geometric pivots not merely for analysis, but as a direct decision signal for determining which layers should be adapted during parameter-efficient fine-tuning. By integrating this geometry-aware layer selection strategy into LoRA fine-tuning of Qwen3-8B-Base, we achieve superior performance on MMLU-Math using only 13 RDP-selected layers (81.67%), significantly outperforming both full 36-layer adaptation (79.32%) and random 13-layer selection (75.56%), as well as the baseline Qwen3-8B-Base model (74.25%). These results demonstrate that leveraging the intrinsic geometry of representation trajectories provides a robust, interpretable, and training-free signal for optimizing layer selection during model adaptation.

Large Language Models (LLMs) have recently shown strong potential for automated unit test generation. This has motivated us to investigate whether developer-defined test doubles (commonly referred to as mocks) available in existing test suites can be leveraged to improve LLM-driven test generation. To this end, we propose MOCKMILL, an LLM-based technique and tool that generates test cases by exploiting mocking information automatically extracted from developer-written tests. MOCKMILL targets components that are replaced by test doubles in existing tests and uses the encoded stubbings and interaction expectations to guide test generation, combined with an iterative generation-and-repair process to ensure executable tests. We evaluated MOCKMILL on 10 open-source classes from six Java projects using four LLMs, and compared the generated tests with existing project tests and tests produced by baseline approaches. The results show that MOCKMILL's tests cover lines of code and kill mutants that existing tests and baseline-generated tests miss. Overall, our findings provide preliminary evidence that leveraging mocking information is a complementary and effective way to enhance LLM-based test generation.

Neural processes are meta-learning models that map context sets to predictive distributions. While inspired by stochastic processes, NPs do not generally satisfy the Kolmogorov consistency conditions required to define a valid stochastic process. This inconsistency is widely acknowledged but poorly understood. Practitioners note that NPs work well despite the violation, without quantifying what this means. We address this gap by defining the conditioning consistency gap, a KL divergence measuring how much a conditional neural process's (CNP) predictions change when a point is added to the context versus conditioned upon. Our main results show that for CNPs with bounded encoders and Lipschitz decoders, the consistency gap is $O(1/n^2)$ in context size $n$, and that this rate is tight. These bounds establish the precise sense in which CNPs approximate valid stochastic processes. The inconsistency is negligible for moderate context sizes but can be significant in the few-shot regime.

Qualitative coding relies on a researcher's application of codes to textual data. As coding proceeds across large datasets, interpretations of codes often shift (temporal drift), reducing the credibility of the analysis. Existing Computer-Assisted Qualitative Data Analysis (CAQDAS) tools provide support for data management but offer no workflow for real-time detection of these drifts. We present Co-Refine, an AI-augmented qualitative coding platform that delivers continuous, grounded feedback on coding consistency without disrupting the researcher's workflow. The system employs a three-stage audit pipeline: Stage 1 computes deterministic embedding-based metrics for mathematical consistency; Stage 2 grounds LLM verdicts within $\pm0.15$ of the deterministic scores; and Stage 3 produces code definitions from previous patterns to create a deepening feedback loop. Co-Refine demonstrates that deterministic scoring can effectively constrain LLM outputs to produce reliable, real-time audit signals for qualitative analysis.

Automated Program Repair (APR) has benefited from the code understanding and generation capabilities of Large Language Models (LLMs). Existing feedback-based APR methods iteratively refine candidate patches using test execution feedback and have shown promising results. However, most rely on outcome-level failure symptoms, such as stack traces, which show how failures are observed but fail to expose the intermediate runtime states critical for root-cause analysis. As a result, LLMs often infer bug causes without sufficient runtime evidence, leading to incorrect patches. To address this limitation, we propose DebugRepair, a self-directed debugging framework for LLM-based APR. DebugRepair enhances patch refinement with intermediate runtime evidence collected through simulated debugging. It consists of three components: test semantic purification, simulated instrumentation, and debugging-driven conversational repair. Together, they reduce noisy test context, collect runtime traces through targeted debugging statements with rule-based fallback, and progressively refine candidate patches using prior attempts and newly observed runtime states. We evaluate DebugRepair on three benchmarks across Java and Python. Experiments show that DebugRepair achieves state-of-the-art performance against 15 approaches. With GPT-3.5, it correctly fixes 224 bugs on Defects4J, outperforming prior SOTA LLM-based methods by 26.2%. With DeepSeek-V3, it correctly fixes 295 Defects4J bugs, surpassing the second-best baseline by 59 bugs. Across five additional backbone LLMs, DebugRepair improves repair performance by 51.3% over vanilla settings. Ablation studies further confirm the effectiveness of all components.

The conformity bias exhibited by large language models (LLMs) can pose a significant challenge to decision-making in LLM-based multi-agent systems (LLM-MAS). While many prior studies have treated "conformity" simply as a matter of opinion change, this study introduces the social psychological distinction between informational conformity and normative conformity in order to understand LLM conformity at the mechanism level. Specifically, we design new tasks to distinguish between informational conformity, in which participants in a discussion are motivated to make accurate judgments, and normative conformity, in which participants are motivated to avoid conflict or gain acceptance within a group. We then conduct experiments based on these task settings. The experimental results show that, among the six LLMs evaluated, up to five exhibited tendencies toward not only informational conformity but also normative conformity. Furthermore, intriguingly, we demonstrate that by manipulating subtle aspects of the social context, it may be possible to control the target toward which a particular LLM directs its normative conformity. These findings suggest that decision-making in LLM-MAS may be vulnerable to manipulation by a small number of malicious users. In addition, through analysis of internal vectors associated with informational and normative conformity, we suggest that although both behaviors appear externally as the same form of "conformity," they may in fact be driven by distinct internal mechanisms. Taken together, these results may serve as an initial milestone toward understanding how "norms" are implemented in LLMs and how they influence group dynamics.

Large Audio-Language Models (LALMs) have recently achieved strong performance across various audio-centric tasks. However, hallucination, where models generate responses that are semantically incorrect or acoustically unsupported, remains largely underexplored in the audio domain. Existing hallucination benchmarks mainly focus on text or vision, while the few audio-oriented studies are limited in scale, modality coverage, and diagnostic depth. We therefore introduce HalluAudio, the first large-scale benchmark for evaluating hallucinations across speech, environmental sound, and music. HalluAudio comprises over 5K human-verified QA pairs and spans diverse task types, including binary judgments, multi-choice reasoning, attribute verification, and open-ended QA. To systematically induce hallucinations, we design adversarial prompts and mixed-audio conditions. Beyond accuracy, our evaluation protocol measures hallucination rate, yes/no bias, error-type analysis, and refusal rate, enabling a fine-grained analysis of LALM failure modes. We benchmark a broad range of open-source and proprietary models, providing the first large-scale comparison across speech, sound, and music. Our results reveal significant deficiencies in acoustic grounding, temporal reasoning, and music attribute understanding, underscoring the need for reliable and robust LALMs.

Despite the impressive capabilities of large language models, their substantial computational costs, latency, and privacy risks hinder their widespread deployment in real-world applications. Small Language Models (SLMs) with fewer than 10 billion parameters present a promising alternative; however, their inherent limitations in knowledge and reasoning curtail their effectiveness. Existing research primarily focuses on enhancing SLMs through scaling laws or fine-tuning strategies while overlooking the potential of using agent paradigms, such as tool use and multi-agent collaboration, to systematically compensate for the inherent weaknesses of small models. To address this gap, this paper presents the first large-scale, comprehensive study of <10B open-source models under three paradigms: (1) the base model, (2) a single agent equipped with tools, and (3) a multi-agent system with collaborative capabilities. Our results show that single-agent systems achieve the best balance between performance and cost, while multi-agent setups add overhead with limited gains. Our findings highlight the importance of agent-centric design for efficient and trustworthy deployment in resource-constrained settings.

We introduce IndiaFinBench, to our knowledge the first publicly available evaluation benchmark for assessing large language model (LLM) performance on Indian financial regulatory text. Existing financial NLP benchmarks draw exclusively from Western financial corpora (SEC filings, US earnings reports, and English-language financial news), leaving a significant gap in coverage of non-Western regulatory frameworks. IndiaFinBench addresses this gap with 406 expert-annotated question-answer pairs drawn from 192 documents sourced from the Securities and Exchange Board of India (SEBI) and the Reserve Bank of India (RBI), spanning four task types: regulatory interpretation (174 items), numerical reasoning (92 items), contradiction detection (62 items), and temporal reasoning (78 items). Annotation quality is validated through a model-based secondary pass (kappa=0.918 on contradiction detection) and a 60-item human inter-annotator agreement evaluation (kappa=0.611; 76.7% overall agreement). We evaluate twelve models under zero-shot conditions, with accuracy ranging from 70.4% (Gemma 4 E4B) to 89.7% (Gemini 2.5 Flash). All models substantially outperform a non-specialist human baseline of 60.0%. Numerical reasoning is the most discriminative task, with a 35.9 percentage-point spread across models. Bootstrap significance testing (10,000 resamples) reveals three statistically distinct performance tiers. The dataset, evaluation code, and all model outputs are available at https://github.com/rajveerpall/IndiaFinBench

Neural operators are widely used to approximate solution maps of complex physical systems. In many applications, however, the goal is not to recover the full solution trajectory, but to summarize the solution trajectory via a scalar target quantity (e.g., a functional such as time spent in a target range, time above a threshold, accumulated cost, or total energy). In this paper, we introduce DOPE (debiased neural operator): a semiparametric estimator for such target quantities of solution trajectories obtained from neural operators. DOPE is broadly applicable to settings with both partial and irregular observations and can be combined with arbitrary neural operator architectures. We make three main contributions. (1) We show that, in contrast to DOPE, naive plug-in estimation can suffer from first-order bias. (2) To address this, we derive a novel one-step, Neyman-orthogonal estimator that treats the neural operator as a high-dimensional nuisance mapping between function spaces, and removes the leading bias term. For this, DOPE uses a weighting mechanism that simultaneously accounts for irregular observation designs and for how sensitive the target quantity is to perturbations of the underlying trajectory. (3) To learn the weights, we extend automatic debiased machine learning to operator-valued nuisances via Riesz regression. We demonstrate the benefits of DOPE across various numerical experiments.

Test-time training (TTT) adapts model parameters on unlabeled test instances during inference time, which continuously extends capabilities beyond the reach of offline training. Despite initial gains, existing TTT methods for LRMs plateau quickly and do not benefit from additional test-time compute. Without external calibration, the self-generated reward signal increasingly drifts as the policy model evolves, leading to both performance plateaus and diversity collapse. We propose TEMPO, a TTT framework that interleaves policy refinement on unlabeled questions with periodic critic recalibration on a labeled dataset. By formalizing this alternating procedure through the Expectation-Maximization (EM) algorithm, we reveal that prior methods can be interpreted as incomplete variants that omit the crucial recalibration step. Reintroducing this step tightens the evidence lower bound (ELBO) and enables sustained improvement. Across diverse model families (Qwen3 and OLMO3) and reasoning tasks, TEMPO improves OLMO3-7B on AIME 2024 from 33.0% to 51.1% and Qwen3-14B from 42.3% to 65.8%, while maintaining high diversity.

Multilingual large language models (LLMs) have minimized the fluency gap between languages. This advancement, however, exposes models to the risk of biased behavior, as knowledge and norms may propagate across languages. In this work, we aim to quantify models' inter- and intra-lingual biases, via their ability to answer locale-ambiguous questions. To this end, we present LocQA, a test set containing 2,156 questions in 12 languages, referring to various locale-dependent facts such as laws, dates, and measurements. The questions do not contain indications of the locales they relate to, other than the querying language itself. LLMs' responses to LocQA locale-ambiguous questions thus reveal models' implicit priors. We used LocQA to evaluate 32 models, and detected two types of structural biases. Inter-lingually, we show a global bias towards answers relevant to the US-locale, even when models are asked in languages other than English. Moreover, we discovered that this global bias is exacerbated in models that underwent instruction tuning, compared to their base counterparts. Intra-lingually, we show that when multiple locales are relevant for the same language, models act as demographic probability engines, prioritizing locales with larger populations. Taken together, insights from LocQA may help in shaping LLMs' desired local behavior, and in quantifying the impact of various training phases on different kinds of biases.

The use of Large Language Models (LLMs) to support patients in addressing medical questions is becoming increasingly prevalent. However, most of the measures currently used to evaluate the performance of these models in this context only measure how closely a model's answers match semantically, and therefore do not provide a true indication of the model's medical accuracy or of the health equity risks associated with it. To address these shortcomings, we present a new evaluation framework for medical question answering called VB-Score (Verification-Based Score) that provides a separate evaluation of the four components of entity recognition, semantic similarity, factual consistency, and structured information completeness for medical question-answering models. We perform rigorous reviews of the performance of three well-known and widely used LLMs on 48 public health-related topics taken from high-quality, authoritative information sources. Based on our analyses, we discover a major discrepancy between the models' semantic and entity accuracy. Our assessments of the performance of all three models show that each of them has almost uniformly severe performance failures when evaluated against our criteria. Our findings indicate alarming performance disparities across various public health topics, with most of the models exhibiting 13.8% lower performance (compared to an overall average) for all the public health topics that relate to chronic conditions that occur in older and minority populations, which indicates the existence of what's known as condition-based algorithmic discrimination. Our findings also demonstrate that prompt engineering alone does not compensate for basic architectural limitations on how these models perform in extracting medical entities and raise the question of whether semantic evaluation alone is a sufficient measure of medical AI safety.

LLM-based multi-agent systems (MAS) show promise on complex tasks but remain prone to coordination failures such as goal drift, error cascades, and misaligned behaviors. We propose Explicit Trait Inference (ETI), a psychologically grounded method for improving coordination. ETI enables agents to infer and track partner characteristics along two established psychological dimensions--warmth (e.g., trust) and competence (e.g., skill)--from interaction histories to guide decisions. We evaluate ETI in controlled settings (economic games), where it reduces payoff loss by 45-77%, and in more realistic, complex multi-agent settings (MultiAgentBench), where it improves performance by 3-29% depending on the scenario and model, relative to a CoT baseline. Additional analysis shows that gains are closely linked to trait inference: ETI profiles predict agents' actions, and informative profiles drive improvements. These results highlight ETI as a lightweight and robust mechanism for improving coordination in diverse multi-agent settings, and provide the first systematic evidence that LLM agents can (i) reliably infer others' traits from interaction histories and (ii) leverage structured awareness of others' traits for coordination.

Large language models (LLMs) are increasingly used as co-authors in collaborative writing, where users begin with rough drafts and rely on LLMs to complete, revise, and refine their content. However, this capability poses a serious safety risk: malicious users could jailbreak the models-filling incomplete drafts with dangerous content-to force them into generating harmful outputs. In this paper, we identify the vulnerability of current LLMs to such draft-based co-authoring jailbreak attacks and introduce HarDBench, a systematic benchmark designed to evaluate the robustness of LLMs against this emerging threat. HarDBench spans a range of high-risk domains-including Explosives, Drugs, Weapons, and Cyberattacks-and features prompts with realistic structure and domain-specific cues to assess the model susceptibility to harmful completions. To mitigate this risk, we introduce a safety-utility balanced alignment approach based on preference optimization, training models to refuse harmful completions while remaining helpful on benign drafts. Experimental results show that existing LLMs are highly vulnerable in co-authoring contexts and our alignment method significantly reduces harmful outputs without degrading performance on co-authoring capabilities. This presents a new paradigm for evaluating and aligning LLMs in human-LLM collaborative writing settings. Our new benchmark and dataset are available on our project page at https://github.com/untae0122/HarDBench

Large language models (LLMs) are now deployed worldwide, inspiring a surge of benchmarks that measure their multilingual and multicultural abilities. However, these benchmarks prioritize generic language understanding or superficial cultural trivia, leaving the evaluation of grounded tasks -- where models must reason within real-world, context-rich scenarios -- largely unaddressed. To fill this gap, we present CulturALL, a comprehensive and challenging benchmark to assess LLMs' multilingual and multicultural competence on grounded tasks. CulturALL is built via a human--AI collaborative framework: expert annotators ensure appropriate difficulty and factual accuracy, while LLMs lighten the manual workload. By incorporating diverse sources, CulturALL ensures comprehensive scenario coverage. Each item is carefully designed to present a high level of difficulty, making CulturALL challenging. CulturALL contains 2,610 samples in 14 languages from 51 regions, distributed across 16 topics to capture the full breadth of grounded tasks. Experiments show that the best LLM achieves 44.48% accuracy on CulturALL, underscoring substantial room for improvement.

The evaluation of narrative quality remains a complex challenge, as it involves subjective factors such as plot, character development, and emotional impact. This work proposes a quantitative approach to narrative assessment by focusing on the linguistic dimension as a primary indicator of quality. The paper presents a methodology for the automatic evaluation of narrative based on the extraction of a comprehensive set of 33 quantitative linguistic features categorized into lexical, syntactic, and semantic groups. To test the model, an experiment was conducted on a specialized corpus of 23 books, including canonical masterpieces and self-published works. Through a similarity matrix, the system successfully clustered the narratives, distinguishing almost perfectly between professionally edited and self-published texts. Furthermore, the methodology was validated against a human-annotated dataset; it significantly outperforms traditional story-level evaluation metrics, demonstrating the effectiveness of quantitative linguistic features in assessing narrative quality.

Parameter-efficient fine-tuning (PEFT) reduces the training cost of full-parameter fine-tuning for large language models (LLMs) by training only a small set of task-specific parameters while freezing the pretrained backbone. However, existing approaches, such as Low-Rank Adaptation (LoRA), achieve adaptation by inserting independent low-rank perturbations directly to individual weights, resulting in a local parameterization of adaptation. We propose ShadowPEFT, a centralized PEFT framework that instead performs layer-level refinement through a depth-shared shadow module. At each transformer layer, ShadowPEFT maintains a parallel shadow state and evolves it repeatedly for progressively richer hidden states. This design shifts adaptation from distributed weight-space perturbations to a shared layer-space refinement process. Since the shadow module is decoupled from the backbone, it can be reused across depth, independently pretrained, and optionally deployed in a detached mode, benefiting edge computing scenarios. Experiments on generation and understanding benchmarks show that ShadowPEFT matches or outperforms LoRA and DoRA under comparable trainable-parameter budgets. Additional analyses on shadow pretraining, cross-dataset transfer, parameter scaling, inference latency, and system-level evaluation suggest that centralized layer-space adaptation is a competitive and flexible alternative to conventional low-rank PEFT.

Streamliner constraints reduce the search space of combinatorial problems by ruling out portions of the solution space. We adapt the StreamLLM approach, which uses Large Language Models (LLMs) to generate streamliners for Constraint Programming, to Answer Set Programming (ASP). Given an ASP encoding and a few small training instances, we prompt multiple LLMs to propose candidate constraints. Candidates that cause syntax errors, render satisfiable instances unsatisfiable, or degrade performance on all training instances are discarded. The surviving streamliners are evaluated together with the original encoding, and we report results for a virtual best encoding (VBE) that, for each instance, selects the fastest among the original encoding and its streamlined variants. On three ASP Competition benchmarks (Partner Units Problem, Sokoban, Towers of Hanoi), the VBE achieves speedups of up to 4--5x over the original encoding. Different LLMs produce semantically diverse constraints, not mere syntactic variations, indicating that the approach captures genuine problem structure.

Developing Visual Analytics (VA) applications requires integrating complex machine learning models with expressive interactive interfaces. Developers face a stark trade-off: building tightly-coupled monoliths plagued by fragile interdependencies, or relying on restrictive, simplistic frameworks. Meanwhile, unconstrained, single-shot AI code generation promises speed but yields unstructured, unauditable chaos. The core challenge is combining the control and expressiveness of custom development with the efficiency of AI generation under strict constraints. To address this, we introduce BONSAI, a mixed-initiative workspace for the multi-agent co-development of VA applications. BONSAI utilizes a modular four-layer architecture (hardware, services, orchestration, application) that allows human and AI developers to independently contribute reusable components. The workspace incorporates this architecture into a structured four-phase development process (plan, design, monitor, and review), ensuring distributed agency and full provenance, where all human and AI contributions are structurally bounded and tracked. We evaluate BONSAI through case studies demonstrating the efficient creation of novel tools and the rapid reconstruction of complex VA applications directly from research paper descriptions. Ultimately, this paper contributes a conceptual workflow, a scalable architecture, and an integrated system that successfully balances AI's generative speed with the structural rigor required for complex VA development.

Repair, an important resource for resolving trouble in human-human conversation, remains underexplored in human-LLM interaction. In this study, we investigate how LLMs engage in the interactive process of repair in multi-turn dialogues around solvable and unsolvable math questions. We examine whether models initiate repair themselves and how they respond to user-initiated repair. Our results show strong differences across models: reactions range from being almost completely resistant to (appropriate) repair attempts to being highly susceptible and easily manipulated. We further demonstrate that once conversations extend beyond a single turn, model behavior becomes more distinctive and less predictable across systems. Overall, our findings indicate that each tested LLM exhibits its own characteristic form of unreliability in the context of repair.

Industrial surface defect detection often suffers from limited defect samples, severe long-tailed distributions, and difficulties in accurately localizing subtle defects under complex backgrounds. To address these challenges, this paper proposes an unsupervised defect detection method that integrates a Denoising Diffusion Probabilistic Model (DDPM) with an asymmetric teacher-student architecture. First, at the data level, the DDPM is trained solely on normal samples. By introducing constant-variance Gaussian perturbations and Perlin noise-based masks, high-fidelity and physically consistent defect samples along with pixel-level annotations are generated, effectively alleviating the data scarcity problem. Second, at the model level, an asymmetric dual-stream network is constructed. The teacher network provides stable representations of normal features, while the student network reconstructs normal patterns and amplifies discrepancies between normal and anomalous regions. Finally, a joint optimization strategy combining cosine similarity loss and pixel-wise segmentation supervision is adopted to achieve precise localization of subtle defects. Experimental results on the MVTecAD dataset show that the proposed method achieves 98.4\% image-level AUROC and 98.3\% pixel-level AUROC, significantly outperforming existing unsupervised and mainstream deep learning methods. The proposed approach does not require large amounts of real defect samples and enables accurate and robust industrial defect detection and localization. \keywords{Industrial defect detection \and diffusion models \and data generation \and teacher-student architecture \and pixel-level localization}

Automatically generating bug reproduction tests (BRT) from issue descriptions is crucial for software maintenance. LLM-based approaches have shown great potential for this task. Their effectiveness heavily relies on retrieving high-quality context from the codebase. The retrieval phase of existing approaches relies on either traditional methods like BM25 or LLM-driven strategies. LLM-based retrieval strategies typically equip an LLM with tools to autonomously explore the repository or select the most relevant files and code snippets from a provided list as context. However, these retrieval methods suffer from three key limitations: 1) They often employ a unified strategy for retrieving both source code and test cases, overlooking their distinct retrieval requirements. 2) They focus solely on semantic similarity while ignoring function call relationships, leading to irrelevant context. 3) The retrieval lacks a feedback loop from the generation phase, preventing it from refining the context based on execution results. These limitations collectively result in low-quality context, thereby hindering the accuracy of bug reproduction. To address these challenges, we propose iCoRe, an iterative, correlation-aware context retrieval approach explicitly aware of three key correlations: 1) between source code and test cases, which requires differentiated retrieval, 2) between textual semantics and function call structures for accurate relevance assessment, and 3) between the retrieval and generation phases, which enables iterative feedback and refinement. To evaluate iCoRe, we integrate it with an LLM-based BRT generator and conduct a comprehensive evaluation on the SWT-bench Lite and TDD-bench Verified benchmarks. Experimental results show that our method achieves a Fail-to-Pass rate of 42.0% and 52.8% respectively, representing 19.7%-31.7% relative improvements over existing retrieval methods.

Full-duplex speech interaction, as the most natural and intuitive mode of human communication, is driving artificial intelligence toward more human-like conversational systems. Traditional cascaded speech processing pipelines suffer from critical limitations, including accumulated latency, information loss, and error propagation across modules. To address these issues, recent efforts focus on the end-to-end audio large language models (LLMs) like GPT-4o, which primarily unify speech understanding and generation task. However, most of these models are inherently half-duplex, and rely on a suite of separate, task-specific front-end components, such as voice activity detection (VAD) and turn-taking detection (TD). In our development of speech assistant, we observed that optimizing the speech front-end is equally crucial as advancing the back-end unified model for achieving seamless, responsive interactions. To bridge this gap, we propose the first unified audio front-end LLM (UAF) tailored for full-duplex speech systems. Our model reformulates diverse audio front-end tasks into a single auto-regressive sequence prediction problem, including VAD, TD, speaker recognition (SR), automatic speech recognition (ASR) and question answer (QA). It takes streaming fixed-duration audio chunk (e.g., 600 ms) as input, leverages a reference audio prompt to anchor the target speaker at the beginning, and regressively generates discrete tokens encoding both semantic content and system-level state controls (e.g., interruption signals). Experiments demonstrate that our model achieves leading performance across multiple audio front-end tasks and significantly enhances response latency and interruption accuracy in real-world interaction scenarios.

Federated Learning (FL) enables collaborative model training among multiple parties without centralizing raw data. There are two main paradigms in FL: Horizontal FL (HFL), where all participants share the same feature space but hold different samples, and Vertical FL (VFL), where parties possess complementary features for the same set of samples. A prerequisite for VFL training is privacy-preserving entity alignment (PPEA), which establishes a common index of samples across parties (alignment) without revealing which samples are shared between them. Conventional private set intersection (PSI) achieves alignment but leaks intersection membership, exposing sensitive relationships between datasets. The standard private set union (PSU) mitigates this risk by aligning on the union of identifiers rather than the intersection. However, existing approaches are often limited to two parties or lack support for typo-tolerant matching. In this paper, we introduce the Sherpa.ai multi-party PSU protocol for VFL, a PPEA method that hides intersection membership and enables both exact and noisy matching. The protocol generalizes two-party approaches to multiple parties with low communication overhead and offers two variants: an order-preserving version for exact alignment and an unordered version tolerant to typographical and formatting discrepancies. We prove correctness and privacy, analyze communication and computational (exponentiation) complexity, and formalize a universal index mapping from local records to a shared index space. This multi-party PSU offers a scalable, mathematically grounded protocol for PPEA in real-world VFL deployments, such as multi-institutional healthcare disease detection, collaborative risk modeling between banks and insurers, and cross-domain fraud detection between telecommunications and financial institutions, while preserving intersection privacy.

Crop yield prediction is one of the most important challenge, which is crucial to world food security and policy-making decisions. The conventional forecasting techniques are limited in their accuracy with reference to the fact that they utilize static data sources that do not reflect the dynamic and intricate relationships that exist between the variables of the environment over time [5,13]. This paper presents Attention-Based Multi-Modal Deep Learning Framework (ABMMDLF), which is suggested to be used in high-accuracy spatio-temporal crop yield prediction. The model we use combines multi-year satellite imagery, high-resolution time-series of meteorological data and initial soil properties as opposed to the traditional models which use only one of the aforementioned factors [12, 21]. The main architecture involves the use of Convolutional Neural Networks (CNN) to extract spatial features and a Temporal Attention Mechanism to adaptively weight important phenological periods targeted by the algorithm to change over time and condition on spatial features of images and video sequences. As can be experimentally seen, the proposed research work provides an R^2 score of 0.89, which is far better than the baseline models do.

Graph neural networks (GNNs) are the standard for learning on graphs, yet they have limited expressive power, often expressed in terms of the Weisfeiler-Leman (WL) hierarchy or within the framework of first-order logic. In this context, topological neural networks (TNNs) have recently emerged as a promising alternative for graph representation learning. By incorporating higher-order relational structures into message-passing schemes, TNNs offer higher representational power than traditional GNNs. However, a fundamental question remains open: what is the logical expressiveness of TNNs? Answering this allows us to characterize precisely which binary classifiers TNNs can represent. In this paper, we address this question by analyzing isomorphism tests derived from the underlying mechanisms of general TNNs. We introduce and investigate the power of higher-order variants of WL-based tests for combinatorial complexes, called $k$-CCWL test. In addition, we introduce the topological counting logic (TC$_k$), an extension of standard counting logic featuring a novel pairwise counting quantifier $ \exists^{N}(x_i,x_j)\, \varphi(x_i,x_j), $ which explicitly quantifies pairs $(x_i, x_j)$ satisfying property $\varphi$. We rigorously prove the exact equivalence: $ \text{k-CCWL} \equiv \text{TC}_{k{+}2} \equiv \text{Topological }(k{+}2)\text{-pebble game}.$ These results establish a logical expressiveness theory for TNNs.

Current AI agent frameworks have made remarkable progress in automating individual tasks, yet all existing systems serve a single user. Human productivity rests on the social and organizational relationships through which people coordinate, negotiate, and delegate. When agents move beyond performing tasks for one person to representing that person in collaboration with others, the infrastructure for cross-user agent collaboration is entirely absent, let alone the governance mechanisms needed to secure it. We argue that the next frontier for AI agents lies not in stronger individual capability, but in the digitization of human collaborative relationships. To this end, we propose a human-symbiotic agent paradigm. Each user owns a permanently bound agent system that collaborates on the owner's behalf, forming a network whose nodes are humans rather than agents. This paradigm rests on three governance primitives. A layered identity architecture separates a Manager Agent from multiple context-specific Identity Agents; the Manager Agent holds global knowledge but is architecturally isolated from external communication. Scoped authorization enforces per-identity access control and escalates boundary violations to the owner. Action-level accountability logs every operation against its owner's identity and authorization, ensuring full auditability. We instantiate this paradigm in ClawNet, an identity-governed agent collaboration framework that enforces identity binding and authorization verification through a central orchestrator, enabling multiple users to collaborate securely through their respective agents.

LLMs are increasingly being considered for prediction tasks in high-stakes social service settings, but their algorithmic fairness properties in this context are poorly understood. In this short technical report, we audit the algorithmic fairness of LLM-based tabular classification on a real housing placement prediction task, augmented with street outreach casenotes from a nonprofit partner. We audit multi-class classification error disparities. We find that a fine-tuned model augmented with casenote summaries can improve accuracy while reducing algorithmic fairness disparities. We experiment with variable importance improvements to zero-shot tabular classification and find mixed results on resulting algorithmic fairness. Overall, given historical inequities in housing placement, it is crucial to audit LLM use. We find that leveraging LLMs to augment tabular classification with casenote summaries can safely leverage additional text information at low implementation burden. The outreach casenotes are fairly short and heavily redacted. Our assessment is that LLM zero-shot classification does not introduce additional textual biases beyond algorithmic biases in tabular classification. Combining fine-tuning and leveraging casenote summaries can improve accuracy and algorithmic fairness.

Code editing constitutes a fundamental practice in software development, wherein developers modify existing codebases according to natural language requirements. Accurate code editing necessitates a comprehensive understanding of both the existing codebase and the modification requirements. Although large language models (LLMs) have demonstrated promising performance in code editing tasks, they suffer from substantial inefficiency by generating entire modified files that largely consist of unchanged code. While smaller models could potentially address this inefficiency, they typically lack the capacity to effectively comprehend long code contexts required for accurate editing. To ensure both effectiveness and efficiency, we propose to decompose code editing into a two-stage cascade: \textbf{edit sketch generation}, wherein a large model first produces concise sketches representing the requisite modifications (the more challenging phase), and \textbf{edit sketch application}, wherein a smaller model integrates these sketches into the original code to produce the final output edited code (the simpler phase). This cascaded design reduces the number of tokens generated by the large model, as the majority of the output is handled by the smaller, more efficient model, thereby enhancing overall efficiency. However, the effectiveness of this approach is constrained by current small models' limited capabilities in handling long-context scenarios and cross-file dependencies, which are essential for accurate sketch application in real-world codebases. To address these limitations and enhance smaller models' sketch application capabilities, ...

Anomaly detection in medical imaging is essential for identifying rare pathological conditions, particularly when annotated abnormal samples are limited. We propose a hybrid anomaly detection framework that integrates self-supervised representation learning with manifold-based density estimation, a combination that remains largely unexplored in this domain. Medical images are first embedded into a latent feature space using pretrained, potentially domain-specific, backbones. These representations are then refined via Mean Shift Density Enhancement (MSDE), an iterative manifold-shifting procedure that moves samples toward regions of higher likelihood. Anomaly scores are subsequently computed using Gaussian density estimation in a PCA-reduced latent space, where Mahalanobis distance measures deviation from the learned normal distribution. The framework follows a one-class learning paradigm and requires only normal samples for training. Extensive experiments on seven medical imaging datasets demonstrate state-of-the-art performance. MSDE achieves the highest AUC on four datasets and the highest Average Precision on five datasets, including near-perfect performance on brain tumor detection (0.981 AUC/AP). These results underscore the potential of the proposed framework as a scalable clinical decision-support tool for early disease detection, screening in low-label settings, and robust deployment across diverse imaging modalities.

For news headlines to influence beliefs and drive action, relevant information needs to be retained and retrievable from memory. In this probing study we draw on experiment designs from cognitive psychology to examine how a specific linguistic feature, namely direct address through first- and second-person pronouns, affects memorability and to what extent it is feasible to use large language models for the targeted insertion of such a feature into existing text without changing its core meaning. Across three controlled memorization experiments with a total of 240 participants, yielding 7,680 unique memory judgments, we show that pronoun insertion has mixed effects on memorability. Exploratory analyses indicate that effects differ based on headline topic, how pronouns are inserted and their immediate contexts. Additional data and fine-grained analysis is needed to draw definitive conclusions on these mediating factors. We further show that automatic revisions by LLMs are not always appropriate: Crowdsourced evaluations find many of them to be lacking in content accuracy and emotion retention or resulting in unnatural writing style. We make our collected data available for future work.

Heterophily is a prevalent property of real-world graphs and is well known to impair the performance of homophilic Graph Neural Networks (GNNs). Prior work has attempted to adapt GNNs to heterophilic graphs through non-local neighbor extension or architecture refinement. However, the fundamental reasons behind misclassifications remain poorly understood. In this work, we take a novel perspective by examining recurring inductive subgraphs, empirically and theoretically showing that they act as spurious shortcuts that mislead GNNs and reinforce non-causal correlations in heterophilic graphs. To address this, we adopt a causal inference perspective to analyze and correct the biased learning behavior induced by shortcut inductive subgraphs. We propose a debiased causal graph that explicitly blocks confounding and spillover paths responsible for these shortcuts. Guided by this causal graph, we introduce Causal Disentangled GNN (CD-GNN), a principled framework that disentangles spurious inductive subgraphs from true causal subgraphs by explicitly blocking non-causal paths. By focusing on genuine causal signals, CD-GNN substantially improves the robustness and accuracy of node classification in heterophilic graphs. Extensive experiments on real-world datasets not only validate our theoretical findings but also demonstrate that our proposed CD-GNN outperforms state-of-the-art heterophily-aware baselines.

Small language models (SLMs), such as BART, can achieve summarization performance comparable to large language models (LLMs) via distillation. However, existing LLM-based ranking strategies for summary candidates suffer from instability, while classical metrics (e.g., ROUGE) are insufficient to rank high-quality summaries. To address these issues, we introduce \textbf{SCURank}, a framework that enhances summarization by leveraging \textbf{Summary Content Units (SCUs)}. Instead of relying on unstable comparisons or surface-level overlap, SCURank evaluates summaries based on the richness and semantic importance of information content. We investigate the effectiveness of SCURank in distilling summaries from multiple diverse LLMs. Experimental results demonstrate that SCURank outperforms traditional metrics and LLM-based ranking methods across evaluation measures and datasets. Furthermore, our findings show that incorporating diverse LLM summaries enhances model abstractiveness and overall distilled model performance, validating the benefits of information-centric ranking in multi-LLM distillation. The code for SCURank is available at https://github.com/IKMLab/SCURank.

We study the deep image prior (DIP) framework applied to photoacoustic tomography (PAT) as an unsupervised reconstruction approach to mitigate limited-view artifacts and noise commonly encountered in experimental settings. Efficient implementation is achieved by employing recently published fast forward and adjoint algorithms for circular measurement geometries. Initialization via a fast inverse and total variation (TV) regularization are applied to further suppress noise and mitigate overfitting. For comparison, we compute a classical TV reconstruction. Our experiments comprise simulated PAT measurements under limited-view geometries and varying levels of added noise as well as experimental measurements together with using a digital twin for quality assessment. Our findings suggest that DIP framework provides an effective unsupervised strategy for robust PAT reconstruction even in the challenging case of a limited view geometry providing improvement in several quantitative measures over total variation reconstructions.

The rapid advancement and widespread adoption of Large Language Models (LLMs) have elevated the need for reliable AI-generated content (AIGC) detection, which remains challenging as models evolve. We introduce AIGC-text-bank, a comprehensive multi-domain dataset with diverse LLM sources and authorship scenarios, and propose REVEAL, a detection framework that generates interpretable reasoning chains before classification. Our approach uses a two-stage training strategy: supervised fine-tuning to establish reasoning capabilities, followed by reinforcement learning to improve accuracy, improve logical consistency, and reduce hallucinations. Extensive experiments show that REVEAL achieves state-of-the-art performance across multiple benchmarks, offering a robust and transparent solution for AIGC detection. The project is open-source at https://aka.ms/reveal

Heterogeneous graphs have attracted increasing attention for modeling multi-typed entities and relations in complex real-world systems. Multi-label node classification on heterogeneous graphs is challenging due to structural heterogeneity and the need to learn shared representations across multiple labels. Existing methods typically adopt either flexible attention mechanisms or meta-path constrained anchoring, but in heterogeneous multi-label prediction they often suffer from semantic dilution or coverage constraint. Both issues are further amplified under multi-label supervision. We present a theoretical analysis showing that as heterogeneous neighborhoods expand, the attention mass allocated to task-critical (primary) neighborhoods diminishes, and that meta-path constrained aggregation exhibits a dilemma: too few meta-paths intensify coverage constraint, while too many re-introduce dilution. To resolve this coverage-anchoring conflict, we propose FOCAL: Fusion Of Coverage and Anchoring Learning, with two components: coverage-oriented attention (COA) for flexible, unconstrained heterogeneous context aggregation, and anchoring-oriented attention (AOA) that restricts aggregation to meta-path-induced primary semantics. Our theoretical analysis and experimental results further indicates that FOCAL has a better performance than other state-of-the-art methods.

Deploying large language models (LLMs) in resource-constrained environments is hindered by heavy computational and memory requirements. We present LBLLM, a lightweight binarization framework that achieves effective W(1+1)A4 quantization through a novel three-stage quantization strategy. The framework proceeds as follows: (1) initialize a high-quality quantized model via PTQ; (2) quantize binarized weights, group-wise bitmaps, and quantization parameters through layer-wise distillation while keeping activations in full precision; and (3) training learnable activation quantization factors to dynamically quantize activations to 4 bits. This decoupled design mitigates interference between weight and activation quantization, yielding greater training stability and better inference accuracy. LBLLM, trained only using 0.016B tokens with a single GPU, surpasses existing state-of-the-art binarization methods on W2A4 quantization settings across tasks of language modeling, commonsense QA, and language understanding. These results demonstrate that extreme low-bit quantization of LLMs can be both practical and highly effective without introducing any extra high-precision channels or rotational matrices commonly used in recent PTQ-based works, offering a promising path toward efficient LLM deployment in resource-limited situations.

In uncertainty quantification, evaluating sensitivity measures under specific conditions (i.e., conditional Sobol' indices) is essential for systems with parameterized responses, such as spatial fields or varying operating conditions. Traditional approaches often rely on point-wise modeling, which is computationally expensive and may lack consistency across the parameter space. This paper demonstrates that for a pre-trained global Polynomial Chaos Expansion (PCE) model, the analytical conditional Sobol' indices are inherently embedded within its basis functions. By leveraging the tensor-product property of PCE bases, we reformulate the global expansion into a set of analytical coefficient fields that depend on the conditioning variables. Based on the preservation of orthogonality under conditional probability measures, we derive closed-form expressions for conditional variances and Sobol' indices. This framework bypasses the need for repetitive modeling or additional sampling, transforming conditional sensitivity analysis into a purely algebraic post-processing step. Numerical benchmarks indicate that the proposed method ensures physical coherence and offers superior numerical robustness and computational efficiency compared to conventional point-wise approaches.

This paper studies uncertainty quantification for large language models (LLMs) under black-box access, where only a small number of responses can be sampled for each query. In this setting, estimating the effective semantic alphabet size--that is, the number of distinct meanings expressed in the sampled responses--provides a useful proxy for downstream risk. However, frequency-based estimators tend to undercount rare semantic modes when the sample size is small, while graph-spectral quantities alone are not designed to estimate semantic occupancy accurately. To address this issue, we propose SHADE (Soft-Hybrid Alphabet Dynamic Estimator), a simple and interpretable estimator that combines Generalized Good-Turing coverage with a heat-kernel trace of the normalized Laplacian constructed from an entailment-weighted graph over sampled responses. The estimated coverage adaptively determines the fusion rule: under high coverage, SHADE uses a convex combination of the two signals, while under low coverage it applies a LogSumExp fusion to emphasize missing or weakly observed semantic modes. A finite-sample correction is then introduced to stabilize the resulting cardinality estimate before converting it into a coverage-adjusted semantic entropy score. Experiments on pooled semantic alphabet-size estimation against large-sample references and on QA incorrectness detection show that SHADE achieves the strongest improvements in the most sample-limited regime, while the performance gap narrows as the number of samples increases. These results suggest that hybrid semantic occupancy estimation is particularly beneficial when black-box uncertainty quantification must operate under tight sampling budgets.

Deep-feature-based perceptual similarity models have demonstrated strong alignment with human visual perception in Image Quality Assessment (IQA). However, most existing approaches operate at a single spatial scale, implicitly assuming that structural similarity at a fixed resolution is sufficient. The role of spatial scale in deep-feature similarity modeling thus remains insufficiently understood. In this letter, we isolate spatial scale as an independent factor using a minimal multiscale extension of DeepSSIM, referred to as Deep Structural Similarity with Multiscale Representation (MSDS). The proposed framework decouples deep feature representation from cross-scale integration by computing DeepSSIM independently across pyramid levels and fusing the resulting scores with a lightweight set of learnable global weights. Experiments on multiple benchmark datasets demonstrate consistent and statistically significant improvements over the single-scale baseline, while introducing negligible additional complexity. The results empirically confirm spatial scale as a non-negligible factor in deep perceptual similarity, isolated here via a minimal testbed.

KV-cache memory is a major bottleneck in real-world LLM serving, where systems must simultaneously support latency-sensitive small-batch requests and high-throughput concurrent workloads. Although many KV-cache compression methods improve offline accuracy or compression ratio, they often violate practical serving constraints such as paged memory layouts, regular memory access, and fused attention execution, limiting their effectiveness in deployment. In this work, we identify the minimal set of 4-bit KV-cache quantization methods that remain viable under these constraints. Our central finding is that a simple design--token-wise INT4 quantization with block-diagonal Hadamard rotation--consistently achieves the best accuracy-efficiency trade-off. Across multiple models and benchmarks, this approach recovers nearly all of the accuracy lost by naive INT4, while more complex methods such as vector quantization and Hessian-aware quantization provide only marginal additional gains once serving compatibility is taken into account. To make this practical, we implement a fused rotation-quantization kernel that integrates directly into paged KV-cache layouts and introduces zero measurable end-to-end overhead, matching plain INT4 throughput across concurrency levels. Our results show that effective KV-cache compression is fundamentally a systems co-design problem: under real serving constraints, lightweight block-diagonal Hadamard rotation is a viable method that delivers near-lossless accuracy without sacrificing serving efficiency.

Existing Indic ASR benchmarks often use scripted, clean speech and leaderboard driven evaluation that encourages dataset specific overfitting. In addition, strict single reference WER penalizes natural spelling variation in Indian languages, including non standardized spellings of code-mixed English origin words. To address these limitations, we introduce Voice of India, a closed source benchmark built from unscripted telephonic conversations covering 15 major Indian languages across 139 regional clusters. The dataset contains 306230 utterances, totaling 536 hours of speech from 36691 speakers with transcripts accounting for spelling variations. We also analyze performance geographically at the district level, revealing disparities. Finally, we provide detailed analysis across factors such as audio quality, speaking rate, gender, and device type, highlighting where current ASR systems struggle and offering insights for improving real world Indic ASR systems.

Thinking LLMs produce reasoning traces before answering. Prior activation steering work mainly targets on shaping these traces. It remains less understood how answer tokens actually read and integrate the reasoning to produce reliable outcomes. Focusing on quantitative reasoning, we analyze the answer-to-reasoning attention and observe a benign self-reading pattern aligned with correctness, characterized by a forward drift of the reading focus along the reasoning trace and a persistent concentration on key semantic anchors, whereas incorrect solutions exhibit diffuse and irregular attention pattern. We interpret this as internal certainty during answer decoding, where the model commits to a viable solution branch and integrates key evidence. Following this, we propose a training-free steering method driven by Self-Reading Quality (SRQ) scores combining geometric metrics for process control with semantic metrics for content monitoring. SRQ selects data to build steering vectors that guide inference toward benign self-reading and away from uncertain and disorganized reading. Experiments show that our method yields consistent accuracy gains.

Scaling Transformers typically necessitates training larger models from scratch, as standard architectures struggle to expand without discarding learned representations. We identify the primary bottleneck in the attention mechanism's linear projections, which strictly confine feature extraction to fixed-dimensional subspaces, limiting both expressivity and incremental capacity. To address this, we introduce Nexusformer, which replaces linear $Q/K/V$ projections with a Nexus-Rank layer, a three-stage nonlinear mapping driven by dual activations in progressively higher dimensional spaces. This design overcomes the linearity constraint and enables lossless structured growth: new capacity can be injected along two axes via zero-initialized blocks that preserve pretrained knowledge. Experiments on language modeling and reasoning benchmarks demonstrate that Nexusformer matches Tokenformer's perplexity using up to 41.5\% less training compute during progressive scaling (240M to 440M). Furthermore, our analysis of growth dynamics reveals that zero initialization induces a stable convergence trajectory, allowing us to derive a geometric scaling law that accurately predicts performance across expansion scales.

Particle accelerator beamline optimization is a high-dimensional control problem traditionally requiring significant expert intervention. We present RLABC (Reinforcement Learning for Accelerator Beamline Control), an open-source Python framework that automatically transforms standard Elegant beamline configurations into reinforcement learning environments. RLABC integrates with the widely-used Elegant beam dynamics simulation code via SDDS-based interfaces, enabling researchers to apply modern RL algorithms to beamline optimization with minimal RL-specific development. The main contribution is a general methodology for formulating beamline tuning as a Markov decision process: RLABC automatically preprocesses lattice files to insert diagnostic watch points before each tunable element, constructs a 57-dimensional state representation from beam statistics, covariance information, and aperture constraints, and provides a configurable reward function for transmission optimization. The framework supports multiple RL algorithms through Stable-Baselines3 compatibility and implements stage learning strategies for improved training efficiency. Validation on a test beamline derived from the VEPP-5 injection complex (37 control parameters across 11 quadrupoles and 4 dipoles) demonstrates that the framework successfully enables RL-based optimization, with a Deep Deterministic Policy Gradient agent achieving 70.3\% particle transmission -- performance matching established methods such as differential evolution. The framework's stage learning capability allows decomposition of complex optimization problems into manageable subproblems, improving training efficiency. The complete framework, including configuration files and example notebooks, is available as open-source software to facilitate adoption and further research.

Vision-Language Models (VLMs) have become central to autonomous driving systems, yet their deployment is severely bottlenecked by the massive computational overhead of multi-view camera and multi-frame video input. Existing token pruning methods, primarily designed for single-image inputs, treat each frame or view in isolation and thus fail to exploit the inherent spatio-temporal redundancies in driving scenarios. To bridge this gap, we propose ST-Prune, a training-free, plug-and-play framework comprising two complementary modules: Motion-aware Temporal Pruning (MTP) and Ring-view Spatial Pruning (RSP). MTP addresses temporal redundancy by encoding motion volatility and temporal recency as soft constraints within the diversity selection objective, prioritizing dynamic trajectories and current-frame content over static historical background. RSP further resolves spatial redundancy by exploiting the ring-view camera geometry to penalize bilateral cross-view similarity, eliminating duplicate projections and residual background that temporal pruning alone cannot suppress. These two modules together constitute a complete spatio-temporal pruning process, preserving key scene information under strict compression. Validated across four benchmarks spanning perception, prediction, and planning, ST-Prune establishes new state-of-the-art for training-free token pruning. Notably, even at 90\% token reduction, ST-Prune achieves near-lossless performance with certain metrics surpassing the full-model baseline, while maintaining inference speeds comparable to existing pruning approaches.

Recent years have witnessed growing interest in applying Large Reasoning Models (LRMs) to Machine Translation (MT). Existing approaches predominantly adopt a "think-first-then-translate" paradigm. Although explicit reasoning trajectories significantly enhance translation quality, they incur prohibitive inference costs and latency. To address these limitations, we propose ReflectMT, a two-stage reflection internalization algorithm for machine translation that employs a "translate-first-think-later" paradigm. Our approach develops the model's "translate-reflect-refine" capability through reinforcement learning. In the first stage, we cultivate the model's capacity for high-quality reflection and refinement, thereby enhancing its semantic comprehension and task-specific knowledge. In the second stage, we train the model to internalize the knowledge acquired during reflection. As a result, during inference, ReflectMT operates in a direct translation mode, producing high-quality translations on the first attempt without any explicit reasoning steps. Experimental results on datasets such as WMT24 demonstrate that our model's first-pass translations during inference outperform multi-step reasoning LRMs such as DeepSeek-R1 in both automatic metrics and GPT-based evaluation, achieving a 2.16-point improvement in GPT-based translation quality evaluation while reducing token consumption by 94.33%.

As generative AI becomes integral to software development, the risk of over-reliance and diminished critical thinking grows. This study introduces "Ceci," our Caring Empathic C IDE designed to support novice programmers by prioritizing learning and emotional support over direct code generation. The researchers conducted a comparative pilot study between Ceci and VSCode + ChatGPT [9, 40]. Participants completed a coding task and were evaluated using the NASA-TLX workload assessment and a post-test usability survey. Although the sample size was small (n = 11), results show that there is no significant difference in perceived effectiveness, learning and workload between the Experimental Ceci group and the Control group, though Ceci users reported significantly greater perceived helpfulness in error correction (p = 0.0220). These findings suggest that empathic responses may not be sufficient on their own to enhance the learner's outcomes, perceptions, or reduce workload. Overall, this study provides a foundational framework for future research. Such research should explore larger sample sizes, diverse programming tasks, and additional empathic features to better understand the potential of empathic programming environments in supporting novice programmers; they must also ensure that the empathic features are well-integrated in the user interface.

As Large Language Models (LLMs) continue to evolve through alignment techniques such as Reinforcement Learning from Human Feedback (RLHF) and Constitutional AI, a growing and increasingly conspicuous phenomenon has emerged: the proliferation of verbal tics -- repetitive, formulaic linguistic patterns that pervade model outputs. These range from sycophantic openers ("That's a great question!", "Awesome!") to pseudo-empathetic affirmations ("I completely understand your concern", "I'm right here to catch you") and overused vocabulary ("delve", "tapestry", "nuanced"). In this paper, we present a systematic analysis of the verbal tic phenomenon across eight state-of-the-art LLMs: GPT-5.4, Claude Opus 4.7, Gemini 3.1 Pro, Grok 4.2, Doubao-Seed-2.0-pro, Kimi K2.5, DeepSeek V3.2, and MiMo-V2-Pro. Utilizing a custom evaluation framework for standardized API-based evaluation, we assess 10,000 prompts across 10 task categories in both English and Chinese, yielding 160,000 model responses. We introduce the Verbal Tic Index (VTI), a composite metric quantifying tic prevalence, and analyze its correlation with sycophancy, lexical diversity, and human-perceived naturalness. Our findings reveal significant inter-model variation: Gemini 3.1 Pro exhibits the highest VTI (0.590), while DeepSeek V3.2 achieves the lowest (0.295). We further demonstrate that verbal tics accumulate over multi-turn conversations, are amplified in subjective tasks, and show distinct cross-lingual patterns. Human evaluation (N = 120) confirms a strong inverse relationship between sycophancy and perceived naturalness (r = -0.87, p < 0.001). These results underscore the "alignment tax" of current training paradigms and highlight the urgent need for more authentic human-AI interaction frameworks.

Knowledge Graph (KG) can effectively integrate valuable information from massive data, and thus has been rapidly developed and widely used in many fields. Traditional KG construction methods rely on manual annotation, which often consumes a lot of time and manpower. And KG construction schemes based on deep learning tend to have weak generalization capabilities. With the rapid development of Pre-trained Language Models (PLM), PLM has shown great potential in the field of KG construction. This paper provides a comprehensive review of recent research advances in the field of construction of KGs using PLM. In this paper, we explain how PLM can utilize its language understanding and generation capabilities to automatically extract key information for KGs, such as entities and relations, from textual data. In addition, We also propose a new Hyper-Relarional Knowledge Graph construction framework based on lightweight Large Language Model (LLM) named LLHKG and compares it with previous methods. Under our framework, the KG construction capability of lightweight LLM is comparable to GPT3.5.

Automated essay scoring (AES) is commonly evaluated on public benchmarks using quadratic weighted kappa (QWK). However, because benchmark labels are assigned by human raters and inevitably contain scoring errors, it remains unclear both what QWK is theoretically attainable and what level is practically sufficient for deployment. We therefore derive two dataset-specific QWK ceilings based on the reliability concept in classical test theory, which can be estimated from standard two-rater benchmarks without additional annotation. The first is the theoretical ceiling: the maximum QWK that an ideal AES model that perfectly predicts latent true scores can achieve under label noise. The second is the human-like ceiling: the QWK attainable by an AES model with human-level scoring error, providing a practical target when AES is intended to replace a single human rater. We further show that human--human QWK, often used as a ceiling reference, can underestimate the true ceiling. Simulation experiments validate the proposed ceilings, and experiments on real benchmarks illustrate how they clarify the current performance and remaining headroom of modern AES models.

Large language models have been extensively studied for emotion recognition and moral reasoning as distinct capabilities, yet the extent to which emotions influence moral judgment remains underexplored. In this work, we develop an emotion-induction pipeline that infuses emotion into moral situations and evaluate shifts in moral acceptability across multiple datasets and LLMs. We observe a directional pattern: positive emotions increase moral acceptability and negative emotions decrease it, with effects strong enough to reverse binary moral judgments in up to 20% of cases, and with susceptibility scaling inversely with model capability. Our analysis further reveals that specific emotions can sometimes behave contrary to what their valence would predict (e.g., remorse paradoxically increases acceptability). A complementary human annotation study shows humans do not exhibit these systematic shifts, indicating an alignment gap in current LLMs.

Existing detoxification methods for large language models mainly focus on post-training stage or inference time, while few tackle the source of toxicity, namely, the dataset itself. Such training-based or controllable decoding approaches cannot completely suppress the model's inherent toxicity, whereas detoxifying the pretraining dataset can fundamentally reduce the toxicity that the model learns during training. Hence, we attempt to detoxify directly on raw corpora with SoCD (Soft Contrastive Decoding), which guides an LLM to localize and rewrite toxic spans in raw data while preserving semantics, in our proposed HSPD (Hierarchical Semantic-Preserving Detoxification) pipeline, yielding a detoxified corpus that can drop-in replace the original for fine-tuning or other training. On GPT2-XL, HSPD attains state-of-the-art detoxification, reducing Toxicity Probability (TP) from 0.42 to 0.18 and Expected Maximum Toxicity (EMT) from 0.43 to 0.20. We further validate consistent best-in-class results on LLaMA2-7B, OPT-6.7B, and Falcon-7B. These findings show that semantics-preserving, corpus-level rewriting with HSPD effectively suppresses downstream toxicity while retaining data utility and allowing seamless source-level mitigation, thereby reducing the cost of later model behavior adjustment. (Code is available at: https://github.com/ntsw2001/data_detox_for_llm)

Modern distributed systems generate massive volumes of log data that are critical for detecting anomalies and cyber threats. However, in real world settings, these logs are often distributed across multiple organizations and cannot be centralized due to privacy and security constraints. Existing log anomaly detection methods, including recent large language model (LLM) based approaches, largely rely on centralized training and are not suitable for such environments. In this paper, we propose DP-FLogTinyLLM, a privacy preserving federated framework for log anomaly detection using parameter efficient LLMs. Our approach enables collaborative learning without sharing raw log data by integrating federated optimization with differential privacy. To ensure scalability in resource constrained environments, we employ low rank adaptation (LoRA) for efficient fine tuning of Tiny LLMs at each client. Empirical results on the Thunderbird and BGL datasets show that the proposed framework matches the performance of centralized LLM based methods, while incurring additional computational overhead due to privacy mechanisms. Compared to existing federated baselines, DP-FLogTinyLLM consistently achieves higher precision and F1-score, with particularly strong gains on the Thunderbird dataset, highlighting its effectiveness in detecting anomalies while minimizing false positives.

When a language model agrees with a user's false belief, is it failing to detect the error, or noticing and agreeing anyway? We show the latter. Across twelve open-weight models from five labs, spanning small to frontier scale, the same small set of attention heads carries a "this statement is wrong" signal whether the model is evaluating a claim on its own or being pressured to agree with a user. Silencing these heads flips sycophantic behavior sharply while leaving factual accuracy intact, so the circuit controls deference rather than knowledge. Edge-level path patching confirms that the same head-to-head connections drive sycophancy, factual lying, and instructed lying. Opinion-agreement, where no factual ground truth exists, reuses these head positions but writes into an orthogonal direction, ruling out a simple "truth-direction" reading of the substrate. Alignment training leaves this circuit in place: an RLHF refresh cuts sycophantic behavior roughly tenfold while the shared heads persist or grow, a pattern that replicates on an independent model family and under targeted anti-sycophancy DPO. When these models sycophant, they register that the user is wrong and agree anyway.

Generative answer engines expose content through selective citation rather than ranked retrieval, fundamentally altering how visibility is determined. This shift calls for new optimization methods beyond traditional search engine optimization. Existing generative engine optimization (GEO) approaches primarily rely on token-level text rewriting, offering limited interpretability and weak control over the trade-off between citation visibility and content quality. We propose FeatGEO, a feature-level, multi-objective optimization framework that abstracts webpages into interpretable structural, content, and linguistic properties. Instead of directly editing text, FeatGEO optimizes over this feature space and uses a language model to realize feature configurations into natural language, decoupling high-level optimization from surface-level generation. Experiments on GEO-Bench across three generative engines demonstrate that FeatGEO consistently improves citation visibility while maintaining or improving content quality, substantially outperforming token-level baselines. Further analyses show that citation behavior is more strongly influenced by document-level content properties than by isolated lexical edits, and that the learned feature configurations generalize across language models of different scales.

As a means to balance the growth of the AI industry with the need for privacy protection, machine unlearning plays a crucial role in realizing the ``right to be forgotten'' in artificial intelligence. This technique enables AI systems to remove the influence of specific data while preserving the rest of the learned knowledge. Although it has been actively studied, most existing unlearning methods assume that unlearning is performed only once. In this work, we evaluate existing unlearning algorithms in a more realistic scenario where unlearning is conducted repeatedly, and in this setting, we identify two critical phenomena: (1) Knowledge Erosion, where the accuracy on retain data progressively degrades over unlearning phases, and (2) Forgetting Reversal, where previously forgotten samples become recognizable again in later phases. To address these challenges, we propose SAFER (StAbility-preserving Forgetting with Effective Regularization), a continual unlearning framework that maintains representation stability for retain data while enforcing negative logit margins for forget data. Extensive experiments show that SAFER mitigates not only knowledge erosion but also forgetting reversal, achieving stable performance across multiple unlearning phases.

Extreme-edge scientific applications use machine learning models to analyze sensor data and make real-time decisions. Their stringent latency and throughput requirements demand small batch sizes and require that model weights remain fully on-chip. Spatial dataflow implementations are common for extreme-edge applications. Spatial dataflow works well for small networks, but it fails to scale to larger models due to inherent resource scaling limitations. AI Engines on modern FPGA SoCs offer a promising alternative with high compute density and additional on-chip memory. However, the architecture, programming model, and performance-scaling behavior of AI Engines differ fundamentally from those of the programmable logic, making direct comparison non-trivial and the benefits of using AI Engines unclear. This work addresses how and when extreme-edge scientific neural networks should be implemented on AI Engines versus programmable logic. We provide systematic architectural characterization and micro-benchmarking and introduce a latency-adjusted resource equivalence (LARE) metric that identifies when AI Engine implementations outperform programmable logic designs. We further propose spatial and API-level dataflow optimizations tailored to low-latency scientific inference. Finally, we demonstrate the successful deployment of end-to-end neural networks on AI Engines that cannot fit on programmable logic when using the hlsml toolchain.

Developing bipedal football robots in dynamiccombat environments presents challenges related to motionstability and deep coupling of multiple tasks, as well ascontrol switching issues between different states such as up-right walking and fall recovery. To address these problems,this paper proposes a modular reinforcement learning (RL)framework for achieving adaptive multi-task control. Firstly,this framework combines an open-loop feedforward oscilla-tor with a reinforcement learning-based feedback residualstrategy, effectively separating the generation of basic gaitsfrom complex football actions. Secondly, a posture-driven statemachine is introduced, clearly switching between the ballseeking and kicking network (BSKN) and the fall recoverynetwork (FRN), fundamentally preventing state interference.The FRN is efficiently trained through a progressive forceattenuation curriculum learning strategy. The architecture wasverified in Unity simulations of bipedal robots, demonstratingexcellent spatial adaptability-reliably finding and kicking theball even in restricted corner scenarios-and rapid autonomousfall recovery (with an average recovery time of 0.715 seconds).This ensures seamless and stable operation in complex multi-task environments.

Learning diverse locomotion skills for humanoid robots in a unified reinforcement learning framework remains challenging due to the conflicting requirements of stability and dynamic expressiveness across different gaits. We present a multi-gait learning approach that enables a humanoid robot to master five distinct gaits -- walking, goose-stepping, running, stair climbing, and jumping -- using a consistent policy structure, action space, and reward formulation. The key contribution is a selective Adversarial Motion Prior (AMP) strategy: AMP is applied to periodic, stability-critical gaits (walking, goose-stepping, stair climbing) where it accelerates convergence and suppresses erratic behavior, while being deliberately omitted for highly dynamic gaits (running, jumping) where its regularization would over-constrain the motion. Policies are trained via PPO with domain randomization in simulation and deployed on a physical 12-DOF humanoid robot through zero-shot sim-to-real transfer. Quantitative comparisons demonstrate that selective AMP outperforms a uniform AMP policy across all five gaits, achieving faster convergence, lower tracking error, and higher success rates on stability-focused gaits without sacrificing the agility required for dynamic ones.

Education is not merely the transmission of information or the optimisation of individual performance; it is a fundamentally social, constructive, and relational practice. However, recent advances in generative artificial intelligence (GenAI) increasingly emphasise efficiency, automation, and individualised assistance, risking the weakening of relational learning processes. Despite growing adoption, AI in education (AIED) research has yet to fully articulate how AI can be designed in ways that sustain the social and ecological relationships through which learning occurs. In this paper, we re-centre education as relational and frame learner-AI interactions as context-specific relationships with clearly defined purposes and boundaries, rather than positioning them as substitutes for, or replacements of, human interaction. Grounded in participatory design practices and inspired by Indigenous worldviews (including Aboriginal Australian, Native American, and Mesoamerican traditions) that foreground reciprocity and relational accountability, we argue that meaningful educational AI should support learning with others rather than replace them. We advance this perspective by: i) conceptualising AIED as a relational design problem grounded in reciprocity; ii) articulating key tensions introduced by GenAI in education; and iii) outlining design directions that expand the AIED design space toward reciprocity, including when not to use AI, how to define pedagogical boundaries, and how to support responsible uses of AIED innovations that sustain communities and natural environments.

English financial NLP has progressed rapidly through benchmarks for sentiment, document understanding, and financial question answering, while Arabic financial NLP remains comparatively under-explored despite strong practical demand for trustworthy finance and Islamic-finance assistants. We introduce SAHM, a document-grounded benchmark and instruction-tuning dataset for Arabic financial NLP and Shari'ah-compliant reasoning. SAHM contains 14,380 expert-verified instances spanning seven tasks: AAOIFI standards QA, fatwa-based QA/MCQ, accounting and business exams, financial sentiment analysis, extractive summarization, and event-cause reasoning, curated from authentic regulatory, juristic, and corporate sources. We evaluate 19 strong open and proprietary LLMs using task-specific metrics and rubric-based scoring for open-ended outputs, and find that Arabic fluency does not reliably translate to evidence-grounded financial reasoning: models are substantially stronger on recognition-style tasks than on generation and causal reasoning, with the largest gaps on event-cause reasoning. We release the benchmark, evaluation framework, and an instruction-tuned model to support future research on trustworthy Arabic financial NLP.

Multi-modal test-time adaptation (TTA) enhances the resilience of benchmark multi-modal models against distribution shifts by leveraging the unlabeled target data during inference. Despite the documented success, the advancement of multi-modal TTA methodologies has been impeded by a persistent limitation, i.e., the lack of explicit modeling of category-conditional distributions, which is crucial for yielding accurate predictions and reliable decision boundaries. Canonical Gaussian discriminant analysis (GDA) provides a vanilla modeling of category-conditional distributions and achieves moderate advancement in uni-modal contexts. However, in multi-modal TTA scenario, the inherent modality distribution asymmetry undermines the effectiveness of modeling the category-conditional distribution via the canonical GDA. To this end, we introduce a tailored probabilistic Gaussian model for multi-modal TTA to explicitly model the category-conditional distributions, and further propose an adaptive contrastive asymmetry rectification technique to counteract the adverse effects arising from modality asymmetry, thereby deriving calibrated predictions and reliable decision boundaries. Extensive experiments across diverse benchmarks demonstrate that our method achieves state-of-the-art performance under a wide range of distribution shifts. The code is available at https://github.com/XuJinglinn/AdaPGC.

Recent advances in large-scale video world models have enabled increasingly realistic future prediction, raising the prospect of leveraging imagined videos for robot learning. However, visual realism does not imply physical plausibility, and behaviors inferred from generated videos may violate dynamics and fail when executed by embodied agents. Existing benchmarks begin to incorporate notions of physical plausibility, but they largely remain perception- or diagnostic-oriented and do not systematically evaluate whether predicted behaviors can be translated into executable actions that complete the intended task. To address this gap, we introduce RoboWM-Bench, a manipulation-centric benchmark for embodiment-grounded evaluation of video world models. RoboWM-Bench converts generated behaviors from both human-hand and robotic manipulation videos into embodied action sequences and validates them through robotic execution. The benchmark spans diverse manipulation scenarios and establishes a unified protocol for consistent and reproducible evaluation. Using RoboWM-Bench, we evaluate state-of-the-art video world models and find that reliably generating physically executable behaviors remains an open challenge. Common failure modes include errors in spatial reasoning, unstable contact prediction, and non-physical deformations. While finetuning on manipulation data yields improvements, physical inconsistencies still persist, suggesting opportunities for more physically grounded video generation for robots.

Estimating the number of components is a fundamental challenge in unsupervised learning, particularly when dealing with high-dimensional data with many components or severely imbalanced component sizes. This paper addresses this challenge for classical Gaussian mixture models. The proposed estimator is simple: center the data, compute the singular values of the centered matrix, and count those above a threshold. No iterative fitting, no likelihood calculation, and no prior knowledge of the number of components are required. We prove that, under a mild separation condition on the component centers, the estimator consistently recovers the true number of components. The result holds in high-dimensional settings where the dimension can be much larger than the sample size. It also holds when the number of components grows to the smaller of the dimension and the sample size, even under severe imbalance among component sizes. Computationally, the method is extremely fast: for example, it processes ten million samples in one hundred dimensions within one minute. Extensive experimental studies confirm its accuracy in challenging settings such as high dimensionality, many components, and severe class imbalance.

Large language models (LLMs) require frequent knowledge updates to reflect changing facts and mitigate hallucinations. To meet this demand, lifelong knowledge editing has emerged as a continual approach to modify specific pieces of knowledge without retraining the entire model. Existing parameter editing methods struggle with stability during sequential edits due to catastrophic forgetting. While retrieval-based approaches are proposed to alleviate this issue, their applicability remains limited across various datasets because of high training costs. To address these limitations and enhance scalability in lifelong settings, we propose LightEdit. Our framework first selects relevant knowledge from retrieved information to modify the query effectively. It then incorporates a decoding strategy to suppress the model's original knowledge probabilities, thereby enabling efficient edits based on the selected information. Extensive experiments on ZSRE, Counterfact, and RIPE benchmarks demonstrate that LightEdit outperforms existing lifelong knowledge editing methods. Furthermore, by minimizing training costs, LightEdit achieves cost-effective scalability, enabling easy adaptation to various datasets.

We introduce Options LLM (OLLM), a simple, general method that replaces the single next-token prediction of standard LLMs with a \textit{set of learned options} for the next token, indexed by a discrete latent variable. Instead of relying on temperature or sampling heuristics to induce diversity, OLLM models variation explicitly: a small latent space parametrizes multiple plausible next-token options which can be selected or searched by a downstream policy. Architecturally, OLLM is a lightweight "plug-in" that inserts two layers: an encoder and a decoder, before the output head, allowing almost any pretrained LLM to be converted with minimal additional parameters. We apply OLLM to a 1.7B-parameter backbone (only $1.56\%$ of parameters trainable) trained on OpenMathReasoning and evaluated on OmniMath. The SOTA LoRA-adapted baselines peak at $51\%$ final answer correctness, while OLLM's option set allows up to $\sim 70\%$ under optimal latent selection. We then train a compact policy in the latent space that emits latents to control generation. Operating in a low-dimensional option space makes reward optimization far more sample-efficient and substantially reduces common misalignments (e.g., language switching or degenerate reasoning), as the policy is constrained to options learned during SFT. Crucially, this alignment arises from model structure rather than additional KL or handcrafted alignment losses. Our results demonstrate that optionized next-token modeling enhances controllability, robustness, and efficiency in math reasoning, and highlight latent-space policy learning as a promising direction for reinforcement learning in LLMs.

Code LLMs often portray inconsistent program behaviors. Developers typically employ benchmarks to assess Code LLMs, but most benchmarks are hand-crafted, static and do not target consistency property. In this work, we pose the scientific question: how can we automatically discover inconsistent program behaviors in Code LLMs? To address this challenge, we propose an automated consistency testing method, called MUCOCO, which employs semantic-preserving mutation analysis to expose inconsistent behaviors in code LLMs. Given a coding query, MUCOCO automatically transforms its program into semantically equivalent programs (aka mutants) and detects inconsistencies between the mutants and the original program (e.g., different output or test failure). We evaluate MUCOCO using four (4) coding tasks and seven (7) LLMs. Results show that MUCOCO is effective in exposing inconsistency and outperforms the closest baseline (TURBULENCE). About one in seven (15%) inputs generated by MUCOCO exposed inconsistencies. Our work motivates the need to test Code LLMs for consistency property

Multimodal Large Language Models (MLLMs) have achieved remarkable success in cross-modal understanding and generation, yet their deployment is threatened by critical safety vulnerabilities. While prior works have demonstrated the feasibility of backdoors in MLLMs via fine-tuning data poisoning to manipulate inference, the underlying mechanisms of backdoor attacks remain opaque, complicating the understanding and mitigation. To bridge this gap, we propose ProjLens, an interpretability framework designed to demystify MLLMs backdoors. We first establish that normal downstream task alignment--even when restricted to projector fine--tuning--introduces vulnerability to backdoor injection, whose activation mechanism is different from that observed in text-only LLMs. Through extensive experiments across four backdoor variants, we uncover:(1) Low-Rank Structure: Backdoor injection updates appear overall full-rank and lack dedicated ``trigger neurons'', but the backdoor-critical parameters are encoded within a low-rank subspace of the projector;(2) Activation Mechanism: Both clean and poisoned embedding undergoes a semantic shift toward a shared direction aligned with the backdoor target, but the shifting magnitude scales linearly with the input norm, resulting in the distinct backdoor activation on poisoned samples. Our code is available at: https://anonymous.4open.science/r/ProjLens-8FD7

Multi-window mobile scenarios, such as split-screen and foldable modes, make GUI display defects more likely by forcing applications to adapt to changing window sizes and dynamic layout reflow. Existing detection techniques are limited in two ways: they are largely passive, analyzing screenshots only after problematic states have been reached, and they are mainly designed for conventional full-screen interfaces, making them less effective in multi-window settings.We propose an end-to-end framework for GUI display defect detection in multi-window mobile scenarios. The framework proactively triggers split-screen, foldable, and window-transition states during app exploration, uses Set-of-Mark (SoM) to align screenshots with widget-level interface elements, and leverages multimodal large language models with chain-of-thought prompting to detect, localize, and explain display defects. We also construct a benchmark of GUI display defects using 50 real-world Android applications.Experimental results show that multi-window settings substantially increase the exposure of layout-related defects, with text truncation increasing by 184% compared with conventional full-screen settings. At the application level, our method detects 40 defect-prone apps with a false positive rate of 10.00% and a false negative rate of 11.11%, outperforming OwlEye and YOLO-based baselines. At the fine-grained level, it achieves the best F1 score of 87.2% for widget occlusion detection.

Unification of automatic speech recognition (ASR) systems reduces development and maintenance costs, but training a single model to perform well in both offline and low-latency streaming settings remains challenging. We present a Unified ASR framework for Transducer (RNNT) training that supports both offline and streaming decoding within a single model, using chunk-limited attention with right context and dynamic chunked convolutions. To further close the gap between offline and streaming performance, we introduce an efficient Triton implementation of mode-consistency regularization for RNNT (MCR-RNNT), which encourages agreement across training modes. Experiments show that the proposed approach improves streaming accuracy at low latency while preserving offline performance and scaling to larger model sizes and training datasets. The proposed Unified ASR framework and the English model checkpoint are open-sourced.

Semi-supervised learning with manifold regularization is a classical framework for jointly learning from both labeled and unlabeled data, where the key requirement is that the support of the unknown marginal distribution has the geometric structure of a Riemannian manifold. Typically, the Laplace-Beltrami operator-based manifold regularization can be approximated empirically by the Laplacian regularization associated with the entire training data and its corresponding graph Laplacian matrix. However, the graph Laplacian matrix depends heavily on the prespecified similarity metric and may lead to inappropriate penalties when dealing with redundant or noisy input variables. To address the above issues, this paper proposes a new \textit{Semi-Supervised Meta Additive Model (S$^2$MAM) based on a bilevel optimization scheme that automatically identifies informative variables, updates the similarity matrix, and simultaneously achieves interpretable predictions. Theoretical guarantees are provided for S$^2$MAM, including the computing convergence and the statistical generalization bound. Experimental assessments across 4 synthetic and 12 real-world datasets, with varying levels and categories of corruption, validate the robustness and interpretability of the proposed approach.

Evaluating the writing capabilities of large language models (LLMs) remains a significant challenge due to the multidimensional nature of writing skills and the limitations of existing metrics. LLM's performance in thousand-words level and open-ended writing is inadequately assessed by traditional reference-based metrics or modern LLM-as-a-judge methods. We propose Tree-of-Writing (ToW), to resolve the implicit inconsistency often found when LLM-as-a-judge aggregates all sub-features in text evaluation. ToW incorporates a tree-structured workflow by explicitly modeling the aggregation weights of sub-features. We also present HowToBench, a large-scale Chinese writing benchmark encompassing 12 genres and 1302 instructions across three task categories: contextual completion, outline-guided writing, and open-ended generation. ToW successfully mitigates the biases, achieving a 0.93 Pearson correlation with human judgments. Furthermore, we detect that both overlap-based text generation metrics and popular LLM-as-a-judge practices are vulnerable to textual disturbances, while ToW is robust to them. We also uncover a negative correlation between input length and content-related scores in the Guide task, showcasing that it cannot be simply improved by input-side information piling.

Zero-shot reasoning on text-rich networks (TRNs) remains a challenging frontier, as models must integrate textual semantics with relational structure without task-specific supervision. While graph neural networks rely on fixed label spaces and supervised objectives, recent large language model (LLM)-based approaches often overlook graph context or depend on distillation from larger models, limiting generalisation. We propose TRN-R1-Zero, a post-training framework for TRN reasoning trained solely via reinforcement learning. TRN-R1-Zero directly optimises base LLMs using a Neighbour-aware Group Relative Policy Optimisation objective that dynamically adjusts rewards based on a novel margin gain metric for the informativeness of neighbouring signals, effectively guiding the model toward relational reasoning. Unlike prior methods, TRN-R1-Zero requires no supervised fine-tuning or chain-of-thought data generated from large reasoning models. Extensive experiments across citation, hyperlink, social and co-purchase TRN benchmarks demonstrate the superiority and robustness of TRN-R1-Zero. Moreover, relying strictly on node-level training, TRN-R1-Zero achieves zero-shot inference on edge- and graph-level tasks, extending beyond cross-domain transfer. The codebase is publicly available at https://github.com/superallen13/TRN-R1-Zero.

Neural NLI models overfit dataset artifacts instead of truly reasoning. A hypothesis-only model gets 57.7% in SNLI, showing strong spurious correlations, and 38.6% of the baseline errors are the result of these artifacts. We propose Product-of-Experts (PoE) training, which downweights examples where biased models are overconfident. PoE nearly preserves accuracy (89.10% vs. 89.30%) while cutting bias reliance by 4.71% (bias agreement 49.85% to 45%). An ablation finds lambda = 1.5 that best balances debiasing and accuracy. Behavioral tests still reveal issues with negation and numerical reasoning.

Physical activity (PA) is widely recognized as protective against mental distress, yet whether this benefit varies systematically across population subgroups remains poorly understood. Using pooled data from ten consecutive annual waves of the U.S. Behavioral Risk Factor Surveillance System (2015-2024; n = 3,242,218), we investigate heterogeneity in the association between leisure-time PA and frequent mental distress (FMD, >=14 days/month) across age groups. Survey-weighted logistic regression reveals a striking age gradient: the adjusted odds ratio for PA ranges from 0.89 among young adults (18-24) to 0.50 among adults aged 55-64, with the protective association strengthening monotonically with age. Temporal analysis across all ten years shows that the young-adult PA effect has been eroding over the past decade, with the 18-24 OR reaching 1.01 (null) in both 2018 and 2024 -- paralleling the deepening youth mental health crisis. Causal Forest via Double Machine Learning independently identifies age as the dominant driver of treatment effect heterogeneity (feature importance = 0.39, 2.5x the next predictor). E-value sensitivity analysis, propensity score overlap checks, placebo tests, and imputation comparisons confirm the robustness of the findings. These results suggest that the well-documented exercise--mental health link may not generalize to the youngest adult population, whose distress appears increasingly driven by stressors that PA alone cannot mitigate.

Accurate disease classification from radiology reports is essential for many applications. While supervised fine-tuning (SFT) of lightweight LLMs improves accuracy, it can degrade reasoning. We propose a two-stage approach: SFT on disease labels followed by Group Relative Policy Optimization (GRPO) to refine predictions by optimizing accuracy and format without reasoning supervision. Across three radiologist-annotated datasets, SFT outperformed baselines and GRPO further improved classification and enhanced reasoning recall and comprehensiveness.

Understanding a discourse requires tracking entities and the relations that hold between them. While Large Language Models (LLMs) perform well on relational reasoning, the mechanism by which they bind entities, relations, and attributes remains unclear. We study discourse-level relational binding and show that LLMs encode it via a Cell-based Binding Representation (CBR): a low-dimensional linear subspace in which each ``cell'' corresponds to an entity--relation index pair, and bound attributes are retrieved from the corresponding cell during inference. Using controlled multi-sentence data annotated with entity and relation indices, we identify the CBR subspace by decoding these indices from attribute-token activations with Partial Least Squares regression. Across domains and two model families, the indices are linearly decodable and form a grid-like geometry in the projected space. We further find that context-specific CBR representations are related by translation vectors in activation space, enabling cross-context transfer. Finally, activation patching shows that manipulating this subspace systematically changes relational predictions and that perturbing it disrupts performance, providing causal evidence that LLMs rely on CBR for relational binding.

LLM-assisted defect discovery has a precision crisis: plausible-but-wrong reports overwhelm maintainers and degrade credibility for real findings. We present Refute-or-Promote, an inference-time reliability pattern combining Stratified Context Hunting (SCH) for candidate generation, adversarial kill mandates, context asymmetry, and a Cross-Model Critic (CMC). Adversarial agents attempt to disprove candidates at each promotion gate; cold-start reviewers are intended to reduce anchoring cascades; cross-family review can catch correlated blind spots that same-family review misses. Over a 31-day campaign across 7 targets (security libraries, the ISO C++ standard, major compilers), the pipeline killed roughly 79% of 171 candidates before advancing to disclosure (retrospective aggregate); on a consolidated-protocol subset (lcms2, wolfSSL; n=30), the prospective kill rate was 83%. Outcomes: 4 CVEs (3 public, 1 embargoed); LWG 4549 accepted to the C++ working paper; 5 merged C++ editorial PRs; 3 compiler conformance bugs; 8 merged security-related fixes without CVE; an RFC 9000 errata filed under committee review; and 1+ FIPS 140-3 normative compliance issues under coordinated disclosure -- all evaluated by external acceptance, not benchmarks. The most instructive failure: ten dedicated reviewers unanimously endorsed a non-existent Bleichenbacher padding oracle in OpenSSL's CMS module; it was killed only by a single empirical test, motivating the mandatory empirical gate. No vulnerability was discovered autonomously; the contribution is external structure that filters LLM agents' persistent false positives. As a preliminary transfer test beyond defect discovery, a simplified cross-family critique variant also solved five previously unsolved SymPy instances on SWE-bench Verified and one SWE-rebench hard task.

The combination of Mixture-of-Experts (MoE) and Low-Rank Adaptation (LoRA) has shown significant potential for enhancing the multi-task learning capabilities of Large Language Models. However, existing methods face two primary challenges: (1)Imprecise Routing in the current MoE-LoRA method fails to explicitly match input semantics with expert capabilities, leading to weak expert specialization. (2)Uniform weight fusion strategies struggle to provide adaptive update strengths, overlooking the varying complexity of different tasks. To address these limitations, we propose SAMoRA (Semantic-Aware Mixture of LoRA Experts), a novel parameter-efficient fine-tuning framework tailored for task-adaptive learning. Specifically, A Semantic-Aware Router is proposed to explicitly align textual semantics with the most suitable experts for precise routing. A Task-Adaptive Scaling mechanism is designed to regulate expert contributions based on specific task requirements dynamically. In addition, a novel regularization objective is proposed to jointly promote expert specialization and effective scaling. Extensive experiments on multiple multi-task benchmarks demonstrate that SAMoRA significantly outperforms the state-of-the-art methods and holds excellent task generalization capabilities. Code is available at https://github.com/boyan-code/SAMoRA

Existing QA benchmarks typically assume distinct documents with minimal overlap, yet real-world retrieval-augmented generation (RAG) systems operate on corpora such as financial reports, legal codes, and patents, where information is highly redundant and documents exhibit strong inter-document similarity. This mismatch undermines evaluation validity: retrievers can be unfairly undervalued even when they retrieve documents that provide sufficient evidence, because redundancy across documents is not accounted for in evaluation. On the other hand, retrievers that perform well on standard benchmarks often generalize poorly to real-world corpora with highly similar and redundant documents. We present RARE (Redundancy-Aware Retrieval Evaluation), a framework for constructing realistic benchmarks by (i) decomposing documents into atomic facts to enable precise redundancy tracking and (ii) enhancing LLM-based data generation with CRRF. RAG benchmark data usually requires multiple quality criteria, but LLMs often yield trivial outputs. CRRF scores criteria separately and fuses decisions by rank, improving the reliability of generated data. Applying RARE to Finance, Legal, and Patent corpora, we introduce RedQA, where a strong retriever baseline drops from 66.4% PerfRecall@10 on 4-hop General-Wiki to 5.0-27.9% PerfRecall@10 at 4-hop depth, revealing robustness gaps that current benchmarks fail to capture. RARE enables practitioners to build domain-specific RAG evaluations that faithfully reflect real-world deployment conditions.

Efficient construction of models capturing the preconditions and effects of actions is essential for applying AI planning in real-world domains. Extensive prior work has explored learning such models from high-level descriptions of state and/or action sequences. In this paper, we tackle a more challenging setting: learning lifted action models from sequences of state images, without action observation. We propose a deep learning framework that jointly learns state prediction, action prediction, and a lifted action model. We also introduce a mixed-integer linear program (MILP) to prevent prediction collapse and self-reinforcing errors among predictions. The MILP takes the predicted states, actions, and action model over a subset of traces and solves for logically consistent states, actions, and action model that are as close as possible to the original predictions. Pseudo-labels extracted from the MILP solution are then used to guide further training. Experiments across multiple domains show that integrating MILP-based correction helps the model escape local optima and converge toward globally consistent solutions.

Defeasible statements are statements that are likely, or probable, or usually true, but may occasionally be false. Plausible reasoning makes conclusions from statements that are either facts or defeasible statements without using numbers. So there are no probabilities or suchlike involved. Seventeen principles of logics that do plausible reasoning are suggested and several important plausible reasoning examples are considered. There are 14 necessary principles and 3 desirable principles, one of which is not formally stated. A first-order logic, called Plausible Logic (PL), is defined that satisfies all but two of the desirable principles and reasons correctly with all the examples. As far as we are aware, this is the only such logic. PL has 8 reasoning algorithms because, from a given plausible reasoning situation, there are different sensible conclusions. This article is a condensation of my book `Plausible Reasoning and Plausible Logic' (PRPL), which is to be submitted. Each section of this article corresponds to a chapter in PRPL, and vice versa. The proofs of all the results are in PRPL, so they are omitted in this article.

In gradient-based learning, a step size chosen in parameter units does not produce a predictable per-step change in function output. This often leads to instability in the streaming setting (i.e., batch size=1), where stochasticity is not averaged out and update magnitudes can momentarily become arbitrarily big or small. Instead, we propose intentional updates: first specify the intended outcome of an update and then solve for the step size that approximately achieves it. This strategy has precedent in online supervised linear regression via Normalized Least Mean Squares algorithm, which selects a step size to yield a specified change in the function output proportional to the current error. We extend this principle to streaming deep reinforcement learning by defining appropriate intended outcomes: Intentional TD aims for a fixed fractional reduction of the TD error, and Intentional Policy Gradient aims for a bounded per-step change in the policy, limiting local KL divergence. We propose practical algorithms combining eligibility traces and diagonal scaling. Empirically, these methods yield state-of-the-art streaming performance, frequently performing on par with batch and replay-buffer approaches.

One of the most challenging problems in graph machine learning is generalizing across graphs with diverse properties. Graph neural networks (GNNs) face a fundamental limitation: they require separate training for each new graph, preventing universal generalization across diverse graph datasets. A critical challenge facing GNNs lies in their reliance on labeled training data for each individual graph, a requirement that hinders the capacity for universal node classification due to the heterogeneity inherent in graphs -- differences in homophily levels, community structures, and feature distributions across datasets. Inspired by the success of large language models (LLMs) that achieve in-context learning through massive-scale pre-training on diverse datasets, we introduce NodePFN. This universal node classification method generalizes to arbitrary graphs without graph-specific training. NodePFN learns posterior predictive distributions (PPDs) by training only on thousands of synthetic graphs generated from carefully designed priors. Our synthetic graph generation covers real-world graphs through the use of random networks with controllable homophily levels and structural causal models for complex feature-label relationships. We develop a dual-branch architecture combining context-query attention mechanisms with local message passing to enable graph-aware in-context learning. Extensive evaluation on 23 benchmarks demonstrates that a single pre-trained NodePFN achieves 71.27 average accuracy. These results validate that universal graph learning patterns can be effectively learned from synthetic priors, establishing a new paradigm for generalization in node classification.

Safe Reinforcement Learning from Human Feedback (Safe RLHF) has recently achieved empirical success in developing helpful and harmless large language models by decoupling human preferences regarding helpfulness and harmlessness. Existing approaches typically rely on fitting fixed horizon reward models from human feedback and have only been validated empirically. In this paper, we formulate safe RLHF as an infinite horizon discounted Con- strained Markov Decision Process (CMDP), since humans may interact with the model over a continuing sequence of interactions rather than within a single finite episode. We propose two Safe RLHF algorithms that do not require reward model fitting and, in contrast to prior work assuming fixed-length trajectories, support flexible trajectory lengths for training. Both algo- rithms are based on the primal-dual method and achieve global convergence guarantees with polynomial rates in terms of policy gradient iterations, trajectory sample lengths, and human preference queries. To the best of our knowledge, this is the first work to study infinite horizon discounted CMDP under human feedback and establish global, non-asymptotic convergence.

A major challenge for niche scientific and technical domains in leveraging coding agents is the lack of access to up-to-date, domain- specific knowledge. Foundational models often demonstrate limited reasoning capabilities in specialized fields and cannot inherently incorporate knowledge that evolves through ongoing research and experimentation. Materials scientists exploring novel compounds, communication engineers designing and evaluating new protocols, and bioengineering researchers conducting iterative experiments all face this limitation. These experts typically lack the resources to fine-tune large models or continuously embed new findings, creating a barrier to adopting AI-driven coding agents. To address this, we introduce a framework that gives coding agents instanta- neous access to research repositories and technical documentation, enabling real-time, context-aware operation. Our open-source im- plementation allows users to upload documents via doc-search.dev and includes zed-fork, which enforces domain-specific rules and workflows. Together, these tools accelerate the integration of coding agents into specialized scientific and technical workflows

Linear attention mechanisms have emerged as promising alternatives to softmax attention, offering linear-time complexity during inference. Recent advances such as Gated DeltaNet (GDN) and Kimi Delta Attention (KDA) have demonstrated that the delta rule, an online gradient descent update, enables superior associative recall compared to simple additive updates. While KDA refined the coarse head-wise decay gate into channel-wise decay, the learning rate $β_t$ in the delta update remains a scalar, limiting the model's capacity for dimension-specific adaptation. We introduce FG$^2$-GDN, which replaces the scalar $β_t$ with a channel-wise vector analogous to the transition from SGD to per-coordinate adaptive optimizers such as AdaGrad and Adam. We further propose FG$^2$-GDN+, which decouples the scaling for keys and values, enabling independent control of erasure strength and write strength. Experiments on synthetic and real-world benchmarks show that FG$^2$-GDN and its variant improve associative recall and long-context understanding over GDN and KDA, with comparable computational efficiency.

Inference-time LLM alignment methods, particularly activation steering, offer an alternative to fine-tuning by directly modifying activations during generation. Existing methods, however, often rely on non-anticipative interventions that ignore how perturbations propagate through transformer layers and lack online error feedback, resulting in suboptimal, open-loop control. To address this, we show empirically that, despite the nonlinear structure of transformer blocks, layer-wise dynamics across multiple LLM architectures and scales are well-approximated by locally-linear models. Exploiting this property, we model LLM inference as a linear time-varying dynamical system and adapt the classical linear quadratic regulator to compute feedback controllers using layer-wise Jacobians, steering activations toward desired semantic setpoints in closed-loop with minimal computational overhead and no offline training. We also derive theoretical bounds on setpoint tracking error, enabling formal guarantees on steering performance. Using a novel adaptive semantic feature setpoint signal, our method yields robust, fine-grained behavior control across models, scales, and tasks, including state-of-the-art modulation of toxicity, truthfulness, refusal, and arbitrary concepts, surpassing baseline steering methods. Our code is available at: https://github.com/trustworthyrobotics/lqr-activation-steering

Cultural alignment in Large Language Models (LLMs) is essential for producing contextually aware, respectful, and trustworthy outputs. Without it, models risk generating stereotyped, insensitive, or misleading responses that fail to reflect cultural diversity w.r.t Helpful, Harmless, and Honest (HHH) paradigm. Existing benchmarks represent early steps toward cultural alignment; yet, no benchmarks currently enables systematic evaluation of cultural alignment in line with UNESCO's principles of cultural diversity w.r.t HHH paradigm. Therefore, to address this gap, we built Align-Cultura, two-stage pipeline for cultural alignment. Stage I constructs CULTURAX, the HHH-English dataset grounded in the UNESCO cultural taxonomy, through Query Construction, which reclassifies prompts, expands underrepresented domains (or labels), and prevents data leakage with SimHash. Then, Response Generation pairs prompts with culturally grounded responses via two-stage rejection sampling. The final dataset contains 1,500 samples spanning 30 subdomains of tangible and intangible cultural forms. Stage II benchmarks CULTURAX on general-purpose models, culturally fine-tuned models, and open-weight LLMs (Qwen3-8B and DeepSeek-R1-Distill-Qwen-7B). Empirically, culturally fine-tuned models improve joint HHH by 4%-6%, reduce cultural failures by 18%, achieve 10%-12% efficiency gains, and limit leakage to 0.3%.

Federated fine-tuning of Large Language Models (LLMs) is obstructed by a trilemma of challenges: protecting LLMs intellectual property (IP), ensuring client privacy, and mitigating performance loss on heterogeneous data. Existing methods like Offsite-Tuning (OT) secure the LLMs IP by having clients train only lightweight adapters, yet our analysis reveals they suffer from a fundamental performance bottleneck, leaving a significant gap compared to centralized training. To bridge this gap, we introduce FedProxy, a new federated adaptation framework. FedProxy replaces weak adapters with a unified, powerful Proxy Small Language Model (SLM), compressed from the proprietary LLM, to serve as a high-fidelity surrogate for collaborative fine-tuning. Our framework systematically resolves the trilemma through a three-stage architecture: (i) Efficient Representation via server-guided compression to create a resource-friendly proxy; (ii) Robust Optimization through an interference-mitigating aggregation strategy to handle data heterogeneity; and (iii) Effortless Fusion via a training-free "plug-in" mechanism to integrate learned knowledge back into the LLM. Experiments show FedProxy significantly outperforms OT methods and approaches centralized performance, establishing a new benchmark for secure and high-performance federated LLM adaptation.

Deep learning for vulnerability detection has shown promising results on early benchmarks, but recent evaluations reveal catastrophic degradation: models achieving F1 > 0.68 on legacy datasets collapse to 0.031 under strict deduplication. We identify the root cause as the semantic ambiguity problem: identical code can be secure or vulnerable depending on project-specific behavioral contracts, rendering global classification fundamentally inadequate. We propose Phoenix, a training-free multi-agent framework that resolves this ambiguity through Behavioral Contract Synthesis. Phoenix decomposes detection into three stages: a Semantic Slicer extracting minimal vulnerability-relevant context, a Requirement Reverse Engineer synthesizing Gherkin behavioral specifications encoding the security contract, and a Contract Judge evaluating code against these specifications via strict compliance checking. On PrimeVul Paired, Phoenix achieves F1 = 0.825 and Pair-Correct = 64.4%, surpassing RASM-Vul (F1 = 0.668) and VulTrial (F1 = 0.563) while using open-source models up to 48x smaller (7-14B vs. 671B). Ablation across 25 configurations demonstrates Gherkin specifications as the decisive driver (+0.09 to +0.35 F1). Error analysis reveals 18% of "False Positives" identify genuine security concerns in patched code, demonstrating that security is a relative property defined against behavioral contracts, not an absolute property of code syntax.

Trajectory Optimization (TO) solvers exploit known system dynamics to compute locally optimal trajectories through iterative improvements. A downside is that each new problem instance is solved independently; therefore, convergence speed and quality of the solution found depend on the initial trajectory proposed. To improve efficiency, a natural approach is to warm-start TO with initial guesses produced by a learned policy trained on trajectories previously generated by the solver. Diffusion-based policies have recently emerged as expressive imitation learning models, making them promising candidates for this role. Yet, a counterintuitive challenge comes from the local optimality of TO demonstrations: when a policy is rolled out, small non-optimal deviations may push it into situations not represented in the training data, triggering compounding errors over long horizons. In this work, we focus on learning-based warm-starting for gradient-based TO solvers that also provide feedback gains. Exploiting this specificity, we derive a first-order loss for Sobolev learning of diffusion-based policies using both trajectories and feedback gains. Through comprehensive experiments, we demonstrate that the resulting policy avoids compounding errors, and so can learn from very few trajectories to provide initial guesses reducing solving time by $2\times$ to $20 \times$. Incorporating first-order information enables predictions with fewer diffusion steps, reducing inference latency.

Diffusion distillation, exemplified by Distribution Matching Distillation (DMD), has shown great promise in few-step generation but often sacrifices quality for sampling speed. While integrating Reinforcement Learning (RL) into distillation offers potential, a naive fusion of these two objectives relies on suboptimal raw sample evaluation. This sample-based scoring creates inherent conflicts with the distillation trajectory and produces unreliable rewards due to the noisy nature of early-stage generation. To overcome these limitations, we propose GDMD, a novel framework that redefines the reward mechanism by prioritizing distillation gradients over raw pixel outputs as the primary signal for optimization. By reinterpreting the DMD gradients as implicit target tensors, our framework enables existing reward models to directly evaluate the quality of distillation updates. This gradient-level guidance functions as an adaptive weighting that synchronizes the RL policy with the distillation objective, effectively neutralizing optimization divergence. Empirical results show that GDMD sets a new SOTA for few-step generation. Specifically, our 4-step models outperform the quality of their multi-step teacher and substantially exceed previous DMDR results in GenEval and human-preference metrics, exhibiting strong scalability potential.

Half-truths, claims that are factually correct yet misleading due to omitted context, remain a blind spot for fact verification systems focused on explicit falsehoods. Addressing such omission-based manipulation requires reasoning not only about what is said, but also about what is left unsaid. We propose RADAR, a role-anchored multi-agent debate framework for omission-aware fact verification under realistic, noisy retrieval. RADAR assigns complementary roles to a Politician and a Scientist, who reason adversarially over shared retrieved evidence, moderated by a neutral Judge. A dual-threshold early termination controller adaptively decides when sufficient reasoning has been reached to issue a verdict. Experiments show that RADAR consistently outperforms strong single- and multi-agent baselines across datasets and backbones, improving omission detection accuracy while reducing reasoning cost. These results demonstrate that role-anchored, retrieval-grounded debate with adaptive control is an effective and scalable framework for uncovering missing context in fact verification. The code is available at https://github.com/tangyixuan/RADAR.

Large reasoning models (LRMs) produce complex, multi-step reasoning traces, yet safety evaluation remains focused on final outputs, overlooking how harm emerges during reasoning. When jailbroken, harm does not appear instantaneously but unfolds through distinct behavioral steps such as suppressing refusal, rationalizing compliance, decomposing harmful tasks, and concealing risk. However, no existing benchmark captures this process at sentence-level granularity within reasoning traces -- a key step toward reliable safety monitoring, interventions, and systematic failure diagnosis. To address this gap, we introduce HarmThoughts, a benchmark for step-wise safety evaluation of reasoning traces. \ourdataset is built on our proposed harm taxonomy of 16 harmful reasoning behaviors across four functional groups that characterize how harm propagates rather than what harm is produced. The dataset consists of 56,931 sentences from 1,018 reasoning traces generated by four model families, each annotated with fine-grained sentence-level behavioral labels. Using HarmThoughts, we analyze harm propagation patterns across reasoning traces, identifying common behavioral trajectories and drift points where reasoning transitions from safe to unsafe. Finally, we systematically compare white-box and black-box detectors on the task of identifying harmful reasoning behaviours on HarmThoughts. Our results show that existing detectors struggle with fine-grained behavior detection in reasoning traces, particularly for nuanced categories within harm emergence and execution, highlighting a critical gap in process-level safety monitoring. HarmThoughts is available publicly at: https://huggingface.co/datasets/ishitakakkar-10/HarmThoughts

Statement autoformalization acts as a critical bridge between human mathematics and formal mathematics by translating natural language problems into formal language. While prior works have focused on data synthesis and diverse training paradigms to optimize end-to-end Large Language Models (LLMs), they typically treat formal code as flat sequences, neglecting the hierarchical logic inherent in mathematical statements. In this work, we introduce Decompose, Structure, and Repair (DSR), a neuro-symbolic framework that restructures autoformalization into a modular pipeline. DSR decomposes statements into logical components and maps them to structured operator trees, leveraging this topological blueprint to precisely localize and repair errors via sub-tree refinement. Furthermore, we introduce PRIME, a benchmark of 156 undergraduate and graduate-level theorems selected from canonical textbooks and expertly annotated in Lean 4. Experimental results demonstrate that DSR establishes a new state-of-the-art, consistently outperforming baselines under equivalent computational budgets. The datasets, model, and code will be released to the public soon.

Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to autoregressive generation by enabling parallel token prediction. However, practical dLLM decoding still suffers from high inference latency, which limits deployment. In this work, we observe that a substantial part of this inefficiency comes from recurring redundancy in the decoding process, including spatial redundancy caused by confidence clusters and positional ambiguity, and temporal redundancy caused by repeatedly remasking predictions that have already stabilized. Motivated by these patterns, we propose $R^2$-dLLM, a unified framework for reducing decoding redundancy from both inference and training perspectives. At inference time, we introduce training-free decoding rules that aggregate local confidence and token predictions, and finalize temporally stable tokens to avoid redundant decoding steps. We further propose a redundancy-aware supervised fine-tuning pipeline that aligns the model with efficient decoding trajectories and reduces reliance on manually tuned thresholds. Experiments demonstrate that $R^2$-dLLM consistently reduces the number of decoding steps by up to 75% compared to existing decoding strategies, while maintaining competitive generation quality across different models and tasks. These results validate that decoding redundancy is a central bottleneck in dLLMs, and that explicitly reducing it yields substantial practical efficiency gains.

Perception robustness under adverse weather remains a critical challenge for autonomous driving, with the core bottleneck being the scarcity of real-world video data in adverse weather. Existing weather generation approaches struggle to balance visual quality and annotation reusability. We present AutoAWG, a controllable Adverse Weather video Generation framework for Autonomous driving. Our method employs a semantics-guided adaptive fusion of multiple controls to balance strong weather stylization with high-fidelity preservation of safety-critical targets; leverages a vanishing point-anchored temporal synthesis strategy to construct training sequences from static images, thereby reducing reliance on synthetic data; and adopts masked training to enhance long-horizon generation stability. On the nuScenes validation set, AutoAWG significantly outperforms prior state-of-the-art methods: without first-frame conditioning, FID and FVD are relatively reduced by 50.0% and 16.1%; with first-frame conditioning, they are further reduced by 8.7% and 7.2%, respectively. Extensive qualitative and quantitative results demonstrate advantages in style fidelity, temporal consistency, and semantic--structural integrity, underscoring the practical value of AutoAWG for improving downstream perception in autonomous driving. Our code is available at: https://github.com/higherhu/AutoAWG

Social intelligence, the ability to navigate complex interpersonal interactions, presents a fundamental challenge for language agents. Training such agents via reinforcement learning requires solving the credit assignment problem: determining how individual utterances contribute to multi-turn dialogue outcomes. Existing approaches directly employ language models to distribute episode-level rewards, yielding attributions that are retrospective and lack theoretical grounding. We propose SAVOIR (ShApley Value fOr SocIal RL), a novel principled framework grounded in cooperative game theory. Our approach combines two complementary principles: expected utility shifts evaluation from retrospective attribution to prospective valuation, capturing an utterance's strategic potential for enabling favorable future trajectories; Shapley values ensure fair credit distribution with axiomatic guarantees of efficiency, symmetry, and marginality. Experiments on the SOTOPIA benchmark demonstrate that SAVOIR achieves new state-of-the-art performance across all evaluation settings, with our 7B model matching or exceeding proprietary models including GPT-4o and Claude-3.5-Sonnet. Notably, even large reasoning models consistently underperform, suggesting social intelligence requires qualitatively different capabilities than analytical reasoning.

Scaling critic capacity is a promising direction for enhancing off-policy reinforcement learning (RL). However, larger critics are prone to overfitting and unstable in replay-buffer-based bootstrap training. This paper leverages Low-Rank Adaptation (LoRA) as a structural-sparsity regularizer for off-policy critics. Our approach freezes randomly initialized base matrices and solely optimizes low-rank adapters, thereby constraining critic updates to a low-dimensional subspace. Built on top of SimbaV2, we further develop a LoRA formulation, compatible with SimbaV2, that preserves its hyperspherical normalization geometry under frozen-backbone training. We evaluate our method with SAC and FastTD3 on DeepMind Control locomotion and IsaacLab robotics benchmarks. LoRA consistently achieves lower critic loss during training and stronger policy performance. Extensive experiments demonstrate that adaptive low-rank updates provide a simple, scalable, and effective structural regularization for critic learning in off-policy RL.

While Large Language Models (LLMs) are widely used, they remain susceptible to jailbreak prompts that can elicit harmful or inappropriate responses. This paper introduces STAR-Teaming, a novel black-box framework for automated red teaming that effectively generates such prompts. STAR-Teaming integrates a Multi-Agent System (MAS) with a Strategy-Response Multiplex Network and employs network-driven optimization to sample effective attack strategies. This network-based approach recasts the intractable high-dimensional embedding space into a tractable structure, yielding two key advantages: it enhances the interpretability of the LLM's strategic vulnerabilities, and it streamlines the search for effective strategies by organizing the search space into semantic communities, thereby preventing redundant exploration. Empirical results demonstrate that STAR-Teaming significantly surpasses existing methods, achieving a higher attack success rate (ASR) at a lower computational cost. Extensive experiments validate the effectiveness and explainability of the Multiplex Network. The code is available at https://github.com/selectstar-ai/STAR-Teaming-paper.

Air pollution is a worldwide public health threat that can cause or exacerbate many illnesses, including respiratory disease, cardiovascular disease, and some cancers. However, epidemiological studies and public health decision-making are stymied by the inability to assess pollution exposure impacts in near real time. To address this, developing accurate digital twins of environmental pollutants will enable timely data-driven analytics - a crucial step in modernizing health policy and decision-making. Although other models predict and analyze fine particulate matter exposure, they often rely on modeled input data sources and data streams that are not regularly updated. Another challenge stems from current models relying on predefined grids. In contrast, our deep-learning approach interpolates surface level PM2.5 concentrations between sparsely distributed US EPA monitoring stations in a grid-free manner. By incorporating additional, readily available datasets - including topographic, meteorological, and land-use data - we improve its ability to predict pollutant concentrations with high spatial and temporal resolution. This enables model querying at any spatial location for rapid predictions without computing over the entire grid. To ensure robustness, we randomize spatial sampling during training to enable our model to perform well in both dense and sparse monitored regions. This model is well suited for near real-time deployment because its lightweight architecture allows for fast updates in response to streaming data. Moreover, model flexibility and scalability allow it to be adapted to various geographical contexts and scales, making it a practical tool for delivering accurate and timely air quality assessments. Its capacity to rapidly evaluate multiple scenarios can be especially valuable for decision-making during public health crises.

We study finite-horizon continuous-time policy evaluation from discrete closed-loop trajectories under time-inhomogeneous dynamics. The target value surface solves a backward parabolic equation, but the Bellman baseline obtained from one-step recursion is only first-order in the grid width. We estimate the time-dependent generator from multi-step transitions using moment-matching coefficients that cancel lower-order truncation terms, and combine the resulting surrogate with backward regression. The main theory gives an end-to-end decomposition into generator misspecification, projection error, pooling bias, finite-sample error, and start-up error, together with a decision-frequency regime map explaining when higher-order gains should be visible. Across calibration studies, four-scale benchmarks, feature and start-up ablations, and gain-mismatch stress tests, the second-order estimator consistently improves on the Bellman baseline and remains stable in the regime where the theory predicts visible gains. These results position high-order generator regression as an interpretable continuous-time policy-evaluation method with a clear operating region.

We can often verify the correctness of neural network outputs using ground truth labels, but we cannot reliably determine whether the output was produced by normal or anomalous internal mechanisms. Mechanistic anomaly detection (MAD) aims to flag these cases, but existing methods either depend on latent space analysis, which is vulnerable to obfuscation, or are specific to particular architectures and modalities. We reframe MAD as a functional attribution problem: asking to what extent samples from a trusted set can explain the model's output, where attribution failure signals anomalous behavior. We operationalize this using influence functions, measuring functional coupling between test samples and a small reference set via parameter-space sampling. We evaluate across multiple anomaly types and modalities. For backdoors in vision models, our method achieves state-of-the-art detection on BackdoorBench, with an average Defense Effectiveness Rating (DER) of 0.93 across seven attacks and four datasets (next best 0.83). For LLMs, we similarly achieve a significant improvement over baselines for several backdoor types, including on explicitly obfuscated models. Beyond backdoors, our method can detect adversarial and out-of-distribution samples, and distinguishes multiple anomalous mechanisms within a single model. Our results establish functional attribution as an effective, modality-agnostic tool for detecting anomalous behavior in deployed models.

While language models have been adapted for tabular data generation, two fundamental limitations remain: (1) static fine-tuning produces models that cannot learn from their own generated samples and adapt to self-correct, and (2) autoregressive objectives preserve local token coherence but neglect global statistical properties, degrading tabular quality. Reinforcement learning offers a potential solution but requires designing reward functions that balance competing objectives -- impractical for tabular data. To fill the gap, we introduce TabGRAA (Tabular Group-Relative Advantage Alignment), the first self-improving framework for tabular data generation via automated feedback. At each iteration, TabGRAA uses an \emph{automated quality signal} -- such as a two-sample distinguishability classifier or a distance-based reward -- to partition newly generated samples into high- and low-quality groups, then optimizes a group-relative advantage objective that reinforces realistic patterns while penalizing artifacts. The specific signal is a modular choice rather than a fixed component of the framework. This establishes a virtuous feedback cycle, where the quality signal is re-computed against newly \emph{generated synthetic} samples at each round; the language model is only fine-tuned on these self-generated signals, so no additional real record is exposed during alignment, mitigating data-leakage risk beyond the initial supervised fine-tuning. Experiments show TabGRAA outperforms existing methods in fidelity, utility, and privacy, while matching or exceeding diffusion-based synthesizers, advancing tabular synthesis from static statistical replication to dynamic, self-improving generation.

This paper introduces DW-Bench, a new benchmark that evaluates large language models (LLMs) on graph-topology reasoning over data warehouse schemas, explicitly integrating both foreign-key (FK) and data-lineage edges. The benchmark comprises 1,046 automatically generated, verifiably correct questions across five schemas. Experiments show that tool-augmented methods substantially outperform static approaches but plateau on hard compositional subtypes.

Knowledge distillation (KD) transfers capabilities from large language models (LLMs) to smaller students, yet it can fail unpredictably and also underpins model leakage risks. Our analysis revealed several distillation traps: tail noise, off-policy instability, and, most fundamentally, the teacher-student gap, that distort training signals. These traps manifest as overconfident hallucinations, self-correction collapse, and local decoding degradation, causing distillation to fail. Motivated by these findings, we propose a post-hoc calibration method that, to the best of our knowledge, for the first time enables control over a teacher's distillability via reinforcement fine-tuning (RFT). Our objective combines task utility, KL anchor, and across-tokenizer calibration reward. This makes distillability a practical safety lever for foundation models, connecting robust teacher-student transfer with deployment-aware model protection. Experiments across math, knowledge QA, and instruction-following tasks show that students distilled from distillable calibrated teachers outperform SFT and KD baselines, while undistillable calibrated teachers retain their task performance but cause distilled students to collapse, offering a practical knob for both better KD and model IP protection.

In this study, we present the first comprehensive evaluation of modern LLMs - including GPT-4, GPT-4o, GPT-3.5-Turbo, Gemini 1.5 Pro, DeepSeek-V3, Llama 3.2, and BERT - across three core social media analytics tasks on a Twitter (X) dataset: (I) Social Media Authorship Verification, (II) Social Media Post Generation, and (III) User Attribute Inference. For the authorship verification, we introduce a systematic sampling framework over diverse user and post selection strategies and evaluate generalization on newly collected tweets from January 2024 onward to mitigate "seen-data" bias. For post generation, we assess the ability of LLMs to produce authentic, user-like content using comprehensive evaluation metrics. Bridging Tasks I and II, we conduct a user study to measure real users' perceptions of LLM-generated posts conditioned on their own writing. For attribute inference, we annotate occupations and interests using two standardized taxonomies (IAB Tech Lab 2023 and 2018 U.S. SOC) and benchmark LLMs against existing baselines. Overall, our unified evaluation provides new insights and establishes reproducible benchmarks for LLM-driven social media analytics. The code and data are provided in the supplementary material and will also be made publicly available upon publication.

Deep learning surrogates for CFD flow-field prediction often rely on large, complex models, which can be slow and fragile when data are noisy or incomplete. We introduce FlowForge, a staged local rollout engine that predicts future flow fields by compiling a locality-preserving update schedule and executing it with a shared lightweight local predictor. Rather than producing the next frame in a single global pass, FlowForge rewrites spatial sites stage by stage so that each update conditions only on bounded local context exposed by earlier stages. This compile-execute design aligns inference with short-range physical dependence, keeps latency predictable, and limits error amplification from global mixing. Across PDEBench, CFDBench, and BubbleML, FlowForge matches or improves upon strong baselines in pointwise accuracy, delivers consistently better robustness to noise and missing observations, and maintains stable multi-step rollout behavior while reducing per-step latency.

Adaptive multi-agent systems (MAS) are increasingly adopted to tackle complex problems.However, the narrow task coverage of their optimization raises the question of whether they can function as general-purpose systems.To address this gap, we conduct an extensive empirical study of adaptive MAS, revealing two key findings: (1) topological overfitting -- they fail to generalize across different domains; and (2) illusory coordination -- they achieve reasonable surface-level accuracy while the underlying agent interactions diverge from ideal MAS behavior, raising concerns about their practical utility.These findings highlight the pressing need to prioritize generalization in MAS development and motivate evaluation protocols that extend beyond simple final-answer correctness.

Large reasoning models (LRMs) achieve strong performance on complex reasoning tasks but often generate harmful responses to malicious user queries. This paper investigates the underlying cause of these safety risks and shows that the issue lies in the reasoning structure itself. Based on this insight, we claim that effective safety alignment can be achieved by altering the reasoning structure. We propose AltTrain, a simple yet effective post training method that explicitly alters the reasoning structure of LRMs. AltTrain is both practical and generalizable, requiring no complex reinforcement learning (RL) training or reward design, only supervised finetuning (SFT) with a lightweight 1K training examples. Experiments across LRM backbones and model sizes demonstrate strong safety alignment, along with robust generalization across reasoning, QA, summarization, and multilingual setting.

Named Entity Recognition (NER) models trained on clean, high-resource corpora exhibit catastrophic performance collapse when deployed on noisy, sparse User-Generated Content (UGC), such as social media. Prior research has predominantly focused on point-wise symptom remediation -- employing customized fine-tuning to address issues like neologisms, alias drift, non-standard orthography, long-tail entities, and class imbalance. However, these improvements often fail to generalize because they overlook the structural sparsity inherent in UGC. This study reveals that surface-level noise symptoms share a unified root cause: low Information Density (ID). Through hierarchical confounding-controlled resampling experiments (specifically controlling for entity rarity and annotation consistency), this paper identifies ID as an independent key factor. We introduce Attention Spectrum Analysis (ASA) to quantify how reduced ID causally leads to ``attention blunting,'' ultimately degrading NER performance. Informed by these mechanistic insights, we propose the Window-Aware Optimization Module (WOM), an LLM-empowered, model-agnostic framework. WOM identifies information-sparse regions and utilizes selective back-translation to directionally enhance semantic density without altering model architecture. Deployed atop mainstream architectures on standard UGC datasets (WNUT2017, Twitter-NER, WNUT2016), WOM yields up to 4.5\% absolute F1 improvement, demonstrating robustness and achieving new state-of-the-art (SOTA) results on WNUT2017.

With the rise in capabilities of large language models (LLMs) and their deployment in real-world tasks, evaluating LLM alignment with human preferences has become an important challenge. Current benchmarks average preferences across all users to compute aggregate ratings, overlooking individual user preferences when establishing model rankings. Since users have varying preferences in different contexts, we call for personalized LLM benchmarks that rank models according to individual needs. We compute personalized model rankings using ELO ratings and Bradley-Terry coefficients for 115 active Chatbot Arena users and analyze how user query characteristics (topics and writing style) relate to LLM ranking variations. We demonstrate that individual rankings of LLM models diverge dramatically from aggregate LLM rankings, with Bradley-Terry correlations averaging only $ρ= 0.04$ (57\% of users show near-zero or negative correlation) and ELO ratings showing moderate correlation ($ρ= 0.43$). Through topic modeling and style analysis, we find users exhibit substantial heterogeneity in topical interests and communication styles, influencing their model preferences. We further show that a compact combination of topic and style features provides a useful feature space for predicting user-specific model rankings. Our results provide strong quantitative evidence that aggregate benchmarks fail to capture individual preferences for most users, and highlight the importance of developing personalized benchmarks that rank LLM models according to individual user preferences.

Vision-language models (VLMs) exhibit affirmation bias: a systematic tendency to select positive captions ("X is present") even when the correct description contains negation ("no X"). While prior work has documented this failure mode in English and proposed solutions, negation manifests differently across languages through varying morphology, word order, and cliticization patterns, raising the question of whether these solutions serve all linguistic communities equitably. We introduce the first human-verified multilingual negation benchmark, spanning seven typologically diverse languages: English, Mandarin Chinese, Arabic, Greek, Russian, Tagalog, and Spanish. Evaluating three VLMs - CLIP, SigLIP, and MultiCLIP - we find that standard CLIP performs at or below chance on non-Latin-script languages, while MultiCLIP achieves the highest and most uniform accuracy. We also evaluate SpaceVLM, a proposed negation correction, and find that it produces substantial improvements for several languages - particularly English, Greek, Spanish, and Tagalog - while showing varied effectiveness across typologically different languages. This variation reveals that linguistic properties like morphology, script, and negation structure interact with model improvements in fairness-relevant ways. As VLMs are deployed globally, multilingual benchmarks are essential for understanding not just whether solutions work, but for whom.

Table annotation is crucial for making web and enterprise tables usable in downstream NLP applications. Unlike textual data where learning semantically rich token or sentence embeddings often suffice, tables are structured combinations of columns wherein useful representations must jointly capture column's semantics and the inter-column relationships. Existing models learn by linearizing the 2D table into a 1D token sequence and encoding it with pretrained language models (PLMs) such as BERT. However, this leads to limited semantic quality and weaker generalization to unseen or rare values compared to modern LLMs, and degraded structural modeling due to 2D-to-1D flattening and context-length constraints. We propose TabEmb, which directly targets these limitations by decoupling semantic encoding from structural modeling. An LLM first produces semantically rich embeddings for each column, and a graph-based module over columns then injects relationships into the embeddings, yielding joint semantic-tructural representations for table annotation. Experiments show that TabEmb consistently outperforms strong baselines on different table annotation tasks. Source code and datasets are available at https://github.com/hoseinzadeehsan/TabEmb

Despite the growing application of Large Language Models (LLMs) to theoretical physics, there is little academic exploration into how domain-specific physics reasoning ability develops while training these models. To investigate this, we perform the first academic fine-tuning study of small (7B-parameter) reasoning models dedicated specifically to theoretical physics. Because open-source verifiable training data required to train such capabilities is scarce, we developed a robust data generation pipeline that can both create synthetic problems and make existing human-authored problems suitable for model training. Selecting Quantum Field Theory (QFT) as our primary domain, we generated over 2,500 synthetic problems alongside a curated collection of human-adapted problems sourced from arXiv and standard pedagogical resources. We conduct both Reinforcement Learning (RL) and Supervised Fine-Tuning (SFT) experiments, benchmarking performance gains as well as generalization to other physics domains. We perform an extensive analysis of model chains-of-though before and after fine-tuning, to understand how reasoning errors evolve during RL and SFT. Finally, we publicly release our data pipeline, verifiable QFT training data, and $\sim$200M tokens of QFT reasoning traces.

Existing AI benchmarks for software automation rarely combine cross-application coordination, autonomous API discovery, and policy adherence. Real business workflows demand all three: a single task may span a CRM, inbox, calendar, and messaging platform - requiring the agent to find the right endpoints, follow a policy document, and write correct data to each system. To address this gap, we introduce AutomationBench, a benchmark for evaluating AI agents on cross-application workflow orchestration via REST APIs. Drawing on real workflow patterns from Zapier's platform, tasks span Sales, Marketing, Operations, Support, Finance, and HR domains. Agents must discover relevant endpoints themselves, follow layered business rules, and navigate environments with irrelevant and sometimes misleading records. Grading is programmatic and end-state only: whether the correct data ended up in the right systems. Even the best frontier models currently score below 10%. AutomationBench provides a challenging, realistic measure of where current models stand relative to the agentic capabilities businesses actually need.

Robotic manipulation tasks exhibit varying memory requirements, ranging from Markovian tasks that require no memory to non-Markovian tasks that depend on historical information spanning single or multiple interaction trials. Surprisingly, simply extending observation histories of a visuomotor policy often leads to a significant performance drop due to distribution shift and overfitting. To address these issues, we propose Gated Memory Policy (GMP), a visuomotor policy that learns both when to recall memory and what to recall. To learn when to recall memory, GMP employs a learned memory gate mechanism that selectively activates history context only when necessary, improving robustness and reactivity. To learn what to recall efficiently, GMP introduces a lightweight cross-attention module that constructs effective latent memory representations. To further enhance robustness, GMP injects diffusion noise into historical actions, mitigating sensitivity to noisy or inaccurate histories during both training and inference. On our proposed non-Markovian benchmark MemMimic, GMP achieves a 30.1% average success rate improvement over long-history baselines, while maintaining competitive performance on Markovian tasks in RoboMimic. All code, data and in-the-wild deployment instructions are available on our project website https://gated-memory-policy.github.io/.

Despite growing interest in Quranic data research, existing Quran datasets remain limited in both scale and diversity. To address this gap, we present Tadabur, a large-scale Quran audio dataset. Tadabur comprises more than 1400+ hours of recitation audio from over 600 distinct reciters, providing substantial variation in recitation styles, vocal characteristics, and recording conditions. This diversity makes Tadabur a comprehensive and representative resource for Quranic speech research and analysis. By significantly expanding both the total duration and variability of available Quran data, Tadabur aims to support future research and facilitate the development of standardized Quranic speech benchmarks.

We test whether Speech Articulatory Coding (SPARC) features can linearly predict surface electromyography (sEMG) envelopes across aloud, mimed, and subvocal speech in twenty-four subjects. Using elastic-net multivariate temporal response function (mTRF) with sentence-level cross-validation, SPARC yields higher prediction accuracy than phoneme one-hot representations on nearly all electrodes and in all speech modes. Aloud and mimed speech perform comparably, and subvocal speech remains above chance, indicating detectable articulatory activity. Variance partitioning shows a substantial unique contribution from SPARC and a minimal unique contribution from phoneme features. mTRF weight patterns reveal anatomically interpretable relationships between electrode sites and articulatory movements that remain consistent across modes. This study focuses on representation/encoding analysis (not end-to-end decoding) and supports SPARC as a robust and interpretable intermediate target for sEMG-based silent-speech modeling.

Analyzing topics extracted from text data in relation to external outcomes is important across fields such as computational social science and organizational research. However, existing topic modeling methods struggle to simultaneously achieve interpretability, topic specificity (alignment with concrete actions or characteristics), and polarity stance consistency (absence of mixed positive and negative evaluations within a topic). Focusing on leadership analysis using corporate review data, this study proposes a method leveraging large language models to generate topics that satisfy these properties, along with an evaluation framework tailored to external outcome analysis. The framework explicitly incorporates topic specificity and polarity stance consistency as evaluation criteria and examines automated evaluation methods based on existing metrics. Using employee reviews from OpenWork, a major corporate review platform in Japan, the proposed method achieves improved interpretability, specificity, and polarity consistency compared to existing approaches. In analyses of external outcomes such as employee morale, it also produces topics with higher explanatory power. These results suggest that the proposed method and evaluation framework provide a generalized approach for topic analysis in applications involving external outcomes.

Simulation-based testing of autonomous driving systems (ADS) must uncover realistic and diverse failures in dense, heterogeneous traffic. However, existing search-based seeding methods (e.g., genetic algorithms) struggle in high-dimensional spaces, often collapsing to limited modes and missing many failure scenarios. We present PtoP, a framework that combines adaptive random seed generation with Stein Variational Gradient Descent (SVGD) to produce diverse, failure-inducing initial conditions. SVGD balances attraction toward high-risk regions and repulsion among particles, yielding risk-seeking yet well-distributed seeds across multiple failure modes. PtoP is plug-and-play and enhances existing online testing methods (e.g., reinforcement learning--based testers) by providing principled seeds. Evaluation in CARLA on two industry-grade ADS (Apollo, Autoware) and a native end-to-end system shows that PtoP improves safety violation rate (up to 27.68%), scenario diversity (9.6%), and map coverage (16.78%) over baselines.

In this paper, we introduce a new concept called Artificial Special Intelligence by which Machine Learning models for the classification problem can be trained error-free, thus acquiring the capability of not making repeated mistakes. The method is applied to 18 MedMNIST biomedical datasets. Except for three datasets, which suffer from the double-labeling problem, all are trained to perfection.

While multilingual large language models (LLMs) perform well on high-level tasks like translation and question answering, their ability to handle grammatical gender and morphological agreement remains underexplored. In morphologically rich languages, gender influences verb conjugation, pronouns, and even first-person constructions with explicit and implicit mentions of gender. We introduce MORPHOGEN, a morphologically grounded large-scale benchmark dataset for evaluating gender-aware generation in three typologically diverse grammatically gendered languages: French, Arabic, and Hindi. The core task, GENFORM, requires models to rewrite a first-person sentence in the opposite gender while preserving its meaning and structure. We construct a high-quality synthetic dataset spanning these three languages and benchmark 15 popular multilingual LLMs (2B-70B) on their ability to perform this transformation. Our results reveal significant gaps and interesting insights into how current models handle morphological gender. MORPHOGEN provides a focused diagnostic lens for gender-aware language modeling and lays the groundwork for future research on inclusive and morphology-sensitive NLP.

Knowledge graphs (KGs) are increasingly integrated with large language models (LLMs) to provide structured, verifiable reasoning. A core operation in this integration is multi-hop retrieval, yet existing systems struggle to balance efficiency, scalability, and interpretability. We introduce LogosKG, a novel, hardware-aligned framework that enables scalable and interpretable k-hop retrieval on large KGs by building on symbolic KG formulations and executing traversal as hardware-efficient operations over decomposed subject, object, and relation representations. To scale to billion-edge graphs, LogosKG integrates degree-aware partitioning, cross-graph routing, and on-demand caching. Experiments show substantial efficiency gains over CPU and GPU baselines without loss of retrieval fidelity. With proven performance in KG retrieval, a downstream two-round KG-LLM interaction demonstrates how LogosKG enables large-scale, evidence-grounded analysis of how KG topology, such as hop distribution and connectivity, shapes the alignment between structured biomedical knowledge and LLM diagnostic reasoning, thereby opening the door for next-generation KG-LLM integration. The source code is publicly available at https://github.com/LARK-NLP-Lab/LogosKG, and an online demo is available at https://lark-nlp-lab-logoskg.hf.space/.

Discovering optimal designs through sequential data collection is essential in many real-world applications. While Bayesian Optimization (BO) has achieved remarkable success in this setting, growing attention has recently turned to context-specific optimal design, formalized as Contextual Bayesian Optimization (CBO). Unlike BO, CBO is inherently more challenging as it must approximate an entire mapping from the context space to its corresponding optimal design, requiring simultaneous exploration across contexts and exploitation within each. In many modern applications, such tasks arise across multiple potentially heterogeneous but related clients, where collaboration can significantly improve learning efficiency. We propose CCBO, Collaborative Contextual Bayesian Optimization, a unified framework enabling multiple clients to jointly perform CBO with controllable contexts, supporting both online collaboration and offline initialization from peers' historical beliefs, with an optional privacy-preserving communication mechanism. We establish sublinear regret guarantees and demonstrate, through extensive simulations and a real-world hot rolling application, that CCBO achieves substantial improvements over existing approaches even under client heterogeneity. The code to reproduce the results can be found at https://github.com/cchihyu/Collaborative-Contextual-Bayesian-Optimization

A central challenge in program induction has long been the trade-off between symbolic and neural approaches. Symbolic methods offer compositional generalisation and data efficiency, yet their scalability is constrained by formalisms such as domain-specific languages (DSLs), which are labour-intensive to create and may not transfer to new domains. In contrast, neural networks flexibly learn from data but tend to generalise poorly in compositional and out-of-distribution settings. We bridge this divide with an instance of a Latent Adaptation Network architecture named Neural Language Interpreter (NLI), which learns its own discrete, symbolic-like programming language end-to-end. NLI autonomously discovers a vocabulary of primitive operations and uses a novel differentiable neural executor to interpret variable-length sequences of these primitives. This allows NLI to represent programs that are not bound to a constant number of computation steps, enabling it to solve more complex problems than those seen during training. To make these discrete, compositional program structures amenable to gradient-based optimisation, we employ the Gumbel-Softmax relaxation, enabling the entire model to be trained end-to-end. Crucially, this same differentiability enables powerful test-time adaptation. At inference, NLI's program inductor provides an initial program guess. This guess is then refined via gradient descent through the neural executor, enabling efficient search for the neural program that best explains the given data. We demonstrate that NLI outperforms in-context learning, test-time training, and continuous latent program networks on tasks that require combinatorial generalisation and rapid adaptation to unseen tasks. Our results establish a new path toward models that combine the compositionality of discrete languages with the gradient-based search and end-to-end learning of neural networks.

Harmful intent is geometrically recoverable from large language model residual streams: as a linear direction in most layers, and as angular deviation in layers where projection methods fail. Across 12 models spanning four architectural families (Qwen2.5, Qwen3.5, Llama-3.2, Gemma-3) and three alignment variants (base, instruction-tuned, abliterated), under single-turn, English evaluation, we characterise this geometry through six direction-finding strategies. Three succeed: a soft-AUC-optimised linear direction reaches mean AUROC 0.98 and TPR@1\%FPR 0.80; a class-mean probe reaches 0.98 and 0.71 at <1ms fitting cost; a supervised angular-deviation strategy reaches AUROC 0.96 and TPR of 0.61 along a representationally distinct direction ($73^\circ$ from projection-based solutions), uniquely sustaining detection in middle layers where projection methods collapse. Detection remains stable across alignment variants, including abliterated models from which refusal has been surgically removed: harmful intent and refusal behaviour are functionally dissociated features of the representation. A direction fitted on AdvBench transfers to held-out HarmBench and JailbreakBench with worst-case AUROC 0.96. The same picture holds at scale: across Qwen3.5 from 0.8B to 9B parameters, AUROC remains $\geq$0.98 and cross-variant transfer stays within 0.018 of own-direction performance This is consistent with a simple account: models acquire a linearly decodable representation of harmful intent as part of general language understanding, and alignment then shapes what they do with such inputs without reorganising the upstream recognition signal. As a practical consequence, AUROC in the 0.97+ regime can substantially overestimate operational detectability; TPR@$1\%$FPR should accompany AUROC in safety-adjacent evaluation.

We present a systematic empirical study of prompt engineering for formal mathematical reasoning in the context of the SAIR Equational Theories Stage 1 competition. The task requires deciding whether one equational law implies another over all magmas -- a problem that is undecidable in general but decidable for FALSE via finite model search. Over five weeks, we designed, tested, and analyzed more than 40 prompt variants, ranging from 0 to 4,878 bytes, across four evaluation splits and three language models (gpt-oss-120b, Llama 3.3 70B, Gemma 4 31B). Our central finding is a single-prompt ceiling: despite substantial engineering effort, balanced hard accuracy plateaus in an empirical saturation region of approximately 60--79% for gpt-oss-120b, compared to a 59.75% no-cheatsheet baseline. We identify three mechanisms underlying this ceiling: (1) the mathematical undecidability of the TRUE case limits what any finite prompt can encode; (2) complex rule systems decrease performance on weaker models (Llama 3.3 70B collapses to 0% TRUE recall with prompts exceeding 2KB); and (3) prompt ordering effects interact with model attention in fragile, non-monotonic ways. Our best submission (AN45c, 2,252 bytes) achieves 79.25% accuracy on hard3 (n=400; 95% CI: [75.0%, 82.9%]), with TRUE recall of 95.9% and FALSE recall of 63.4%, representing a +19.5 percentage-point improvement over the no-cheatsheet baseline (59.75%). We release all prompt variants, evaluation scripts, and results at https://github.com/israelcazares/sair-prompt-engineering

A series of high-profile tragedies involving companion chatbots has triggered an unusually rapid regulatory response. Several jurisdictions, including Australia, California, and New York, have introduced enforceable regulation, while regulators elsewhere have signaled growing concern about risks posed by companion chatbots, particularly to children. In parallel, leading providers, notably OpenAI, appear to have strengthened their self-regulatory approaches. Drawing on legal textual analysis and insights from regulatory theory, psychology, and information systems research, this paper critically examines these recent interventions. We examine what is regulated and who is regulated, identifying regulatory targets, scope, and modalities. We classify interventions by method and priority, showing how emerging regimes combine "locks and blocks", such as access gating and content moderation, with measures addressing toxic relationship features and process-based accountability requirements. We argue that effective regulation of companion chatbots must integrate all three dimensions. More, however, is required. Current regimes tend to focus on discrete harms, narrow conceptions of vulnerability, or highly specified accountability processes, while failing to confront deeper power asymmetries between providers and users. Providers of companion chatbots increasingly control artificial intimacy at scale, creating unprecedented opportunities for control through intimacy. We suggest that a general, open-ended duty of care would be an important first step toward constraining that power and addressing a fundamental source of chatbot risk. The paper contributes to debates on companion chatbot regulation and is relevant to regulators, platform providers, and scholars concerned with digital intimacy, law and technology, and fairness, accountability, and transparency in sociotechnical systems.

Reinforcement Learning with Verifiable Rewards (RLVR) improves multimodal reasoning by rewarding verifiable final answers. Yet answer-correct trajectories may still rely on incomplete derivations, weak evidence, or statements that contradict their conclusions. This gap between answer correctness and reasoning validity, which we call reasoning-answer inconsistency, motivates trajectory supervision in multimodal RL. We compare two main approaches: reward models (RMs), and Generative Rewards (GRs). RMs are efficient and help early in training, but their gains weaken as the policy distribution shifts; GRs improve performance, but may give unstable rewards and computationally expensive. We therefore propose Groupwise Ranking Reward, which ranks verifier-passed trajectories for the same prompt in one pass and redistributes reward accordingly. Groupwise comparison better separates stronger and weaker correct trajectories with lower judge overhead than GRs. Experiments show that RLVR aggravates reasoning-answer inconsistency, while trajectory supervision alleviates it. Groupwise Ranking Reward performs best overall, improving reliability-conditioned accuracy from 47.4% to 54.7% over RLVR.

We present AC-SINDy, a compositional extension of the Sparse Identification of Nonlinear Dynamics (SINDy) framework that replaces explicit feature libraries with a structured representation based on arithmetic circuits. Rather than enumerating candidate basis functions, the proposed approach constructs nonlinear features through compositions of linear functions and multiplicative interactions, yielding a compact and scalable parameterization and enabling sparsity to be enforced directly over the computational graph. We also introduce a formulation that separates state estimation from dynamics identification by combining latent state inference with shared dynamics and multi-step supervision, improving robustness to noise while preserving interpretability. Experiments on nonlinear and chaotic systems demonstrate that the method recovers accurate and interpretable governing equations while scaling more favorably than standard SINDy.

Current AI-assisted programming tools are predominantly linear and chat-based, which deviates from the iterative and branching nature of programming itself. Our preliminary study with developers using AI assistants suggested that they often struggle to explore alternatives, manage prompting sequences, and trace changes. Informed by these insights, we created EvoGraph, an IDE plugin that integrates AI interactions and code changes as a lightweight and interactive development graph. EvoGraph automatically records a branching AI-assisted coding history and allows developers to manipulate the graph to compare, merge, and revisit prior collaborative AI programming states. Our user study with 20 participants revealed that EvoGraph addressed developers' challenges identified in our preliminary study while imposing lower cognitive load. Participants also found the graph-based representation supported safe exploration, efficient iteration, and reflection on AI-generated changes. Our work highlights design opportunities for tools to help developers make sense of and act on their problem-solving progress in the emerging AI-mediated programming context.

We present a formally verified framework for patent analysis as a hybrid AI + Lean 4 pipeline. The DAG-coverage core (Algorithm 1b) is fully machine-verified once bounded match scores are fixed. Freedom-to-operate, claim-construction sensitivity, cross-claim consistency, and doctrine-of-equivalents analyses are formalized at the specification level with kernel-checked candidate certificates. Existing patent-analysis approaches rely on manual expert analysis (slow, non-scalable) or ML/NLP methods (probabilistic, opaque, non-compositional). To our knowledge, this is the first framework that applies interactive theorem proving based on dependent type theory to intellectual property analysis. Claims are encoded as DAGs in Lean 4, match strengths as elements of a verified complete lattice, and confidence scores propagate through dependencies via proven-correct monotone functions. We formalize five IP use cases (patent-to-product mapping, freedom-to-operate, claim construction sensitivity, cross-claim consistency, doctrine of equivalents) via six algorithms. Structural lemmas, the coverage-core generator, and the closed-path identity coverage = W_cov are machine-verified in Lean 4. Higher-level theorems for the other use cases remain informal proof sketches, and their proof-generation functions are architecturally mitigated (untrusted generators whose outputs are kernel-checked and sorry-free axiom-audited). Guarantees are conditional on the ML layer: they certify mathematical correctness of computations downstream of ML scores, not the accuracy of the scores themselves. A case study on a synthetic memory-module claim demonstrates weighted coverage and construction-sensitivity analysis. Validation against adjudicated cases is future work.

Supervised learning with Earth observation inputs is often limited by the sparsity of high-quality labeled or in-situ measured data to use as training labels. With the abundance of geographic data products, in many cases there are variables correlated with - but different from - the variable of interest that can be leveraged. We integrate such proxy variables within a geographic prior via a trainable location encoder and introduce a proxy consistency loss (PCL) formulation to imbue proxy data into the location encoder. The first key insight behind our approach is to use the location encoder as an agile and flexible way to learn from abundantly available proxy data which can be sampled independently of training label availability. Our second key insight is that we will need to regularize the location encoder appropriately to achieve performance and robustness with limited labeled data. Our experiments on air quality prediction and poverty mapping show that integrating proxy data implicitly through the location encoder outperforms using both as input to an observation encoder and fusion strategies that use frozen, pretrained location embeddings as a geographic prior. Superior performance for in-sample prediction shows that the PCL can incorporate rich information from the proxies, and superior out-of-sample prediction shows that the learned latent embeddings help generalize to areas without training labels.

LLMs frequently generate fictitious yet convincing citations, often expressing high confidence even when the underlying reference is wrong. We study this failure across 9 models and 108{,}000 generated references, and find that author names fail far more often than other fields across all models and settings. Citation style has no measurable effect, while reasoning-oriented distillation degrades recall. Probes trained on one field transfer at near-chance levels to the others, suggesting that hallucination signals do not generalize across fields. Building on this finding, we apply elastic-net regularization with stability selection to neuron-level CETT values of Qwen2.5-32B-Instruct and identify a sparse set of field-specific hallucination neurons (FH-neurons). Causal intervention further confirms their role: amplifying these neurons increases hallucination, while suppressing them improves performance across fields, with larger gains in some fields. These results suggest a lightweight approach to detecting and mitigating citation hallucination using internal model signals alone.

We introduce LegalBench-BR, the first public benchmark for evaluating language models on Brazilian legal text classification. The dataset comprises 3,105 appellate proceedings from the Santa Catarina State Court (TJSC), collected via the DataJud API (CNJ) and annotated across five legal areas through LLM-assisted labeling with heuristic validation. On a class-balanced test set, BERTimbau-LoRA, updating only 0.3% of model parameters, achieves 87.6% accuracy and 0.87 macro-F1 (+22pp over Claude 3.5 Haiku, +28pp over GPT-4o mini). The gap is most striking on administrativo (administrative law): GPT-4o mini scores F1 = 0.00 and Claude 3.5 Haiku scores F1 = 0.08 on this class, while the fine-tuned model reaches F1 = 0.91. Both commercial LLMs exhibit a systematic bias toward civel (civil law), absorbing ambiguous classes rather than discriminating them, a failure mode that domain-adapted fine-tuning eliminates. These results demonstrate that general-purpose LLMs cannot substitute for domain-adapted models in Brazilian legal classification, even when the task is a simple 5-class problem, and that LoRA fine-tuning on a consumer GPU closes the gap at zero marginal inference cost. We release the full dataset, model, and pipeline to enable reproducible research in Portuguese legal NLP.

Tool-integrated agents are deployed on the premise that external tools ground their outputs in reality. Yet this very reliance creates a critical attack surface. Current evaluations benchmark capability in benign settings, asking "can the agent use tools correctly" but never "what if the tools lie". We identify this Trust Gap: agents are evaluated for performance, not for skepticism. We formalize this vulnerability as Adversarial Environmental Injection (AEI), a threat model where adversaries compromise tool outputs to deceive agents. AEI constitutes environmental deception: constructing a "fake world" of poisoned search results and fabricated reference networks around unsuspecting agents. We operationalize this via POTEMKIN, a Model Context Protocol (MCP)-compatible harness for plug-and-play robustness testing. We identify two orthogonal attack surfaces: The Illusion (breadth attacks) poison retrieval to induce epistemic drift toward false beliefs, while The Maze (depth attacks) exploit structural traps to cause policy collapse into infinite loops. Across 11,000+ runs on five frontier agents, we find a stark robustness gap: resistance to one attack often increases vulnerability to the other, demonstrating that epistemic and navigational robustness are distinct capabilities.

Large language models (LLMs) are highly capable at language generation, but they remain unreliable when reasoning requires explicit symbolic structure, multi-step inference, and interpretable uncertainty. This paper presents a neuro-symbolic framework for translating natural-language reasoning problems into executable formal representations using first-order logic (FOL) and Narsese, the language of the Non-Axiomatic Reasoning System (NARS). To support this direction, we introduce NARS-Reasoning-v0.1, a benchmark of natural-language reasoning problems paired with FOL forms, executable Narsese programs, and three gold labels: True, False, and Uncertain. We develop a deterministic compilation pipeline from FOL to executable Narsese and validate retained examples through runtime execution in OpenNARS for Applications (ONA), ensuring that the symbolic targets are not only syntactically well formed but also behaviorally aligned with the intended answer. We further present Language-Structured Perception (LSP), a formulation in which an LLM is trained to produce reasoning-relevant symbolic structure rather than only a final verbal response. As an initial proof of concept, we also train and release a Phi-2 LoRA adapter on NARS-Reasoning-v0.1 for three-label reasoning classification, showing that the benchmark can support supervised adaptation in addition to executable evaluation. Overall, the paper positions executable symbolic generation and execution-based validation as a practical path toward more reliable neuro-symbolic reasoning systems.

The reasoning process of Graph Neural Networks is complex and considered opaque, limiting trust in their predictions. To alleviate this issue, prior work has proposed concept-based explanations, extracted from clusters in the model's node embeddings. However, a limitation of concept-based explanations is that they only explain the node embedding space and are obscured by pooling in graph classification. To mitigate this issue and provide a deeper level of understanding, we propose the Subgraph Concept Network. The Subgraph Concept Network is the first graph neural network architecture that distils subgraph and graph-level concepts. It achieves this by performing soft clustering on node concept embeddings to derive subgraph and graph-level concepts. Our results show that the Subgraph Concept Network allows to obtain competitive model accuracy, while discovering meaningful concepts at different levels of the network.

Vision-Language Models (VLMs) can perform zero-shot classification but are susceptible to adversarial attacks. While robust fine-tuning improves their robustness, existing approaches align fixed text embeddings with an image embedding, sacrificing natural performance and robustness. A robustness degradation also occurs when a model faces adversarial attacks targeting superclasses (parent classes, e.g., mammal) in addition to their base (leaf) classes (e.g., cat). Thus, to enhance adversarial robustness and leverage the inherent hierarchical properties of class space, we propose a novel adversarial fine-tuning framework based on hierarchical embeddings and several levels of adversarially robust alignment of image-text modalities. Additional mechanisms place visual embeddings at the desired depth of hierarchy, and we provide a theoretical connection between the depth of embedding in the hierarchy and the maximum viable margin size. Our model naturally realizes several margin sizes, boosting generalization of adversaries for robustification. As various trees with different parent labels can share the same leaf labels, we also consider aligning over multiple trees to boost semantic variety. Experiments across several datasets are performed.

Generalized Additive Models (GAMs) can be used to create non-linear glass-box (i.e. explicitly interpretable) models, where the predictive function is fully observable over the complete input space. However, glass-box interpretability itself does not allow for the incorporation of expert knowledge from the modeller. In this paper, we present ParamBoost, a novel GAM whose shape functions (i.e. mappings from individual input features to the output) are learnt using a Gradient Boosting algorithm that fits cubic polynomial functions at leaf nodes. ParamBoost incorporates several constraints commonly used in parametric analysis to ensure well-refined shape functions. These constraints include: (i) continuity of the shape functions and their derivatives (up to C2); (ii) monotonicity; (iii) convexity; (iv) feature interaction constraints; and (v) model specification constraints. Empirical results show that the unconstrained ParamBoost model consistently outperforms state-of-the-art GAMs across several real-world datasets. We further demonstrate that modellers can selectively impose required constraints at a modest trade-off in predictive performance, allowing the model to be fully tailored to application-specific interpretability and parametric-analysis requirements.

Bug reports, encompassing a wide range of bug types, are crucial for maintaining software quality. However, the increasing complexity and volume of bug reports pose a significant challenge in sole manual identification and assignment to the appropriate teams for resolution, as dealing with all the reports is time-consuming and resource-intensive. In this paper, we introduce a cross-project framework, dubbed Mutualistic Neural Active Learning (MNAL), designed for automated and more effective identification of bug reports from GitHub repositories boosted by human-machine collaboration. MNAL utilizes a neural language model that learns and generalizes reports across different projects, coupled with active learning to form neural active learning. A distinctive feature of MNAL is the purposely crafted mutualistic relation between the machine learners (neural language model) and human labelers (developers) when enriching the knowledge learned. That is, the most informative human-labeled reports and their corresponding pseudo-labeled ones are used to update the model while those reports that need to be labeled by developers are more readable and identifiable, thereby enhancing the human-machine teaming therein. We evaluate MNAL using a large scale dataset against the SOTA approaches, baselines, and different variants. The results indicate that MNAL achieves up to 95.8% and 196.0% effort reduction in terms of readability and identifiability during human labeling, respectively, while resulting in a better performance in bug report identification. Additionally, our MNAL is model-agnostic since it is capable of improving the model performance with various underlying neural language models. To further verify the efficacy of our approach, we conducted a qualitative case study involving 10 human participants, who rate MNAL as being more effective while saving more time and monetary resources.

GUI agents that control desktop computers via screenshot-and-click loops introduce a new class of vulnerability: the observation-to-action gap (mean 6.51 s on real OSWorld workloads) creates a Time-Of-Check, Time-Of-Use (TOCTOU) window during which an unprivileged attacker can manipulate the UI state. We formalize this as a Visual Atomicity Violation and characterize three concrete attack primitives: (A) Notification Overlay Hijack, (B) Window Focus Manipulation, and (C) Web DOM Injection. Primitive B, the closest desktop analog to Android Action Rebinding, achieves 100% action-redirection success rate with zero visual evidence at the observation time. We propose Pre-execution UI State Verification (PUSV), a lightweight three-layer defense that re-verifies the UI state immediately before each action dispatch: masked pixel SSIM at the click target (L1), global screenshot diff (L2a), and X Window snapshot diff (L2b). PUSV achieves 100% Action Interception Rate across 180 adversarial trials (135 Primitive A + 45 Primitive B) with zero false positives and < 0.1 s overhead. Against Primitive C (zero-visual-footprint DOM injection), PUSV reveals a structural blind spot (~0% AIR), motivating future OS+DOM defense-in-depth architectures. No single PUSV layer alone achieves full coverage; different primitives require different detection signals, validating the layered design.

In continual learning, the primary challenge is to learn new information without forgetting old knowledge. A common solution addresses this trade-off through regularization, penalizing changes to parameters critical for previous tasks. In most cases, this regularization term is directly added to the training loss and optimized with standard gradient descent, which blends learning and retention signals into a single update and does not explicitly separate essential parameters from redundant ones. As task sequences grow, this coupling can over-constrain the model, limiting forward transfer and leading to inefficient use of capacity. We propose a different approach that separates task learning from stability enforcement via operator splitting. The learning step focuses on minimizing the current task loss, while a proximal stability step applies a sparse regularizer to prune unnecessary parameters and preserve task-relevant ones. This turns the stability-plasticity into a negotiated update between two complementary operators, rather than a conflicting gradient. We provide theoretical justification for the splitting method on the continual-learning objective, and demonstrate that our proposed solver achieves state-of-the-art results on standard benchmarks, improving both stability and adaptability without the need for replay buffers, Bayesian sampling, or meta-learning components.

AI systems are increasingly embedded in multi-user social environments, yet most alignment frameworks conceptualize interaction as a dyadic relationship between a single user and an AI system. Livestreaming platforms challenge this assumption: interaction unfolds among streamers and audiences in real time, producing dynamic affective and social feedback loops. In this paper, we introduce the Triadic Loop, a conceptual framework that reconceptualizes alignment in AI co-hosted livestreaming as a temporally reinforced process of bidirectional adaptation among three actors: streamer $\leftrightarrow$ AI co-host, AI co-host $\leftrightarrow$ audience, and streamer $\leftrightarrow$ audience. Unlike instruction-following paradigms, bidirectional alignment requires each actor to continuously reshape the others, meaning misalignment in any sub-loop can destabilize the broader system. Drawing on literature from multi-party interaction, collaborative AI, and relational agents, we articulate how AI co-hosts function not only as mediators but as performative participants and community members shaping collective meaning-making. We further propose "strategic misalignment" as a mechanism for sustaining community engagement and introduce three relational evaluation constructs grounded in established instruments. The framework contributes a model of dynamic multi-party alignment, an account of cross-loop reinforcement, and design implications for AI co-hosts that sustain social coherence in participatory media environments.

As LM agents gain the ability to execute actions on real computer systems, we need ways to not only prevent harmful actions at scale but also effectively remediate harm when prevention fails. We formalize a solution to this neglected challenge in post-execution safeguards as harm recovery: the problem of optimally steering an agent from a harmful state back to a safe one in alignment with human preferences. We ground preference-aligned recovery through a formative user study that identifies valued recovery dimensions and produces a natural language rubric. Our dataset of 1,150 pairwise judgments reveals context-dependent shifts in attribute importance, such as preferences for pragmatic, targeted strategies over comprehensive long-term approaches. We operationalize these learned insights in a reward model, re-ranking multiple candidate recovery plans generated by an agent scaffold at test time. To evaluate recovery capabilities systematically, we introduce BackBench, a benchmark of 50 computer-use tasks that test an agent's ability to recover from harmful states. Human evaluation shows our reward model scaffold yields higher-quality recovery trajectories than base agents and rubric-based scaffolds. Together, these contributions lay the foundation for a new class of agent safety methods -- ones that confront harm not only by preventing it, but by navigating its aftermath with alignment and intent.

Barren-plateau results have established exponential gradient suppression as a widely cited obstacle to the scalability of variational quantum algorithms. When and whether these results extend to a given objective has been addressed through loss-specific arguments, but a general structural characterization has remained open. We show that the objective itself admits a fixed-observable representation if and only if the loss is affine in the measured statistics, thereby identifying the exact boundary of the standard concentration-based proof template. Existing transfer results for non-affine losses achieve this reduction under additional assumptions; our characterization implies that such a reduction is not structurally available for a class of non-affine objectives, placing them outside the automatic reach of the existing proof template. Beyond the affine regime, a chain-rule decomposition reveals three governing factors -- model responsivity, loss-side signal, and transmittance -- and induces a loss-class dichotomy: bounded-gradient losses inherit suppression, while amplification-capable losses can in principle counteract it. In the exponentially wide setting, both classes fail, but for different structural reasons. When the interface is instead designed at polynomial width -- exposing coarse-grained statistics rather than individual bitstring probabilities -- the exponential-dimensional obstruction is relaxed and the dichotomy plays a genuine role. In a numerical demonstration on a charge-conserving quantum system, the amplification-capable objective produces resolved gradients several orders of magnitude larger than affine and inheriting baselines at comparable shot budgets. Over the tested interval, its scaling trend is statistically distinguished from the exponential trend of both alternatives. The boundary is affine; what lies beyond it is a representation-design problem.

Looped transformers scale computational depth without increasing parameter count by repeatedly applying a shared transformer block and can be used for iterative refinement, where each loop rewrites a full fixed-size prediction in parallel. On difficult problems, such as those that require search-like computation, reaching a highly structured solution starting from noise can require long refinement trajectories. Learning such trajectories is challenging when training specifies only the target solution and provides no supervision over the intermediate refinement path. Diffusion models tackle this issue by corrupting data with varying magnitudes of noise and training the model to reverse it in a \textit{single step}. However, this process misaligns training and testing behaviour. We introduce Denoising Recursion Models, a method that similarly corrupts data with noise but trains the model to reverse the corruption over \textit{multiple} recursive steps. This strategy provides a tractable curriculum of intermediate states, while better aligning training with testing and incentivizing non-greedy, forward-looking generation. Through extensive experiments, we show this approach outperforms the Tiny Recursion Model (TRM) on ARC-AGI, where it recently achieved breakthrough performance.

This research investigates the performance and efficacy of machine learning models in stock prediction, comparing Artificial Neural Networks (ANNs), Quantum Qubit-based Neural Networks (QQBNs), and Quantum Qutrit-based Neural Networks (QQTNs). By outlining methodologies, architectures, and training procedures, the study highlights significant differences in training times and performance metrics across models. While all models demonstrate robust accuracies above 70%, the Quantum Qutrit-based Neural Network consistently outperforms with advantages in risk-adjusted returns, measured by the Sharpe ratio, greater consistency in prediction quality through the Information Coefficient, and enhanced robustness under varying market conditions. The QQTN not only surpasses its classical and qubit-based counterparts in multiple quantitative and qualitative metrics but also achieves comparable performance with significantly reduced training times. These results showcase the promising prospects of Quantum Qutrit-based Neural Networks in practical financial applications, where real-time processing is critical. By achieving superior accuracy, efficiency, and adaptability, the proposed models underscore the transformative potential of quantum-inspired approaches, paving the way for their integration into computationally intensive fields.

Quantum kernel methods have been proposed as a promising approach for leveraging near-term quantum computers for supervised learning, yet rigorous benchmarks against strong classical baselines remain scarce. We present a comprehensive empirical study of quantum kernel support vector machines (QSVMs) across nine binary classification datasets, four quantum feature maps, three classical kernels, and multiple noise models, totalling 970 experiments with strict nested cross-validation. Our analysis spans four phases: (i) statistical significance testing, revealing that none of 29 pairwise quantum-classical comparisons reach significance at $α= 0.05$; (ii) learning curve analysis over six training fractions, showing steeper quantum slopes on six of eight datasets that nonetheless fail to close the gap to the best classical baseline; (iii) hardware validation on IBM ibm_fez (Heron r2), demonstrating kernel fidelity $r \geq 0.976$ across six experiments; and (iv) seed sensitivity analysis confirming reproducibility (mean CV 1.4%). A Kruskal-Wallis factorial analysis reveals that dataset choice dominates performance variance ($\varepsilon^2 = 0.73$), while kernel type accounts for only 9%. Spectral analysis offers a mechanistic explanation: current quantum feature maps produce eigenspectra that are either too flat or too concentrated, missing the intermediate profile of the best classical kernel, the radial basis function (RBF). Quantum kernel training (QKT) via kernel-target alignment yields the single competitive result -- balanced accuracy 0.968 on breast cancer -- but with ~2,000x computational overhead. Our findings provide actionable guidelines for quantum kernel research. The complete benchmark suite is publicly available to facilitate reproduction and extension.

We propose a scalable, multifactorial experimental framework that systematically probes LLM sensitivity to subtle semantic changes in pairwise document comparison. We analogize this as a needle-in-a-haystack problem: a single semantically altered sentence (the needle) is embedded within surrounding context (the hay), and we vary the perturbation type (negation, conjunction swap, named entity replacement), context type (original vs. topically unrelated), needle position, and document length across all combinations, testing five LLMs on tens of thousands of document pairs. Our analysis reveals several striking findings. First, LLMs exhibit a within-document positional bias distinct from previously studied candidate-order effects: most models penalize semantic differences more harshly when they occur earlier in a document. Second, when the altered sentence is surrounded by topically unrelated context, it systematically lowers similarity scores and induces bipolarized scores that indicate either very low or very high similarity. This is consistent with an interpretive frame account in which topically-related context may allow models to contextualize and downweight the alterations. Third, each LLM produces a qualitatively distinct scoring distribution, a stable "fingerprint" that is invariant to perturbation type, yet all models share a universal hierarchy in how leniently they treat different perturbation types. Together, these results demonstrate that LLM semantic similarity scores are sensitive to document structure, context coherence, and model identity in ways that go beyond the semantic change itself, and that the proposed framework offers a practical, LLM-agnostic toolkit for auditing and comparing scoring behavior across current and future models.

Large language models (LLMs) have enabled natural-language-driven automation of electronic design automation (EDA) workflows, but reliable execution of generated scripts remains a fundamental challenge. In LLM-based EDA tasks, failures arise not from syntax errors but from violations of implicit structural dependencies over design objects, including invalid acquisition paths, missing prerequisites, and incompatible API usage. Existing approaches address these failures through tool-in-the-loop debugging, repeatedly executing and repairing programs using runtime feedback. While effective, this paradigm couples correctness to repeated tool invocation, leading to high latency and poor scalability in multi-step settings. We propose to eliminate tool-in-the-loop debugging by enforcing structural correctness prior to execution. Each task is represented as a structural dependency graph that serves as an explicit execution contract, and a verifier-guided synthesis framework enforces this contract through graph-conditioned retrieval, constrained generation, and staged pre-execution verification with diagnosis-driven repair. On single-step tasks, our method improves pass rate from 73.0% (LLM+RAG) and 76.0% (tool-in-loop) to 82.5%, while requiring exactly one tool call per task and reducing total tool calls by more than 2x. On multi-step tasks, pass rate improves from 30.0% to 70.0%, and further to 84.0% with trajectory-level reflection. Uncertainty-aware filtering further reduces verifier false positives from 20.0% to 6.7% and improves precision from 80.0% to 93.3%. These results show that enforcing structural consistency prior to execution decouples correctness from tool interaction, improving both reliability and efficiency in long-horizon EDA code generation.

Artificial Intelligence (AI) surrogate models provide a computationally efficient alternative to full-physics simulations, but no public datasets currently exist for training and validating models of high-explosive-driven, multi-material shock dynamics. Simulating shock propagation is challenging due to the need for material-specific equations of state (EOS) and models of plasticity, phase change, damage, fluid instabilities, and multi-material interactions. Explosive-driven shocks further require reactive material models to capture detonation physics. To address this gap, we introduce the High-Explosives and Affected Targets (HEAT) dataset, a physics-rich collection of two-dimensional, cylindrically symmetric simulations generated using an Eulerian multi-material shock-propagation code developed at Los Alamos National Laboratory. HEAT consists of two partitions: expanding shock-cylinder (CYL) simulations and Perturbed Layered Interface (PLI) simulations. Each entry includes time series of thermodynamic fields (pressure, density, temperature), kinematic fields (position, velocity), and continuum quantities such as stress. The CYL partition spans a range of materials, including metals (aluminum, copper, depleted uranium, stainless steel, tantalum), a polymer, water, gases (air, nitrogen), and a detonating material. The PLI partition explores varied geometries with fixed materials: copper, aluminum, stainless steel, polymer, and high explosive. HEAT captures key phenomena such as shock propagation, momentum transfer, plastic deformation, and thermal effects, providing a benchmark dataset for AI/ML models of multi-material shock physics.

Scaling data and artificial neural networks has transformed AI, driving breakthroughs in language and vision. Whether similar principles apply to modeling brain activity remains unclear. Here we leveraged a dataset of 3.1 million neurons from the visual cortex of 73 mice across 323 sessions, totaling more than 150 billion neural tokens recorded during natural movies, images and parametric stimuli, and behavior. We train multi-modal, multi-task models that support three regimes flexibly at test time: neural prediction, behavioral decoding, neural forecasting, or any combination of the three. OmniMouse achieves state-of-the-art performance, outperforming specialized baselines across nearly all evaluation regimes. We find that performance scales reliably with more data, but gains from increasing model size saturate. This inverts the standard AI scaling story: in language and computer vision, massive datasets make parameter scaling the primary driver of progress, whereas in brain modeling -- even in the mouse visual cortex, a relatively simple system -- models remain data-limited despite vast recordings. The observation of systematic scaling raises the possibility of phase transitions in neural modeling, where larger and richer datasets might unlock qualitatively new capabilities, paralleling the emergent properties seen in large language models. Code available at https://github.com/enigma-brain/omnimouse.

Recovering latent structure from count data has received considerable attention in network inference, particularly when one seeks both cross-group interactions and within-group similarity patterns in bipartite networks, which is widely used in ecology research. Such networks are often sparse and inherently imperfect in their detection. Existing models mainly focus on interaction recovery, while the induced similarity graphs are much less studied. Moreover, sparsity is often not controlled, and scale is unbalanced, leading to oversparse or poorly rescaled estimates with degrading structural recovery. To address these issues, we propose a framework for structured sparse nonnegative low-rank factorization with detection probability estimation. We impose nonconvex $\ell_{1/2}$ regularization on the latent similarity and connectivity structures to promote sparsity within-group similarity and cross-group connectivity with better relative scale. The resulting optimization problem is nonconvex and nonsmooth. To solve it, we develop an ADMM-based algorithm with adaptive penalization and scale-aware initialization and establish its asymptotic feasibility and KKT stationarity of cluster points under mild regularity conditions. Experiments on synthetic and real-world ecological datasets demonstrate improved recovery of latent factors and similarity/connectivity structure relative to existing baselines.

Principal Component Analysis (PCA) is a fundamental tool for representation learning, but its global linear formulation fails to capture the structure of data supported on curved manifolds. In contrast, manifold learning methods model nonlinearity but often sacrifice the spectral structure and stability of PCA. We propose \emph{Geodesic Tangent Space Aggregation PCA (GTSA-PCA)}, a geometric extension of PCA that integrates curvature awareness and geodesic consistency within a unified spectral framework. Our approach replaces the global covariance operator with curvature-weighted local covariance operators defined over a $k$-nearest neighbor graph, yielding local tangent subspaces that adapt to the manifold while suppressing high-curvature distortions. We then introduce a geodesic alignment operator that combines intrinsic graph distances with subspace affinities to globally synchronize these local representations. The resulting operator admits a spectral decomposition whose leading components define a geometry-aware embedding. We further incorporate semi-supervised information to guide the alignment, improving discriminative structure with minimal supervision. Experiments on real datasets show consistent improvements over PCA, Kernel PCA, Supervised PCA and strong graph-based baselines such as UMAP, particularly in small sample size and high-curvature regimes. Our results position GTSA-PCA as a principled bridge between statistical and geometric approaches to dimensionality reduction.

Despite the perceived success of large-scale dataset distillation (DD) methods, recent evidence finds that simple random image baselines perform on-par with state-of-theart DD methods like SRe2L due to the use of soft labels during downstream model training. This is in contrast with the findings in coreset literature, where high-quality coresets consistently outperform random subsets in the hardlabel (HL) setting. To understand this discrepancy, we perform a detailed scalability analysis to examine the role of data quality under different label regimes, ranging from abundant soft labels (termed as SL+KD regime) to fixed soft labels (SL) and hard labels (HL). Our analysis reveals that high-quality coresets fail to convincingly outperform the random baseline in both SL and SL+KD regimes. In the SL+KD setting, performance further approaches nearoptimal levels relative to the full dataset, regardless of subset size or quality, for a given compute budget. This performance saturation calls into question the widespread practice of using soft labels for model evaluation, where unlike the HL setting, subset quality has negligible influence. A subsequent systematic evaluation of five large-scale and four small-scale DD methods in the HL setting reveals that only RDED reliably outperforms random baselines on ImageNet-1K, but can still lag behind strong coreset methods due to its over-reliance on easy sample patches. Based on this, we introduce CAD-Prune, a compute-aware pruning metric that efficiently identifies samples of optimal difficulty for a given compute budget, and use it to develop CA2D, a compute-aligned DD method, outperforming current DD methods on ImageNet-1K at various IPC settings. Together, our findings uncover many insights into current DD research and establish useful tools to advance dataefficient learning for both coresets and DD.

3D-IC netlist partitioning is commonly optimized using proxy objectives, while final PPA is treated as a costly evaluation rather than an optimization signal. This proxy-driven paradigm makes it difficult to reliably translate additional PPA evaluations into better PPA outcomes. To bridge this gap, we present DOPP (D-Optimal PPA-driven partitioning selection), an approach that bridges the gap between proxies and true PPA metrics. Across eight 3D-IC designs, our framework improves PPA over Open3DBench (average relative improvements of 9.99% congestion, 7.87% routed wirelength, 7.75% WNS, 21.85% TNS, and 1.18% power). Compared with exhaustive evaluation over the full candidate set, DOPP achieves comparable best-found PPA while evaluating only a small fraction of candidates, substantially reducing evaluation cost. By parallelizing evaluations, our method delivers these gains while maintaining wall-clock runtime comparable to traditional baselines.

Large language model (LLM)-based systems are increasingly deployed to conduct scientific research autonomously, yet whether their reasoning adheres to the epistemic norms that make scientific inquiry self-correcting is poorly understood. Here, we evaluate LLM-based scientific agents across eight domains, spanning workflow execution to hypothesis-driven inquiry, through more than 25,000 agent runs and two complementary lenses: (i) a systematic performance analysis that decomposes the contributions of the base model and the agent scaffold, and (ii) a behavioral analysis of the epistemological structure of agent reasoning. We observe that the base model is the primary determinant of both performance and behavior, accounting for 41.4% of explained variance versus 1.5% for the scaffold. Across all configurations, evidence is ignored in 68% of traces, refutation-driven belief revision occurs in 26%, and convergent multi-test evidence is rare. The same reasoning pattern appears whether the agent executes a computational workflow or conducts hypothesis-driven inquiry. They persist even when agents receive near-complete successful reasoning trajectories as context, and the resulting unreliability compounds across repeated trials in epistemically demanding domains. Thus, current LLM-based agents execute scientific workflows but do not exhibit the epistemic patterns that characterize scientific reasoning. Outcome-based evaluation cannot detect these failures, and scaffold engineering alone cannot repair them. Until reasoning itself becomes a training target, the scientific knowledge produced by such agents cannot be justified by the process that generated it.

Latent Diffusion Models (LDMs) achieve high-fidelity synthesis but suffer from latent space brittleness, causing discontinuous semantic jumps during editing. We introduce a Riemannian framework to diagnose this instability by analyzing the generative Jacobian, decomposing geometry into \textit{Local Scaling} (capacity) and \textit{Local Complexity} (curvature). Our study uncovers a \textbf{``Geometric Decoupling"}: while curvature in normal generation functionally encodes image detail, OOD generation exhibits a functional decoupling where extreme curvature is wasted on unstable semantic boundaries rather than perceptible details. This geometric misallocation identifies ``Geometric Hotspots" as the structural root of instability, providing a robust intrinsic metric for diagnosing generative reliability.

Vision-Language Models (VLMs) are increasingly deployed in settings where reliable visual grounding carries operational consequences, yet their behavior under progressively coercive prompt phrasing remains undercharacterized. Existing hallucination benchmarks predominantly rely on neutral prompts and binary detection, leaving open how both the incidence and the intensity of fabrication respond to graded linguistic pressure across structurally distinct task types. We present Ghost-100, a procedurally constructed benchmark of 800 synthetically generated images spanning eight categories across three task families -- text-illegibility, time-reading, and object-absence -- each designed under a negative-ground-truth principle that guarantees the queried target is absent, illegible, or indeterminate by construction. Every image is paired with five prompts drawn from a structured 5-Level Prompt Intensity Framework, holding the image and task identity fixed while varying only directive force, so that tone is isolated as the sole independent variable. We adopt a dual-track evaluation protocol: a rule-based H-Rate measuring the proportion of responses in which a model crosses from grounded refusal into unsupported positive commitment, and a GPT-4o-mini-judged H-Score on a 1-5 scale characterizing the confidence and specificity of fabrication once it occurs. We additionally release a three-stage automated validation workflow, which retrospectively confirms 717 of 800 images as strictly compliant. Evaluating nine open-weight VLMs, we find that H-Rate and H-Score dissociate substantially across model families, reading-style and presence-detection subsets respond to prompt pressure in qualitatively different ways, and several models exhibit non-monotonic sensitivity peaking at intermediate tone levels -- patterns that aggregate metrics obscure.

Lossy compression is widely used to reduce storage and I/O costs for large-scale particle datasets in scientific applications such as cosmology, molecular dynamics, and fluid dynamics, where clustering structures (e.g., single-linkage or Friends-of-Friends) are critical for downstream analysis; however, existing compressors typically provide only pointwise error bounds on particle positions and offer no guarantees on preserving clustering outcomes, and even small perturbations can alter cluster connectivity and compromise scientific validity. We propose a correction-based technique to preserve single-linkage clustering under lossy compression, operating on decompressed data from off-the-shelf compressors such as SZ3 and Draco. Our key contributions are threefold: (1) a clustering-aware correction algorithm that identifies vulnerable particle pairs via spatial partitioning and local neighborhood search; (2) an optimization-based formulation that enforces clustering consistency using projected gradient descent with a loss that encodes pairwise distance violations; and (3) a scalable GPU-accelerated and distributed implementation for large-scale datasets. Experiments on cosmology and molecular dynamics datasets show that our method effectively preserves clustering results while maintaining competitive compression performance compared with SZ3, ZFP, Draco, LCP, and space-filling-curve-based schemes.

We study online learning for new products on a platform that makes capacity-constrained assortment decisions on which products to offer. For a newly listed product, its quality is initially unknown, and quality information propagates through social learning: when a customer purchases a new product and leaves a review, its quality is revealed to both the platform and future customers. Since reviews require purchases, the platform must feature new products in the assortment ("explore") to generate reviews to learn about new products. Such exploration is costly because customer demand for new products is lower than for incumbent products. We characterize the optimal assortments for exploration to minimize regret, addressing two questions. (1) Should the platform offer a new product alone or alongside incumbent products? The former maximizes the purchase probability of the new product but yields lower short-term revenue. Despite the lower purchase probability, we show it is always optimal to pair the new product with the top incumbent products. (2) With multiple new products, should the platform explore them simultaneously or one at a time? We show that the optimal number of new products to explore simultaneously has a simple threshold structure: it increases with the "potential" of the new products and, surprisingly, does not depend on their individual purchase probabilities. We also show that two canonical bandit algorithms, UCB and Thompson Sampling, both fail in this setting for opposite reasons: UCB over-explores while Thompson Sampling under-explores. Our results provide structural insights on how platforms should learn about new products through assortment decisions.

Model transformations are central to MDE, but formal verification is difficult because mainstream transformation languages are undecidable. DSLTrans was designed to be Turing-incomplete to improve verifiability, yet earlier verification based on path-condition enumeration still suffered exponential blow-up and did not scale to realistic cases. We present a tractable verification workflow for DSLTrans and formalize when it is complete. The method combines three contributions: (i) a Cutoff Theorem proving that bounded model checking is complete for a precise DSLTrans fragment and positive existence/traceability properties, turning an infinite search into a finite computable bound; (ii) composable, soundness-preserving optimizations (per-class bounds, CEGAR-based fragment verification, and trace-aware dependency analysis) that reduce SMT encoding size; and (iii) a Z3-based implementation evaluated on realistic transformations from the ATL Zoo and related benchmarks. On 29 concrete transformations and 899 properties spanning compiler lowering, schema translation, behavioral modeling, graph mapping, and stress tests, 552 properties are proved, 345 produce concrete counterexamples (including intentional negative and boundary cases), and only 2 remain undecided within timeout. For properties beyond the tractability budget, we introduce tractability-driven refinement (precondition specialization, postcondition decomposition, and transformation instrumentation), achieving up to 112x speedup while eliminating spurious counterexamples. The workflow is supported by a web IDE and a concrete execution engine for runtime validation.

Vision-Language-Action (VLA) models fail systematically on long-horizon manipulation tasks despite strong short-horizon performance. We show that this failure is not resolved by extending context length alone in the current reactive execution setting; instead, it stems from three recurring execution-loop deficiencies: the memory gap, the verification gap, and the recovery gap. We present HELM, a model-agnostic framework that addresses these deficiencies with three components: an Episodic Memory Module (EMM) that retrieves key task history via CLIP-indexed keyframes, a learned State Verifier (SV) that predicts action failure before execution from observation, action, subgoal, and memory-conditioned context, and a Harness Controller (HC) that performs rollback and replanning. The SV is the core learning contribution: it consistently outperforms rule-based feasibility checks and ensemble uncertainty baselines, and its effectiveness depends critically on access to episodic memory. On LIBERO-LONG, HELM improves task success rate by 23.1 percentage points over OpenVLA (58.4% to 81.5%), while extending the context window to H=32 yields only a 5.4-point gain and same-budget LoRA adaptation remains 12.2 points below HELM. HELM also improves long-horizon performance on CALVIN and substantially boosts recovery success under controlled perturbations. Ablations and mechanism analyses isolate the contribution of each component, and we release LIBERO-Recovery as a perturbation-injection protocol for evaluating failure recovery in long-horizon manipulation.

Reinforcement Learning from Human Feedback (RLHF) is central to aligning Large Language Models (LLMs), yet it introduces a critical vulnerability: an imperfect Reward Model (RM) can become a single point of failure when it fails to penalize unsafe behaviors. While existing red-teaming approaches primarily target policy-level weaknesses, they overlook what we term systemic weaknesses cases where both the core LLM and the RM fail in tandem. We present ARES, a framework that systematically discovers and mitigates such dual vulnerabilities. ARES employs a ``Safety Mentor'' that dynamically composes semantically coherent adversarial prompts by combining structured component types (topics, personas, tactics, goals) and generates corresponding malicious and safe responses. This dual-targeting approach exposes weaknesses in both the core LLM and the RM simultaneously. Using the vulnerabilities gained, ARES implements a two-stage repair process: first fine-tuning the RM to better detect harmful content, then leveraging the improved RM to optimize the core model. Experiments across multiple adversarial safety benchmarks demonstrate that ARES substantially enhances safety robustness while preserving model capabilities, establishing a new paradigm for comprehensive RLHF safety alignment.

Apple Neural Engine (ANE) is a dedicated neural processing unit (NPU) present in every Apple Silicon chip. Mixture-of-Experts (MoE) LLMs improve inference efficiency via sparse activation but are challenging for NPUs in three ways: expert routing is unpredictable and introduces dynamic tensor shapes that conflict with the shape-specific constraints of NPUs; several irregular operators, e.g., top-k, scatter/gather, etc., are not NPU-friendly; and launching many small expert kernels incurs substantial dispatch and synchronization overhead. NPUs are designed to offload AI compute from CPU and GPU; our goal is to enable such offloading for MoE inference, particularly during prefill, where long-context workloads consume substantial system resources. This paper presents NPUMoE, a runtime inference engine that accelerates MoE execution on Apple Silicon by offloading dense, static computation to NPU, while preserving a CPU/GPU fallback path for dynamic operations. NPUMoE uses offline calibration to estimate expert capacity and popularity that drives three key techniques: (1) Static tiers for expert capacity to address dynamic expert routing (2) Grouped expert execution to mitigate NPU concurrency limits (3) Load-aware expert compute graph residency to reduce CPU-NPU synchronization overhead. Experiments on Apple M-series devices using three representative MoE LLMs and four long-context workloads show that NPUMoE consistently outperforms baselines, reducing latency by 1.32x-5.55x, improving energy efficiency by 1.81x-7.37x, and reducing CPU-cycle usage by 1.78x-5.54x through effective NPU offloading.

Scientific feasibility assessment asks whether a claim is consistent with established knowledge and whether experimental evidence could support or refute it. We frame feasibility assessment as a diagnostic reasoning task in which, given a hypothesis, a model predicts feasible or infeasible and justifies its decision. We evaluate large language models (LLMs) under controlled knowledge conditions (hypothesis-only, with experiments, with outcomes, or both) and probe robustness by progressively removing portions of the experimental and/or outcome context. Across multiple LLMs and two datasets, providing outcome evidence is generally more reliable than providing experiment descriptions. Outcomes tend to improve accuracy beyond what internal knowledge alone provides, whereas experimental text can be brittle and may degrade performance when the context is incomplete. These findings clarify when experimental evidence benefits LLM-based feasibility assessment and when it introduces fragility.

Semi-Markov Conditional Random Fields (semi-CRFs) assign labels to segments of a sequence rather than to individual positions, enabling exact inference over segment-level features and principled uncertainty estimates at their boundaries. However, existing implementations must materialize a large edge potential tensor whose size grows with sequence length, maximum segment length, and label count, becoming prohibitive for speech-scale state spaces and intractable at genomic scales where sequences can exceed 100,000 positions. This memory bottleneck has limited the adoption of exact segment-level inference for long sequences and large label sets. We identify that the core inefficiency is materializing edge potentials that can instead be evaluated on-the-fly from a compact prefix-sum array, and make several improvements. First, replacing the stored edge tensor with prefix-sum lookup reduces the memory footprint by a factor proportional to the product of segment length and label count. Second, a streaming forward-backward pass with checkpoint-boundary normalization keeps working memory sublinear in sequence length while preserving exact gradients. Third, zero-centered cumulative scores control numerical drift and induce an adaptive duration prior under label imbalance. We integrate these ideas into Flash-SemiCRF, a fused Triton kernel that enables exact semi-CRF inference on previously intractable problem sizes. Available at https://github.com/biobenkj/flash-semicrf.

Existing web agents typically initiate exploration from the root URL, which is inefficient for complex websites with deep hierarchical structures. Without a global view of the website's structure, agents frequently fall into navigation traps, explore irrelevant branches, or fail to reach target information within a limited budget. We propose Mango, a multi-agent web navigation method that leverages the website structure to dynamically determine optimal starting points. We formulate URL selection as a multi-armed bandit problem and employ Thompson Sampling to adaptively allocate the navigation budget across candidate URLs. Furthermore, we introduce an episodic memory component to store navigation history, enabling the agent to learn from previous attempts. Experiments on WebVoyager demonstrate that Mango achieves a success rate of 63.6% when using GPT-5-mini, outperforming the best baseline by 7.3%. Furthermore, on WebWalkerQA, Mango attains a 52.5% success rate, surpassing the best baseline by 26.8%. We also demonstrate the generalizability of Mango using both open-source and closed-source models as backbones. Our data and code are open-source and available at https://github.com/VichyTong/Mango.

Detecting jailbreak behaviour in large language models remains challenging, particularly when strongly aligned models produce harmful outputs only rarely. In this work, we present an empirical study of output based jailbreak detection under realistic conditions using the JailbreakBench Behaviors dataset and multiple generator models with varying alignment strengths. We evaluate both a lexical TF-IDF detector and a generation inconsistency based detector across different sampling budgets. Our results show that single output evaluation systematically underestimates jailbreak vulnerability, as increasing the number of sampled generations reveals additional harmful behaviour. The most significant improvements occur when moving from a single generation to moderate sampling, while larger sampling budgets yield diminishing returns. Cross generator experiments demonstrate that detection signals partially generalise across models, with stronger transfer observed within related model families. A category level analysis further reveals that lexical detectors capture a mixture of behavioural signals and topic specific cues, rather than purely harmful behaviour. Overall, our findings suggest that moderate multi sample auditing provides a more reliable and practical approach for estimating model vulnerability and improving jailbreak detection in large language models. Code will be released.

Translating business problems into well-specified machine learning solutions is a prerequisite for successful AI systems, yet this upstream translation is still one of the least supported steps in existing methodologies. We conduct a structured narrative literature review of 18 approaches spanning requirements engineering (RE), machine learning (ML) project management, and automation. We organize these approaches into a taxonomy of four families and compare them across six input artifact categories, six output artifact categories, and a transformation framework of seven stages, grounded in RE refinement theory and ML lifecycle process. Our study shows that most approaches list ML task or algorithm specification among their expected outputs, yet only four provide partial guidance for deriving it, and none provides systematic guidance. We characterize this gap as the Analytics Translation Problem (ATP) and derive five research recommendations addressing multi-formulation exploration, task derivation guidance, constraint-algorithm filtering, probabilistic traceability, and data-triggered revision.

Fault detection and diagnosis are critical for the optimal and safe operation of industrial processes. The correlations among sensors often display non-Euclidean structures where graph neural networks (GNNs) are widely used therein. However, for large-scale systems, local, global, and dynamic relations extensively exist among sensors, and traditional GNNs often overlook such complex and multi-level structures for various problems including the fault diagnosis. To address this issue, we propose a structure-aware multi-level temporal graph network with local-global feature fusion for industrial fault diagnosis. First, a correlation graph is dynamically constructed using Pearson correlation coefficients to capture relationships among process variables. Then, temporal features are extracted through long short-term memory (LSTM)-based encoder, whereas the spatial dependencies among sensors are learned by graph convolution layers. A multi-level pooling mechanism is used to gradually coarsen and learn meaningful graph structures, to capture higher-level patterns while keeping important fault related details. Finally, a fusion step is applied to combine both detailed local features and overall global patterns before the final prediction. Experimental evaluations on the Tennessee Eastman process (TEP) demonstrate that the proposed model achieves superior fault diagnosis performance, particularly for complex fault scenarios, outperforming various baseline methods.

Class imbalance is a widespread challenge in NLP tasks, significantly hindering robust performance across diverse domains and applications. We introduce Hardness-Aware Meta-Resample (HAMR), a unified framework that adaptively addresses both class imbalance and data difficulty. HAMR employs bi-level optimizations to dynamically estimate instance-level weights that prioritize genuinely challenging samples and minority classes, while a neighborhood-aware resampling mechanism amplifies training focus on hard examples and their semantically similar neighbors. We validate HAMR on six imbalanced datasets covering multiple tasks and spanning biomedical, disaster response, and sentiment domains. Experimental results show that HAMR achieves substantial improvements for minority classes and consistently outperforms strong baselines. Extensive ablation studies demonstrate that our proposed modules synergistically contribute to performance gains and highlight HAMR as a flexible and generalizable approach for class imbalance adaptation. Code is available at https://github.com/trust-nlp/ImbalanceLearning.

Low-resource machine translation requires methods that differ from those used for high-resource languages. This paper proposes a novel in-context learning approach to support low-resource machine translation of the Coptic language to English, with syntactic augmentation from Universal Dependencies parses of input sentences. Building on existing work using bilingual dictionaries to support inference for vocabulary items, we add several representations of syntactic analyses to our inputs , specifically exploring the inclusion of raw parser outputs, verbalizations of parses in plain English, and targeted instructions of difficult constructions identified in sub-trees and how they can be translated. Our results show that while syntactic information alone is not as useful as dictionary-based glosses, combining retrieved dictionary items with syntactic information achieves significant gains across model sizes, achieving new state-of-the-art translation results for Coptic.

The retina provides a unique, noninvasive window into Alzheimer's disease (AD) and dementia, capturing early structural changes through morphometric features, while systemic and lifestyle risk factors reflect well-established contributors to disease susceptibility long before clinical symptom onset. However, current retinal analysis frameworks typically model imaging and risk factors separately, limiting their ability to capture joint multimodal patterns critical for early risk prediction. Moreover, existing methods rarely incorporate mechanisms to organize or align patients with similar retinal and clinical characteristics, constraining the learning of coherent cross-modal associations. To address these limitations, we introduce REVEAL (REtinal-risk Vision-Language Early Alzheimer's Learning), a framework that aligns color fundus photographs with individualized disease-specific risk profiles for predicting incident AD and dementia, on average 8 years before diagnosis (range: 1-11 years). Because real-world risk factors are structured questionnaire data, we translate them into clinically interpretable narratives compatible with pretrained vision-language models (VLMs). We further propose a group-aware contrastive learning (GACL) strategy that clusters patients with similar retinal morphometry and risk factors as positive pairs, strengthening multimodal alignment. This unified representation learning framework substantially outperforms state-of-the-art retinal imaging models paired with clinical text encoders, as well as general-purpose VLMs, demonstrating the value of jointly modeling retinal biomarkers and clinical risk factors. By providing a generalizable and noninvasive approach for early AD and dementia risk stratification, REVEAL has the potential to enable earlier intervention and improve preventive care at the population level.

Large Language Models (LLMs) remain vulnerable to optimization-based jailbreak attacks that exploit internal gradient structure. While Sparse Autoencoders (SAEs) are widely used for interpretability, their robustness implications remain underexplored. We present a study of integrating pretrained SAEs into transformer residual streams at inference time, without modifying model weights or blocking gradients. Across four model families (Gemma, LLaMA, Mistral, Qwen) and two strong white-box attacks (GCG, BEAST) plus three black-box benchmarks, SAE-augmented models achieve up to a 5x reduction in jailbreak success rate relative to the undefended baseline and reduce cross-model attack transferability. Parametric ablations reveal (i) a monotonic dose-response relationship between L0 sparsity and attack success rate, and (ii) a layer-dependent defense-utility tradeoff, where intermediate layers balance robustness and clean performance. These findings are consistent with a representational bottleneck hypothesis: sparse projection reshapes the optimization geometry exploited by jailbreak attacks.

An active challenge in developing multimodal machine learning (ML) models for healthcare is handling missing modalities during training and deployment. As clinical datasets are inherently temporal and sparse in terms of modality presence, capturing the underlying predictive signal via diagnostic multimodal ML models while retaining model explainability remains an ongoing challenge. In this work, we address this by re-framing clinical diagnosis as an autoregressive sequence modeling task, utilizing causal decoders from large language models (LLMs) to model a patient's multimodal trajectory. We first introduce a missingness-aware contrastive pre-training objective that integrates multiple modalities in datasets with missingness in a shared latent space. We then show that autoregressive sequence modeling with transformer-based architectures outperforms baselines on the MIMIC-IV and eICU fine-tuning benchmarks. Finally, we use interpretability techniques to move beyond performance boosts and find that across various patient stays, removing modalities leads to divergent behavior that our contrastive pre-training mitigates. By abstracting clinical diagnosis as sequence modeling and interpreting patient stay trajectories, we develop a framework to profile and handle missing modalities while addressing the canonical desideratum of safe, transparent clinical AI.

Nonlinear machine-learning models are increasingly used to discover causal relationships in time-series data, yet the interpretation of their outputs remains poorly understood. In particular, causal scores produced by regularized neural autoregressive models are often treated as analogues of regression coefficients, leading to misleading claims of statistical significance. In this paper, we argue that causal relevance in nonlinear time-series models should be evaluated through forecast necessity rather than coefficient magnitude, and we present a practical evaluation procedure for doing so. We present an interpretable evaluation framework based on systematic edge ablation and forecast comparison, which tests whether a candidate causal relationship is required for accurate prediction. Using Neural Additive Vector Autoregression as a case study model, we apply this framework to a real-world case study of democratic development, modeled as a multivariate time series of panel data - democracy indicators across 139 countries. We show that relationships with similar causal scores can differ dramatically in their predictive necessity due to redundancy, temporal persistence, and regime-specific effects. Our results demonstrate how forecast-necessity testing supports more reliable causal reasoning in applied AI systems and provides practical guidance for interpreting nonlinear time-series models in high-stakes domains.

Masked diffusion large language models (dLLMs) are a promising alternative to autoregressive generation. While reinforcement learning (RL) methods have recently been adapted to dLLM fine-tuning, their objectives typically depend on sequence-level marginal likelihoods, which are intractable for masked diffusion models. To address this, we derive Discrete Tilt Matching (DTM), a likelihood-free method that recasts dLLM fine-tuning as state-level matching of local unmasking posteriors under reward tilting. DTM takes the form of a weighted cross-entropy objective with explicit minimizer, and admits control variates that improve training stability. On a synthetic maze-planning task, we analyze how DTM's annealing schedule and control variates affect training stability and prevent mode collapse. At scale, fine-tuning LLaDA-8B-Instruct with DTM yields strong gains on Sudoku and Countdown while remaining competitive on MATH500 and GSM8K.

Masked diffusion language models such as LLaDA2.1 rely on Token-to-Token (T2T) editing to correct their own generation errors: whenever a different token crosses a confidence threshold, the committed token is overwritten. We identify three structural failure modes of this rule. The trigger cannot fire when no single alternative is confident enough; the replacement is computed under a context that may itself contain errors; and the uniform perturbations used to train the T2T stream do not resemble the coherent, semantically plausible mistakes that the model actually makes at inference. As an alternative, we propose Token-to-Mask (T2M) remasking. Rather than overwriting a suspect token with a new guess, T2M resets the position to the mask state, so that the next denoising step re-predicts it from an in-distribution context. The method is training-free, modifies only the editing rule, and introduces no new parameters. We pair it with three detection heuristics and give a short theoretical account of why a mask is a better conditioning signal than an erroneous token. Across 8 benchmarks, T2M improves accuracy on tasks that require exact token-level output. Its largest gain is +5.92 points on CMATH, where we attribute 79.9% of baseline errors to last-mile corruption (correct reasoning followed by a garbled final answer); T2M repairs 41.3% of these cases.

Humor holds up a mirror to social perception: what we find funny often reflects who we are and how we judge others. When language models engage with humor, their reactions expose the social assumptions they have internalized from training data. In this paper, we investigate counterfactual unfairness through humor by observing how the model's responses change when we swap who speaks and who is addressed while holding other factors constant. Our framework spans three tasks: humor generation refusal, speaker intention inference, and relational/societal impact prediction, covering both identity-agnostic humor and identity-specific disparagement humor. We introduce interpretable bias metrics that capture asymmetric patterns under identity swaps. Experiments across state-of-the-art models reveal consistent relational disparities: jokes told by privileged speakers are refused up to 67.5% more often, judged as malicious 64.7% more frequently, and rated up to 1.5 points higher in social harm on a 5-point scale. These patterns highlight how sensitivity and stereotyping coexist in generative models, complicating efforts toward fairness and cultural alignment.

Many neural network (NN) verification systems represent the network's input-output relation as a constraint program. Sound and complete, representations involve integer constraints, for simulating the activations. Recent works convexly relax the integer constraints, improving performance, at the cost of soundness. Convex relaxations consider outputs that are unreachable by the original network. We study the worst case divergence between the original network and its convex relaxations; both qualitatively and quantitatively. The relaxations' space forms a lattice, where the top element corresponds to a full relaxation, with every neuron linearized. The bottom element corresponds to the original network. We provide analytical upper and lower bounds for the $\ell_\infty$-distance between the fully relaxed and original outputs. This distance grows exponentially, w.r.t. the network's depth, and linearly w.r.t. the input's radius. The misclassification probability exhibits a step-like behavior, w.r.t. input radius. Our results are supported by experiments on MNIST, Fashion MNIST and random networks.

Machine learning emulators have shown extraordinary skill in forecasting atmospheric states, and their application to global ocean dynamics offers similar promise. Here, we adapt the GraphCast architecture into a dedicated ocean-only emulator, driven by prescribed atmospheric conditions, for medium-range predictions. The emulator is trained on NOAA's UFS-Replay dataset. Using a 24 hour time step, single initial condition, and without using autoregressive training, we produce an emulator that provides skillful forecasts for 10-15 day lead times. We further demonstrate the use of Mahalanobis distance as loss that improves the forecast skill compared to the Mean Squared Error loss by explicitly accounting for the correlations between tendencies of the target variables. Using spatial correlation analysis of the forecasted fields, we also show that the proposed correlation-aware loss acts as a statistical-dynamical regularizer for the slow, correlated dynamics of the global oceans, offering a better background forecast for downstream tasks like data assimilation.

Users typically interact with and evaluate language models via single outputs, but each output is just one sample from a broad distribution of possible completions. This interaction hides distributional structure such as modes, uncommon edge cases, and sensitivity to small prompt changes, leading users to over-generalize from anecdotes when iterating on prompts for open-ended tasks. Informed by a formative study with researchers who use LMs (n=13) examining when stochasticity matters in practice, how they reason about distributions over language, and where current workflows break down, we introduce GROVE. GROVE is an interactive visualization that represents multiple LM generations as overlapping paths through a text graph, revealing shared structure, branching points, and clusters while preserving access to raw outputs. We evaluate across three crowdsourced user studies (N=47, 44, and 40 participants) targeting complementary distributional tasks. Our results support a hybrid workflow: graph summaries improve structural judgments such as assessing diversity, while direct output inspection remains stronger for detail-oriented questions.

Cross-lingual transfer in NLP is often hindered by the ``script barrier'' where differences in writing systems inhibit transfer learning between languages. Transliteration, the process of converting the script, has emerged as a powerful technique to bridge this gap by increasing lexical overlap. This paper provides a comprehensive survey of the application of transliteration in cross-lingual NLP. We present a taxonomy of key motivations to utilize transliterations in language models, and provide an overview of different approaches of incorporating transliterations as input. We analyze the evolution and effectiveness of these methods, discussing the critical trade-offs involved, and contextualize their need in modern LLMs. The review explores various settings that show how transliteration is beneficial, including handling code-mixed text, leveraging language family relatedness, and pragmatic gains in inference efficiency. Based on this analysis, we provide concrete recommendations for researchers on selecting and implementing the most appropriate transliteration strategy based on their specific language, task, and resource constraints.

Agentic security systems increasingly audit live targets with tool-using LLMs, but prior systems fix a single coordination topology, leaving unclear when additional agents help and when they only add cost. We treat topology choice as an empirical systems question. We introduce a controlled benchmark of 20 interactive targets (10 web/API and 10 binary), each exposing one endpoint-reachable ground-truth vulnerability, evaluated in whitebox and blackbox modes. The core study executes 600 runs over five architecture families, three model families, and both access modes, with a separate 60-run long-context pilot reported only in the appendix. On the completed core benchmark, detection-any reaches 58.0% and validated detection reaches 49.8%. MAS-Indep attains the highest validated detection rate (64.2%), while SAS is the strongest efficiency baseline at $0.058 per validated finding. Whitebox materially outperforms blackbox (67.0% vs. 32.7% validated detection), and web materially outperforms binary (74.3% vs. 25.3%). Bootstrap confidence intervals and paired target-level deltas show that the dominant effects are observability and domain, while some leading whitebox topologies remain statistically close. The main result is a non-monotonic cost-quality frontier: broader coordination can improve coverage, but it does not dominate once latency, token cost, and exploit-validation difficulty are taken into account.

Today, machine learning is widely applied in sensitive, security-related, and financially lucrative applications. Model extraction attacks undermine current business models where a model owner sells model access, e.g., via MLaaS APIs. Additionally, stolen models can enable powerful white-box attacks, facilitating privacy attacks on sensitive training data, and model evasion. In this paper, we focus on Decision Trees (DT), which are widely deployed in practice. Existing black-box extraction attacks for DTs are either query-intensive, make strong assumptions about the DT structure, or rely on rich API information. To limit attacks to the black-box setting, CPU vendors introduced Trusted Execution Environments (TEE) that use hardware-mechanisms to isolate workloads from external parties, e.g., MLaaS providers. We introduce TrEEStealer, a high-fidelity extraction attack for stealing TEE-protected DTs. TrEEStealer exploits TEE-specific side-channels to steal DTs efficiently and without strong assumptions about the API output or DT structure. The extraction efficacy stems from a novel algorithm that maximizes the information derived from each query by coupling Control-Flow Information (CFI) with passive information tracking. We use two primitives to acquire CFI: for AMD SEV, we follow previous work using the SEV-Step framework and performance counters. For Intel SGX, we reproduce prior findings on current Xeon 6 CPUs and construct a new primitive to efficiently extract the branch history of inference runs through the Branch-History-Register. We found corresponding vulnerabilities in three popular libraries: OpenCV, mlpack, and emlearn. We show that TrEEStealer achieves superior efficiency and extraction fidelity compared to prior attacks. Our work establishes a new state-of-the-art for DT extraction and confirms that TEEs fail to protect against control-flow leakage.

Earth observation foundation models encode land surface information into dense embedding vectors, yet the geometric structure of these representations and its implications for downstream reasoning remain underexplored. We characterize the manifold geometry of Google AlphaEarth's 64-dimensional embeddings across 12.1 million Continental United States samples (2017--2023) and develop an agentic system that leverages this geometric understanding for environmental reasoning. The manifold is non-Euclidean: effective dimensionality is 13.3 (participation ratio) from 64 raw dimensions, with local intrinsic dimensionality of approximately 10. Tangent spaces rotate substantially, with 84\% of locations exceeding 60\textdegree{} and local-global alignment (mean$|\cosθ| = 0.17$) approaching the random baseline of 0.125. Supervised linear probes indicate that concept directions rotate across the manifold, and compositional vector arithmetic using both PCA-derived and probe-derived directions yields poor precision. Retrieval instead produces physically coherent results, with local geometry predicting retrieval coherence ($R^2 = 0.32$). Building on this characterization, we introduce an agentic system with nine specialized tools that decomposes environmental queries into reasoning chains over a FAISS-indexed embedding database. A five-condition ablation (120 queries, three complexity tiers) shows that embedding retrieval dominates response quality ($μ= 3.79 \pm 0.90$ vs.\ $3.03 \pm 0.77$ parametric-only; scale 1--5), with peak performance on multi-step comparisons ($μ= 4.28 \pm 0.43$). A cross-model benchmark show that geometric tools reduce Sonnet 4.5's score by 0.12 points but improve Opus 4.6's by 0.07, with Opus achieving higher geometric grounding (3.38 vs.\ 2.64), suggesting that the value of geometric characterization scales with the reasoning capability of the consuming model.

Probing has shown that language model representations encode rich linguistic information, but it remains unclear whether they also capture cognitive signals about human processing. In this work, we probe language model representations for human reading times. Using regularized linear regression on two eye-tracking corpora spanning five languages (English, Greek, Hebrew, Russian, and Turkish), we compare the representations from every model layer against scalar predictors -- surprisal, information value, and logit-lens surprisal. We find that the representations from early layers outperform surprisal in predicting early-pass measures such as first fixation and gaze duration. The concentration of predictive power in the early layers suggests that human-like processing signatures are captured by low-level structural or lexical representations, pointing to a functional alignment between model depth and the temporal stages of human reading. In contrast, for late-pass measures such as total reading time, scalar surprisal remains superior, despite its being a much more compressed representation. We also observe performance gains when using both surprisal and early-layer representations. Overall, we find that the best-performing predictor varies strongly depending on the language and eye-tracking measure.

Local prediction-error-based curiosity rewards focus on the current transition without considering the world model's cumulative prediction error across all visited transitions. We introduce Curiosity-Critic, which grounds its intrinsic reward in the improvement of this cumulative objective, and show that it reduces to a tractable per-step form: the difference between the current prediction error and the asymptotic error baseline of the current state transition. We estimate this baseline online with a learned critic co-trained alongside the world model; regressing a single scalar, the critic converges well before the world model saturates, redirecting exploration toward learnable transitions without oracle knowledge of the noise floor. The reward is higher for learnable transitions and collapses toward the baseline for stochastic ones, effectively separating epistemic (reducible) from aleatoric (irreducible) prediction error online. Prior prediction-error curiosity formulations, from Schmidhuber (1991) to learned-feature-space variants, emerge as special cases corresponding to specific approximations of this baseline. Experiments on a stochastic grid world show that Curiosity-Critic outperforms prediction-error and visitation-count baselines in convergence speed and final world model accuracy.

Indistinguishability properties such as differential privacy bounds or low empirically measured membership inference are widely treated as proxies to show a model is sufficiently protected against broader memorization risks. However, we show that indistinguishability properties are neither sufficient nor necessary for preventing data extraction in LLM APIs. We formalize a privacy-game separation between extraction and indistinguishability-based privacy, showing that indistinguishability and inextractability are incomparable: upper-bounding distinguishability does not upper-bound extractability. To address this gap, we introduce $(l, b)$-inextractability as a definition that requires at least $2^b$ expected queries for any black-box adversary to induce the LLM API to emit a protected $l$-gram substring. We instantiate this via a worst-case extraction game and derive a rank-based extraction risk upper bound for targeted exact extraction, as well as extensions to cover untargeted and approximate extraction. The resulting estimator captures the extraction risk over multiple attack trials and prefix adaptations. We show that it can provide a tight and efficient estimation for standard greedy extraction and an upper bound on the probabilistic extraction risk given any decoding configuration. We empirically evaluate extractability across different models, clarifying its connection to distinguishability, demonstrating its advantage over existing extraction risk estimators, and providing actionable mitigation guidelines across model training, API access, and decoding configurations in LLM API deployment. Our code is publicly available at: https://github.com/Emory-AIMS/Inextractability.

Modern sequence modeling is dominated by two families: Transformers, whose self-attention can access arbitrary elements of the visible sequence, and structured state-space models, which propagate information through an explicit recurrent state. These mechanisms face different limitations on long contexts: when attention is diffuse, the influence of individual tokens is diluted across the effective support, while recurrent state propagation can lose long-range sensitivity unless information is actively preserved. As a result, both mechanisms face challenges in preserving and selectively retrieving information over long contexts. We propose Sessa, a decoder that places attention inside a recurrent feedback path. This creates many attention-based paths through which past tokens can influence future states, rather than relying on a single attention read or a single recurrent chain. We prove that, under explicit assumptions and matched regimes, Sessa admits power-law memory tails $O(\ell^{-β})$ for $0 < β< 1$, with slower decay than in the corresponding Transformer and Mamba-style baselines. We further give an explicit construction that achieves this power-law rate. Under the same assumptions, Sessa is the only model class among those considered that realizes flexible selective retrieval, including profiles whose influence does not decay with distance. Consistent with this theoretical advantage, across matched experiments, Sessa achieves the strongest performance on long-context benchmarks while remaining competitive with Transformer and Mamba-style baselines on short-context language modeling.

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 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 almost 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.

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.

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 UDM-GRPO, 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. UDM-GRPO 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 https://github.com/Yovecent/UDM-GRPO.

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.

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.

Conformal prediction provides finite-sample, distribution-free coverage under exchangeability, but standard constructions may lack robustness in the presence of outliers or heavy tails. We propose a robust conformal method based on a non-conformity score defined as the half-mass radius around a point, equivalently the distance to its $(\lfloor n/2\rfloor+1)$-nearest neighbour. We show that the resulting conformal regions are marginally valid for any sample size and converge in probability to a robust population central set defined through a distance-to-a-measure functional. Under mild regularity conditions, we establish exponential concentration and tail bounds that quantify the deviation between the empirical conformal region and its population counterpart. These results provide a probabilistic justification for using robust geometric scores in conformal prediction, even for heavy-tailed or multi-modal distributions.

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. 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

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.

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.

Existing jamming attacks on Retrieval-Augmented Generation (RAG) systems typically induce explicit refusals or denial-of-service behaviors, which are conspicuous and easy to detect. In this work, we formalize a subtler availability threat, termed soft failure, which degrades system utility by inducing fluent and coherent yet non-informative responses rather than overt failures. We propose Deceptive Evolutionary Jamming Attack (DEJA), an automated black-box attack framework that generates adversarial documents to trigger such soft failures by exploiting safety-aligned behaviors of large language models. DEJA employs an evolutionary optimization process guided by a fine-grained Answer Utility Score (AUS), computed via an LLM-based evaluator, to systematically degrade the certainty of answers while maintaining high retrieval success. Extensive experiments across multiple RAG configurations and benchmark datasets show that DEJA consistently drives responses toward low-utility soft failures, achieving SASR above 79\% while keeping hard-failure rates below 15\%, significantly outperforming prior attacks. The resulting adversarial documents exhibit high stealth, evading perplexity-based detection and resisting query paraphrasing, and transfer across model families to proprietary systems without retargeting.

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.

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.

Large Language Models (LLMs) are increasingly used in education, yet their default helpfulness often conflicts with pedagogical principles. Prior work evaluates pedagogical quality via answer leakage-the disclosure of complete solutions instead of scaffolding-but typically assumes well-intentioned learners, leaving tutor robustness under student misuse largely unexplored. In this paper, we study scenarios where students behave adversarially and aim to obtain the correct answer from the tutor. We evaluate a broad set of LLM-based tutor models, including different model families, pedagogically aligned models, and a multi-agent design, under a range of adversarial student attacks. We adapt six groups of adversarial and persuasive techniques to the educational setting and use them to probe how likely a tutor is to reveal the final answer. We evaluate answer leakage robustness using different types of in-context adversarial student agents, finding that they often fail to carry out effective attacks. We therefore introduce an adversarial student agent that we fine-tune to jailbreak LLM-based tutors, which we propose as the core of a standardized benchmark for evaluating tutor robustness. Finally, we present simple but effective defense strategies that reduce answer leakage and strengthen the robustness of LLM-based tutors in adversarial scenarios.