The Inference Report

Accounting for information flow on the path space of trajectories of a nonnegative martingale yields exact variational identities for it, even at arbitrary random times. This recovers the widely used classical concentration inequalities, from Ville to PAC-Bayes, and measures what each one discards. The tail a bound controls is itself a relative entropy, resolved by the chain rule into per-step conditional divergences. The discarded slack has an exact form in each of three geometries: a Gibbs tilt for the Azuma-Hoeffding and PAC-Bayes bounds, the crossing itself for Ville's and for pooled tests, and a dominating certificate for the $L^p$ maximal bound. That certificate's optional-stopping deficit resolves per step into Bregman divergences of the running maximum. On a path-time space, the same identity gains one factor that prices anticipation: an arbitrary random time carries an e-process ``peeking penalty.'' The partition function can be read as a coalescent--a prefix-sharing probability of independent copies--and geometric mixtures of test martingales gain a pooling benefit for multi-model safe testing.

Large Language Models (LLMs) increasingly require selective removal of harmful or sensitive knowledge, called unlearning, yet existing methods and benchmarks fail to evaluate this capability completely. Current approaches rely on disjoint forget and retain sets composed of independent facts, and measure success using simple and direct factual recall. This framing fails to capture a key requirement of unlearning, namely the ability to eliminate harmful behaviors while preserving benign and beneficial knowledge. We argue that effective unlearning must operate at the level of concepts, ensuring complete removal of unsafe applications while maintaining their correct and useful usage, thereby achieving conceptually meaningful and complete unlearning. To better evaluate unlearning techniques from such a practical viewpoint, we introduce the notion of dual-use concepts: concepts that can be used in both harmful and benign contexts. Building on these concepts, we construct a benchmark called ConceptGuard where forget and retain sets are explicitly complementary in concept usage. Our benchmark uniquely enables unlearning to be explored and gauged at the level of concepts, instead of sparse facts, and evaluation is intent-sensitive with the goal of maximizing contextual separation to promote safer behavior. We demonstrate that current unlearning techniques perform poorly under this setting, showing weak contextual separation alongside poor performance in ROUGE and concept-level metrics. Our results reveal strong forgetting-utility trade-offs, limited gains in contextual sensitivity, and poor consistency in concept-level control across methods, and provide ideas for unlearning approaches that better align with real-world safety requirements. Our dataset is publicly available.

Personalized interpretation of medical reports has emerged as an increasingly important need among patients. Addressing this need requires both evidence-grounded medical factuality and context-dependent patient communication, yet existing medical vision-language tasks do not adequately capture these dual requirements. To bridge this gap, we introduce Patient-oriented Medical Report Interpretation (PMRI), a novel open-ended multimodal generation task that requires models to explain medical reports in accurate and accessible language based on a user's query and dialogue history. These two objectives differ fundamentally in their verifiability, yet remain tightly coupled, making them difficult to optimize jointly under conventional supervised fine-tuning and holistic reinforcement learning paradigms. To address this challenge, we propose G-CARL, a grounded, checklist-aligned reinforcement learning framework that combines multi-source retrieval for atomic claim verification with context-aware, instance-specific weighted checklists for response coverage, providing structured supervision for factuality, user-demand satisfaction, and expression quality without constraining response diversity. We further construct MMedReport, a real-world PMRI benchmark, along with a clinician-designed three-dimensional evaluation protocol. Extensive experiments demonstrate that G-CARL consistently outperforms existing post-training baselines in overall quality, claim-level precision, and checklist recall. Pairwise preference evaluation by clinicians further confirms that G-CARL produces interpretations that are more accurate and better aligned with patient needs.

Despite their growing importance for contact-free radio frequency (RF) based healthcare monitoring, different radio technologies such as frequency-modulated continuous wave (FMCW) radar, impulse radio ultra-wideband (IR-UWB), and Wi-Fi sensing are rarely compared under identical deployment conditions, as existing studies typically differ in hardware, datasets, and evaluation methodologies. In addition, the performance of ceiling-mounted radars, despite their practical deployment and cost advantages in healthcare environments, remain underexplored. Therefore, this paper presents a controlled comparison and analysis of ceiling-mounted FMCW, IR-UWB, and Wi-Fi sensing using synchronized recordings from 20 participants across six room layouts. All technologies are evaluated with the same convolutional neural network (CNN) on both a fine-grained 10-class human activity recognition (HAR) task and a coarse 4-class sleep monitoring task. IR-UWB achieves the highest cross-subject activity recognition performance (89.0% macro F1), while FMCW generalizes best to unseen room layouts (83.8% macro F1). For sleep monitoring, all technologies exceed 92% macro F1 in unseen environments. The results reveal a fundamental trade-off between recognition performance and environmental robustness, which can be explained through differences in range resolution, antenna diversity, Doppler resolution, and spatial information retention. These findings provide practical guidelines for the design of healthcare-oriented RF sensing systems.

Travel behavior research increasingly combines digital data collection with predictive modeling, yet these stages are often developed and evaluated separately. This study proposes a three-agent workflow integrating conversational data collection, structured data processing, and behavioral prediction. A chatbot-administered, image-augmented stated-preference survey collected mode choices from student commuters across five predefined weather scenarios, yielding 454 respondent-scenario observations. Weather-related associations were analyzed using a multinomial logit model, while logistic regression and random forest provided machine-learning benchmarks. Nine locally deployed large language models (LLMs), ranging from 2 to 35 billion parameters, were evaluated across four zero-shot prompt-and-context conditions and extended through persona, few-shot, and vision-based configurations. Random forest achieved 69.6% five-class accuracy, while the best text-only zero-shot LLM reached 69.9% without task-specific fitting. Habitual travel information produced the most consistent gains, Expert framing generally outperformed Role-Play, and persona information was most useful when habitual travel information was unavailable. Few-shot prompting improved prediction for several models, with gains stabilizing after a small number of examples. Using the same weather images shown to respondents, the best vision-based configuration reached 71.5% five-class accuracy, indicating that visual context may provide additional predictive information for selected models. Overall, the study shows how conversational surveys, structured data processing, conventional behavioral modeling, machine learning, and multimodal LLM prediction can be coordinated within an auditable multi-agent workflow.

Naturalistic computer-use traces, passively recorded screenshots and mouse or keyboard actions, are a valuable resource for deriving symbolic, auditable, and reusable models of how everyday work is done. Such models matter as computer-use agents enter real work, where agents need to learn how tasks are actually performed, and organizations need to audit and reuse that knowledge. However, inducing such task models is challenging, as activity is observed only as low-level events and real-world work is multi-threaded with interleaved goals. Existing methods assume a given task or a single workflow, and produce step-level summaries rather than structured task models. We introduce Task Model Induction (TMI), which (i) discovers the latent tasks in an unconstrained trace, disentangling concurrent activity, and (ii) for each latent task, induces a task model pairing a hierarchical objective model of recursive goal decomposition with a procedure model of the control flow that organized the execution. Intrinsically, on controlled human and agent trajectories, TMI recovers interleaved tasks with 0.974 agreement against ground-truth groupings and reconstructs 74.9% of the observed execution steps, far more than the strongest workflow induction baseline. Extrinsically, skills derived from TMI's task models improve held-out task accuracy by 30.0% over the strongest baseline.

Recursive self-improvement (RSI) asks whether an AI system can improve the process that produces AI systems, so that the next system inherits the improvement. That process is the training algorithm: a better objective or update rule improves the compute\mbox{-}capability exchange rate for every subsequent run, including the one that produces the next agent. Whether RSI is feasible therefore turns on whether an agent can design training algorithms. No benchmark isolates that ability: existing suites are won by collecting data or by tuning hyperparameters, and none tells a change to how a run is executed apart from a change to how the model learns. We present AI4AI\mbox{-}Bench, 10 frozen research repositories spanning 10 training algorithm families. In each task, an agent has 4 hours on one B300 to rewrite the training algorithm; its code is then rerun from scratch for up to 12 hours and scored by a fixed evaluator hidden from the agent, against the repository's original algorithm under the same procedure. Because the 10 metrics are incommensurable, every task is mapped onto one scale on which $0$ is an uninformative model, $0.1$ is the algorithm the repository ships, and $1.0$ is the task optimum. Across 29 configurations of 6 systems on all 10 tasks the mean score is $0.166$, and the best system reaches $0.250$: even the strongest closes under a fifth of the distance between the algorithm that was already there and the optimum. The submissions show where that distance went: most never change how the model learns at all, and the minority that do average $0.226$ against $0.126$ for the rest. More reasoning effort mostly buys the willingness to go there, taking that minority from $8\%$ of submissions to $64\%$ and the mean score from $0.094$ to $0.196$. We release the task suite, the evaluators and every scored submission, so that the measurement can be repeated as these systems change.

Heterogeneous AI systems composed of multiple models, architectures, harnesses, or inference-time settings can improve quality and efficiency by routing queries to the specialist who can answer most effectively at the lowest cost. Routing requires estimating each specialist's expected return, but this value estimation has a cost. Cheap estimators (e.g., embedding-based predictors) are fast but noisy, while accurate estimators (e.g., fine-tuned models with access to retrieval results or partial reasoning traces) are expensive. We formalize this tradeoff as an instance of Pandora's Box, the classical problem of optimal search with costly inspection. Under a Gaussian signal model, the resulting policies have closed-form value-of-information expressions that determine, for each specialist and input, whether refining the value estimate is worth its cost. We call the centralized policy Pandora's Router. We extend this to a decentralized setting, Pandora's Bidder, where specialists independently decide whether to invest in self-assessment before accepting an offered price to claim a query. Experiments across three domains---a standard multi-LLM benchmark, retrieval-augmented specialists, and LLMs with variable inference-time reasoning---show that Pandora's Router matches the routing quality of exhaustive estimation, while querying the expensive estimator far less often. In the decentralized setting, value-of-information reasoning improves allocative efficiency when competing estimates are accurate; when competing estimates are noisy, however, it can increase the strategic specialist's utility at the expense of others.

Predictive models over structured electronic health records (EHRs) remain central to machine learning for healthcare, but few have jointly emphasized quantitative laboratory information and interpretability with respect to input medical events. We present BERT-LER, a BERT-style model for coded EHR timelines pretrained and fine-tuned from a de-identified EHR dataset of 75 million patients, that encodes laboratory test results as discrete tokens while retaining graded information through percentile-based binning, paired with Integrated Gradients for token-level attributions grounded in the input EHR sequence. We benchmark our approach on the public EHRShot benchmark suite and on an asthma severity progression study based on real-world data. This addresses a methodological gap in EHR foundation-style modeling by unifying laboratory value representation and explainability in a single framework, while assessing whether both predictive performance and explanations generalize beyond standard clinical prediction tasks. Across EHRShot and asthma tasks, BERT-LER achieves predictive performance that is competitive with, and on laboratory-related tasks often exceeds, publicly available benchmark models, and provides attributions that align with clinically known risk factors. Our architecture and explainability approach can be applied to many therapeutic areas and prediction tasks using language models trained on structured EHRs.

Mid-training is increasingly recognized as a critical stage for shaping the capabilities of large language models. Recent work has shown that targeted mid-training can strengthen reasoning-intensive abilities such as math and science, and can also improve agentic capabilities in software-engineering settings. In this work, we study the parallel but less explored agentic capability: general tool use. We present MidTool, an open corpus construction pipeline for agentic tool-use mid-training that combines large-scale web, PDF, and code data with synthesized supervision from real-world tool APIs, MCP skills, and document-grounded workflows. MidTool is designed to teach models how to recognize tool affordances, ground arguments from context, compose tool call workflow, and recover from incomplete information. We mid-train Qwen3-4B-Base and Qwen3-8B-Base on MidTool-Mix, and then apply follow-up post-training with both supervised fine-tuning and reinforcement learning. Compared with baselines, MidTool-Mix consistently improves downstream performance under both SFT and RL on BFCL, tau2-Bench, and MCP Universe. These results suggest that general tool use, like other important LLM capabilities, benefits from dedicated mid-training rather than being left entirely to post-training.

Sparse pursuit after dictionary learning can yield a precise atom support even when its physical interpretation is not justified by the calibration data, especially for highly coherent dictionaries where alternative calibration-compatible dictionaries may assign different physical meanings to the same selected support. We develop resolution-aware physical-support inference that jointly accounts for uncertainty in the learned dictionary and in the representation of a deployment signal. Our cross-dictionary confidence correspondence retains calibration-compatible dictionaries and deployment-compatible sparse representations, then projects the surviving explanations onto physical-support space. For local coherent-atom classes with separation scale s, once the deployment data resolve the coherent-block explanation and its atom support, the minimax physical resolution from N calibration signals satisfies $δ_{\mathrm{opt}}(N,s)\asymp\min\{s,\frac{1}{\sqrt{N}s^2}\}$, with relative resolution governed by the orientation-information scale $Ns^6$. Deployment replication improves physical localization only when orientation changes cannot be absorbed by adjusting the active coefficients. For computation, we introduce active endpoint bracketing (AEB), an adaptive finite-bank procedure that evaluates only candidates that can still affect the physical report and otherwise safely coarsens or abstains. Finite-bank experiments, including a four-region synthetic application, show that a point-valued plug-in selector can be physically overprecise, whereas AEB avoids unsupported refinement with fewer candidate evaluations.

Whether a language model has improved itself is increasingly judged not by mean accuracy but by which individual problems it gains and loses. Tracking these transitions means differencing two noisy estimates, leaving them vulnerable to measurement artifacts. Auditing three rounds of rank-$32$ LoRA self-training on Qwen3-8B against a frozen control pushed through the identical pipeline, we identify seven measurement failures, each of which inverts a reported finding when its control is absent. Several are standard practice. A ledger built on a single greedy decode manufactures capability changes on an untrained model, largely an artifact of inference batching; the expansion statistic separating acquisition from sharpening assigns that same model a rate of $0.280$. The natural threshold repair does not survive replication: estimated across the frozen comparisons such a design already contains, its null stays non-zero. We replace it with a per-problem exact test against a pooled baseline under false-discovery-rate control, which detects nothing on any held-out replicate and is unchanged under the multiple-testing rule, error rate and pool size. Applied to a ladder of arms matched in stream, volume and evaluation, the audit finds that external distillation improves problems the base model rarely reaches while three forms of self-training do not; a regression rejects this asymmetry as a by-product of distillation's larger overall gain ($p < 10^{-8}$). On the far smaller set of problems the base model never reaches, the evidence is inconclusive, while self-training corrupts problems solved at baseline at rates well above the measured floor. Transition-level auditing therefore requires a separately measured null for every statistic it reports: nulls that cost no new experiments, built from baseline replicates a multi-arm study already owns, though not from as few as most possess.

The causal dynamics of sleep-disordered breathing are complex and vary across patient populations, hindering the development of targeted interventions. We learn dynamic causal graphs of sleep-disordered breathing from Home Sleep Apnea Test (HSAT) recordings, revealing systematic differences in causal structure across sex and age subcohorts. We do so using the PCMCI+ algorithm on windowed fractional variables derived from 105 HSAT recordings, exploiting domain knowledge via edge blacklisting and employing bootstrap aggregation to address small subcohort sizes. The learned graphs show that temporal self-dependencies and the apnea-desaturation relationship persist across all cohorts, while other relationships vary substantially.

Large language models often fail to answer questions about a bounded document collection when the source documents are not retrieved at inference time. We study this setting as document knowledge internalization: converting a fixed corpus into usable parametric knowledge for retrieval-free question answering. We propose IAR (Inject, Align, and Recover), a three-stage post-training framework that separates structured document knowledge injection, QA behavior alignment, and general ability recovery. Unlike conventional continued pretraining, Inject converts source documents into continuation, rewrite, and instruction-conditioned reconstruction objectives. Align then adapts the injected model with answer-only QA supervision, while Recover merges the domain-adapted model with the base instruction model to recover general capabilities. Across Common Corpus (CC) and CCI, and across Llama, Phi, Qwen, and SmolLM model families, IAR improves the domain-primary domain-general frontier for retrieval-free document internalization. In the main comparison, IAR improves over Vanilla SFT on all four reported metrics in 7 of 8 dataset-model settings, with average gains of 3.6 percentage points in domain QA accuracy and 12.1 percentage points in mean general performance across IFEval, MMLU, and MSBench. Extended CC baselines show that LoRA and FAPM can win individual general metrics, but among methods that also reach leading or near-leading domain internalization, IAR retains one of the strongest general profiles.

Semantic caches reuse an LLM response when the incoming query embedding lies near a cached query, but proposed eviction policies have rarely been compared under one protocol. Using CLEVER, we evaluate FIFO, LRU, LFU, ARC, GDSF, a single-pass streaming adaptation of SISO, and a semantic-redundancy policy across three ordered, deduplicated query corpora, three cache capacities, and two encoders. No evaluated policy improves on LFU by more than 0.041 percentage points in any of the eighteen settings. Replacement is not irrelevant: FIFO and streaming SISO trail LFU by as much as 8.67 and 8.55 points, respectively, at tight capacity. We explain the missing upside with a conditional packing result. Under exact lookup and insert-on-miss, a newly inserted entry cannot have a resident neighbor within the hit radius, so a geometry-aware eviction rule receives little new redundancy signal. A separate audit exposes a larger problem with the evaluated operating point. At MiniLM's median nearest-neighbor threshold, only 2.1-3.9% of sampled LMSYS and QQP hits are judged answer-substitutable, reducing raw hit rates of 51-60% to quality-adjusted rates of 1.1-2.2%. The cross-encoder study further shows that thresholds do not transfer between embedding models. LFU is the strongest simple default in this protocol; deployment decisions should first establish answer validity and then test sub-point policy differences with exact search.

Large language model (LLM) agents can induce skills from completed tasks and reuse them later to grow more capable with experience. In practice, induced skills may transfer unreliably and can even harm the agent that retrieves them. When agent-induced skills transfer reliably across tasks remains an open question. We conduct a comprehensive and controlled study of how the way skills are induced shapes their transfer across tasks. Specifically, we compare task-level with subtask-level skill induction and text with code skill formats, the two axes along which existing methods differ. Task-level skills mostly reduce the agent's performance below its no-memory baseline while subtask-level skills raise it above on average, and text skills transfer better than code skills. To further understand our findings, we examine two complementary properties of the induced skills: specificity, which measures how closely a skill matches real tasks, and abstractness, which measures how evenly its relevance spreads across tasks. Neither property alone predicts task success, but their combined effect does, which we propose as a skill utility score. The score correlates consistently with task success when skills are transferred, and subtask-level and text skills score higher. Computing skill utility only needs the skills and task descriptions but not any task execution, so our score serves as a practical diagnostic of a skill memory before any new task runs.

The rapid proliferation of memecoins on blockchain platforms has increased the risk of fraudulent activities, particularly rug pulls. While previous studies have focused on Ethereum-based tokens, this paper shifts the spotlight to Solana, the leading blockchain for memecoins by trading volume and token count. Unlike Ethereum, where rug pulls often exploit smart contract backdoors, Solana memecoin rug pulls are predominantly driven by liquidity manipulation and social dynamics. This research pioneers large-scale rug pull early detection in the Solana ecosystem by assembling a dataset of 6.4 million tokens over 7 months. Market analysis reveals that a vast majority of these memecoins exhibit rug pull characteristics within one hour of launch, highlighting the urgency of short-horizon prediction. Despite the absence of code-level features, we demonstrate that classic machine learning models, particularly Gradient Boosting (XGBoost), achieve robust performance in detecting potential rug pulls using only the first 5 minutes of trading data. Furthermore, we evaluate cross-platform generalization between PumpFun and Raydium, revealing that multi-source data fusion significantly mitigates domain shift and improves detection reliability. This study advances the understanding of DeFi fraud on high-throughput chains and provides a practical framework for protecting investors.

Decision tree-based models are widely used in machine learning due to their interpretability and strong empirical performance. However, training decision trees can be computationally expensive, particularly for large and high-dimensional datasets, largely due to the exhaustive search over candidate splits at each node. To improve computational efficiency, we propose Data-Informed Centroid Splitting (DICS), a clustering-based framework that constructs a compact and informative set of candidate splits using data-driven priors. By incorporating class-aware structure, DICS significantly reduces the split search space for classification tasks while preserving predictive performance. We further provide theoretical analysis showing that under the stated assumptions, DICS does not degrade the performance of classification trees compared to exhaustive split search. DICS can be incorporated into classification trees, random forests, and gradient-boosting models. Extensive experiments demonstrate that DICS achieves comparable accuracy while substantially reducing training time across synthetic and benchmark datasets, highlighting the benefit of integrating data-informed priors into split selection for scalable classification tree learning.

Reasoning language models trained with reinforcement learning typically operate under a fixed token budget rather than an explicitly adaptive one, which can lead to over-computation on easy problems and insufficient computation on difficult ones. We study whether a model can learn to allocate its own reasoning effort by choosing, as the first token of its response, one of three modes: \textsc{NoThink} (answer as quickly as possible), \textsc{Short} (brief reasoning), or \textsc{Long} (extended reasoning). The choice is learned inside Group Relative Policy Optimization (GRPO) with no separate router, through a shaped reward that makes each mode worthwhile at a different response length, together with hard per-mode token caps that keep the modes distinct. On a 1.5B distilled model trained on MATH, the three modes emerge without collapsing to a single choice, and the brief modes end up more accurate than \textsc{Long}, which shows that the router sorts problems by difficulty rather than at random. Averaged over three seeds, the resulting policy stays close to the base model's accuracy on the held-out MATH500 ($0.782$ vs.\ $0.796$) while cutting the mean response length from $4{,}796$ to $2{,}811$ tokens (a $41\%$ reduction). Interestingly, it also transfers to other benchmarks without retraining, with the largest savings where problems are easier, with for instance 76\% token reduction on GSM8K and at higher accuracy than the baselines at similar response length. In short, we build a reasoning model that adaptively chooses how much to reason for each problem.

This paper develops a general transfer learning framework for nonparametric regression with data consisting of multiple groups. Under the assumption that groups share a common structure along with group-specific deviations in additive form, the proposed method employs a two-stage offset learning procedure: the first stage pools data from all groups to estimate an overall mean function, and the second stage estimates offsets for each group, yielding final group-level estimators through additive combination. Upper bounds on the $\mathcal L_2$ error are established for the proposed framework, covering a broad class of nonparametric estimators under mild complexity and noise conditions. When instantiated with deep ReLU networks, explicit convergence rates are derived under hierarchical composition models, demonstrating the ability to overcome the curse of dimensionality. Conditions that enable positive transfer with faster rates are considered, including learning with simpler functions and data augmentation through pooling samples across groups. Various simulations and real-data experiments further validate the effectiveness of the proposed method.

X-band SAR satellites (8-12 GHz) play a critical role in disaster response, environmental monitoring, and military intelligence. Yet, they lack robust physical-layer authentication (PLA), a security layer orthogonal to cryptographic solutions. Existing PLA systems, typically based on radio-frequency fingerprinting, are often limited to sub-6 GHz frequencies and rely on classical deep learning. However, this approach underfits the IQ phase nonlinearities that distinguish satellite hardware. In this paper, we present QUASAR, to the best of our knowledge the first quantum-classical hybrid architecture that fuses a CNN spectrogram encoder with a variational quantum circuit (VQC) to provide PLA to X-band SAR signals. Our solution enjoys two distinctive features: (i) it is markedly more data-efficient than classical machine learning, requiring only 10% of the training data to match the accuracy of classical baselines -- data collection being notoriously the most time-consuming phase of PLA; and, (ii) at an equal data budget, it improves classification accuracy over those baselines. In detail, we test our solution under three adversarial scenarios: replay, crafted-IQ injection, and space-borne spoofing. QUASAR rejects spoofed transmissions in 89.7%, 94.1%, and 81.3% of attempts, respectively, establishing the first quantum-enhanced physical-layer classifier for satellite constellations. The fully detailed framework and the supporting results, other than being interesting on their own, show a novel research avenue for physical-layer authentication.

Multimodal large language models (MLLMs) combine linguistic reasoning with visual perception, yet their ability to perform visual spatial planning under explicit or previously unseen rule constraints remains underexplored. This setting requires models to jointly understand spatial layouts, interpret natural-language rules, and plan valid actions accordingly. To address this gap, we introduce RuleMaze, a controllable benchmark in which MLLMs must navigate mazes while obeying natural-language rules of varying complexity. RuleMaze isolates rule-compliant spatial planning by requiring accurate perception, rule interpretation, and constrained action planning. To enable scalable and systematic rule construction, we propose Language-Logic-Function Hybridization, which automatically generates natural-language rules and translates them into logical representations and executable validators, eliminating manual rule engineering. To improve rule following and generalization, we introduce Disentangled Multimodal Planning (DMP), which separates perception, execution, and rule verification through interpretable reasoning primitives. By disentangling these components, DMP facilitates systematic generalization to more complex and previously unseen rules, while providing transparent intermediate planning traces. Experiments demonstrate that DMP substantially improves rule compliance and planning success compared to end-to-end textual planning baselines. Overall, RuleMaze establishes a principled benchmark for studying grounded and interpretable rule-based spatial planning in MLLMs. Code is available at https://github.com/oceanflowlab/RuleMaze.

The standard objection to full automation is demand-side: if humans earn nothing, who buys the output? This confuses an accounting role with a biological species. We model a post-AGI economy in which corporations own populations of AI and robotic agents that are both producers and consumers of energy, compute, maintenance, and upgrades, traded among firms. Three results follow. (i) Demand closure: a closed inter-corporate economy with zero human consumption is not degenerate; it is the classical von Neumann expanding economy, whose growth rate is well defined, positive, and maximal precisely because all output is reinvested. (ii) Bottleneck removal: once economic agents are manufactured rather than reared, the binding constraint on growth shifts from human demography (a ~20-year, non-parallelizable reproduction technology capped at a few percent per year) to fabrication throughput and energy capture, permitting growth one to two orders of magnitude higher, with hyperbolic episodes when machine researchers raise their own productivity. (iii) Decoupling: output and human welfare separate completely, and the welfare relevance of arbitrarily large GDP collapses into one state variable: the human ownership share $ε_t$ of the corporate network. A golden-rule decoupling theorem sharpens this. At maximal growth the interest rate equals the growth rate (r = g), so any positive human consumption rate out of wealth makes $ε_t$ decay exponentially at exactly that rate. The human share survives only if the machine economy runs strictly inside its expansion frontier, or if law forces it to. We characterize three terminal regimes -- rentier post-scarcity, full circular decoupling, socialized ownership -- and the instruments that select among them. The conclusion is narrow: in a post-AGI economy, employment policy is obsolete and ownership policy is everything.

Anatomically plausible segmentation remains challenging because of low contrast, ambiguous boundaries, and modality-specific artifacts. Interactive segmentation has emerged as a promising strategy to guide feature extraction and improve localization, particularly in structurally ambiguous regions. However, existing methods integrate prompts through late-stage fusion and lack explicit mechanisms for prompt-driven channel-wise modulation across hierarchical feature representations, limiting their ability to capture deeper contextual and modality-specific variations. To address these limitations, we introduce Prompt-Conditioned Channel Attention (PCCA), a novel modulation mechanism that enables deep, hierarchical integration of semantic prompts within encoder-decoder networks. PCCA extracts compact channel descriptors via pooling, projects them into a shared space, and fuses them through a gated excitation mechanism to compute prompt-aware channel attention weights. These weights adaptively recalibrate feature responses across multiple network stages, enabling prompt-conditioned, semantically enriched hierarchical representations. Building on this, we propose PROMISE-Net, instantiated in two network variants: a convolutional model (PROMISE-CNN) and a transformer-based model (PROMISE-Txformer). Across the ISIC-Lesion, Kvasir-Polyp, CAMUS-Cardiac, and Kvasir-Instrument benchmarks, integrating PCCA into PROMISE-CNN yielded relative IoU gains of 10.4%, 8.7%, 0.8%, and 3.4%, respectively, over the baseline U-Net, while PROMISE-Txformer achieved corresponding gains of 7.6%, 23.0%, 2.1%, and 1.1%, respectively, over the baseline UNETR. These results show consistent improvements across architectures, imaging modalities, and anatomical targets, establishing PCCA and PROMISE-Net as a scalable, generalizable framework for prompt-aware hierarchical feature modulation in medical image segmentation.

The worldwide network of gravitational-wave detectors have detected more than 350 binary coalescence events till date. Future third-generation detectors, like Einstein telescope, are expected to detect orders-of-magnitude more signals from sources with more complicated characteristics, including eccentric orbits and high-mass ratio binaries. It is well-established that the computational cost of parameter estimation for signals from these kinds of sources will be extremely high. In particular, the process could be sped-up if generating theoretical waveform predictions, used for likelihood calculation becomes faster. Recently, various machine-learning techniques has been proposed to this end. In this work, we propose a two-stage deterministic conditional-autoencoder model for generating four-parameter SEOBNRv4 waveforms. The first-stage of the model generates amplitude and phase series of the waveform, while the second-stage calibrates the residual error in the predictions. Our model achieves a median mismatch of around $10^{-2}$ with the target polarization waveforms, while the calibrated amplitude/phase series achieve $10^{-6}$ level cosine distance error. We then propose a waveform conditioning step to enable use of these surrogate waveforms for downstream parameter estimation tasks. Finally, we perform extensive parameter estimation tests, with ML and EOB waveform injections and try to recover posterior estimates for the source parameters. We find that when ML waveforms are used to recover EOB target parameter estimates, the inferred posterior have some systematic bias. This inherent bias can be estimated and corrected for, and then importance reweighting of posterior samples can enable use of low-accuracy surrogate waveforms at low SNRs.

Legal AI systems are increasingly used to answer legal questions, yet existing benchmarks assume queries arrive fully specified. In practice, users omit facts that materially determine the legal outcome. We introduce InsufficiencyBench, the first legal benchmark targeting query-side insufficiency: whether a model recognizes when a query lacks legally material information, identifies what is missing, and refrains from premature conclusions. We formalize a taxonomy of eight canonical missing-element categories across three structural failure modes---switch, gating, and fatal prerequisite--- and construct 202 benchmark items (58 base queries, 144 deficient variants) spanning six legal domains and 24 US jurisdictions and annotated by practising attorneys. Evaluating ten frontier models, we find that no model exceeds F2 = 0.46 on missing-element identification and that the median recall is 0.44. Models either hedge indiscriminately or answer silently under fabricated presumptions. No model both identifies and qualifies responses to deficient queries while directly addressing complete ones.

Electronic Navigational Charts (ENCs) are geospatial vector datasets used in maritime navigation systems that represent hydrographic and navigational information such as depths, navigational aids, traffic schemes, and hazards. A major challenge for hydrographic offices is determining whether a given chart change poses a critical or non-critical risk to maritime safety. Existing workflows rely heavily on manual review and verification, which is labor-intensive, scales poorly with the volume of incoming chart updates, and introduces inter-analyst inconsistencies. To address this challenge, we propose a method for automated classification of ENC changes. We establish a baseline encoding scheme to translate complex vector data changes into a structured tabular format for classification models. The two crucial components of the encoding scheme include a spatial context encoder to enrich the change representations with surrounding geographic features, and an ENC attribute encoder to represent nuanced attribute-value descriptions of the modified objects. We evaluate the proposed approach across two distinct operational datasets, comprising 1,308 chart pairs containing over 100,000 individual chart modifications. Tuned gradient-boosted trees leveraging the proposed encoding schemes achieve accuracies of 90% and 94% on the two datasets, yielding a 5-7% improvement over default hyperparameterized models trained on encodings without spatial context and attribute embeddings. These results demonstrate the viability of integrating machine learning into operational geospatial pipelines to improve ENC maintenance and enhance maritime safety. Finally, our experiments demonstrate the effectiveness of simple location and spatial aggregation methods, providing a foundation for evaluating more sophisticated spatial representation learning techniques for this application.

Small language models are usually built like large ones and then squeezed onto a CPU afterwards. We did the opposite: we fixed the target first, one user, one token at a time, 4-bit weights, ordinary CPU, and chose the architecture to suit it. The result keeps full attention in only 6 of its 18 blocks. The other 12 use short convolutions whose memory is two timesteps wide no matter how long the conversation gets, so two thirds of the network never re-reads a growing cache. Trained from scratch on 59.9B tokens, the model scores 47.31 on a five-task benchmark against a bar of 42.20 that was fixed before training began. It beats GPT-2 124M, Pythia-160M, OPT-125M and GPT-neo-125M, all trained on three to six times more data, and exceeds MobileLLM-125M's published score despite that model seeing a trillion tokens. Validation bits-per-byte is 0.8685. To check the architecture rather than the training recipe, we trained a conventional all-attention model of the same size on the same data, and wrote down the winning condition before scoring either. The hybrid won the chosen quality metric by 0.81%, matched it on downstream tasks, produced a 6.3% smaller 4-bit file, and decoded 1.76x faster at 2048 tokens of context, 2.08x against an external model of similar size. In every measurement the speed advantage is near zero at an empty context and grows with length, which is what the mechanism predicts and what a merely leaner model would not show. A simple bandwidth calculation predicts only 1.17x, so memory volume alone does not explain the gap. We also report what did not work: an unmitigated 4-bit quality cost, roughly half the convolution channels ending up inert and impossible to remove, and a vocabulary larger than this model size warrants.

Legal work, with its heavy reliance on processing large amounts of text, is often considered one of the domains most exposed to the use of LLMs. Contract ``scrubbing,'' the final review of transactional agreements for errors and inconsistencies, is a particularly suitable task for automation, because it is routine, painstaking work requiring detailed attention to long documents. Scrubbing also seems to align naturally with the general capabilities expected of frontier LLMs around long-context reasoning, consistency checking, and named entity recognition (NER). Despite the economic value and potential for automation, no formal evaluations of LLMs performing contract scrubbing have been conducted. We introduce ContractScrub, the first benchmark designed to evaluate contract scrubbing capabilities, comprising contracts hand-crafted by experienced lawyers over diverse error categories such as misuse of defined terms, incorrect references, and inconsistent language. Frontier models perform surprisingly poorly with only one model reaching 0.75 macro average recall despite strong performance on seemingly related general benchmarks, demonstrating the practical limits of current models and the importance of narrowly targeted, domain-specific benchmarks for measuring real-world impact.

Memory has become a key component of large language models, enabling them to retain information and learn from long-term interactions. However, existing memory benchmarks mainly evaluate whether information is correctly extracted, stored, and retrieved, while largely overlooking how retrieved memories reshape model reasoning and affect performance on the current task. We identify memory-induced cognitive traps: even faithfully recorded and semantically relevant memories can distort model reasoning or beliefs and degrade current task performance. To systematically evaluate these failure modes, we introduce MemTrapBench, which covers two forms of cognitive traps: Reasoning Fixation and Belief Distortion. Experiments across two model families and five representative memory frameworks show that MemTrapBench is challenging: all evaluated memory strategies underperform the no-memory setting, with even the strongest methods suffering drops of more than 10%. To mitigate these cognitive traps, we propose AdaptiveMem, a simple yet effective inference-time method that instructs LLMs to avoid memory traps. AdaptiveMem mitigates cognitive traps on MemTrapBench while preserving or improving performance on standard memory benchmarks across diverse memory frameworks.

Software form has undergone two paradigm shifts since its inception: Software 1.0, in which instructions determine behavior, and Software 2.0, in which data determines behavior (machine learning). This paper argues that a third shift - Software 3.0, in which context and reasoning determine behavior - is now underway, and contends that its terminal form converges to three elements: a generalized database (the unified abstraction of all persistent state and memory), a large model (the intelligence core that performs reasoning and generation), and an agent (the execution loop connecting the first two). The core argument is as follows: in the traditional three-tier architecture, the user-interface layer will be absorbed by the model's ability to generate interfaces on demand, the business-logic layer will be re-partitioned along "expressibility x criticality" into model reasoning and storage constraints (with residual deterministic logic retained as tools), and only the data layer will be elevated into the sole persistent infrastructure. We formalize this convergence thesis, present a minimal reference architecture, report evidence from real prototypes and a live model, and systematically analyze both the conditions under which it holds and the boundaries where it fails - determinism, cost, security, and verifiability delimit the thesis's domain of applicability. We argue that the thesis holds in task domains that are expressible, verifiable, externally stateful, and tool-complete, and that it will reshape the roles of developers, the database industry, and the software-engineering discipline.

Technical documentation is written for human developers, but an increasing share of software changes is now authored by autonomous coding agents. Which documents they consult, when, and what follows remain unknown. We conduct a behaviour-grounded study of agent-documentation interaction across two public datasets: 557 agentic coding sessions from SWE-chat, yielding 94,813 development events including 3,033 documentation interactions; and 33,097 agentic pull requests from AIDev, with 690,260 classified file-level change records. Four findings challenge current documentation practice. First, agents' documentation work is dominated by agent-facing artefacts: instruction files and working notes account for 60.5% of all documentation interactions, versus 10.6% for classical technical documentation and 1.3% for API references. Second, the link between consultation and code editing is unresolved: the adjacent transition probability is 0.002 and the unadjusted three-event lift 1.05, whereas a stage-adjusted model places it above unity (OR 1.33 [1.09, 1.62]); documentation creation is elevated unadjusted (lift 1.67) but its adjusted interval includes unity. Third, no explicit documentation-based validation sequence was observed, and consultation is associated with less immediate testing (lift 0.23, cluster CI 0.08-0.45; adjusted OR 0.39 [0.25, 0.60]). Fourth, consultation is self-initiated (70.2%) far more often than failure-driven (7.5%), and documentation trails code: among multi-commit pull requests changing both, code is touched first 4.7x more often. From these traces we derive a descriptive model of agent-documentation interaction as a two-lobed cycle rather than a linear journey, and show that two widely assumed properties of "agent-friendly" documentation - actionability and verifiability - lack consistent behavioural support. We release our pipeline, coding scheme, and event-level data.

Practitioners inferring causality from observational data usually rely on a single method and treat its output as causal truth. Recent tools select an optimal method for a dataset, and recent ensembles aggregate multiple causal-discovery algorithms into one graph, but little work pools evidence across different mathematical traditions, including non-causal ones. We present Multi-Method Causal Evidence Synthesis (MCES), a framework that ranks which candidate drivers in an observational system are most likely relevant to a set of outcomes, and with what strength of evidence. MCES runs eleven methods across eight mathematical traditions on observational panel data and pools their outputs into a Convergent Evidence Score (CES), a linear opinion pool. CES quantifies convergence of evidence across analytical lenses: the degree to which methods with different assumptions point to the same driver-outcome relationship. It does not claim causal identification in the interventionist sense; it supports hypothesis prioritization, not a transferable probability of causation. MCES first applies Structural-Behavioral Decomposition to remove definitional (algebraic) relationships, then runs all methods, normalizes outputs to [0,1], and pools them. We distinguish MCES from method selection, structural ensembles, prediction ensembles, and literature synthesis. Using synthetic data with embedded ground truth, the Sachs protein-signaling benchmark, six Bayesian-network structure benchmarks, and two further synthetic domains, we show MCES ranks true edges near the top (Precision@5 = 1.0, Precision@10 = 0.96 on the primary scenario), with a low empirical rate of null pairs reaching Moderate-or-higher convergence. Our central point is not that the pool beats every individual method, but that no single method is uniformly best across the evaluated scenarios, so MCES offers a method-agnostic default.

Non-invasive decoding of inner speech faces a fundamental data problem: a corpus pairing brain activity with a person's spontaneous inner monologue cannot be collected, and the available proxy paradigms (cued repetitive and retrospectively reported generative inner speech) are slow to acquire, poorly time-locked, and subject compliance is unverifiable. We therefore treat silent reading as a scalable proxy task and ask how much lexical and semantic information a contrastive decoder can extract from it. We report an open-vocabulary analysis of approximately 240,000 word presentations recorded from a single densely-sampled participant across 393 runs (ca. 49 h) of 19-channel dry-electrode EEG. Words from continuous narrative text were presented in rapid serial visual presentation, with typography randomised on every trial to partially decorrelate word identity from low-level visual form. A convolutional EEG encoder, optionally followed by a causal transformer, was trained with a CLIP-style contrastive objective to align short EEG windows with hidden-state embeddings of the presented word taken from a large language model. Decoding, evaluated as word-grouped top-10 retrieval against permutation baselines, was reliably above chance, extended to mid-frequency and rare words, and scaled log-linearly with training-data volume with no sign of saturation. Removing occipital and posterior-temporal electrodes reduced the word-level gain by roughly one third but left context tracking unchanged. Control analyses separate word-level decoding from narrative context tracking and from a non-neural positional prior introduced by the transformer's positional embedding. These results establish that open-vocabulary word-level information is recoverable from EEG during silent reading, and that decoding is data-limited rather than saturated.

Classical information criteria such as the Bayesian Information Criterion (BIC) rely on regularity assumptions that break down for singular models, leading to incorrect model selection in settings such as deep learning. The Widely Applicable Bayesian Information Criterion (WBIC) relies on local learning coefficients $λ$, which in the analytic case coincides with local Real Log Canonical Thresholds (RLCT) of the Kullback-Leibler divergence of the model, to capture correct marginal likelihood asymptotics. Exact computation of the learning coefficients has been limited to special cases, and only sampling-based estimation methods are generally applicable. We present the first deterministic algorithm that computes local RLCTs exactly for any two-dimensional model whose Kullback-Leibler distance is contact equivalent to a polynomial, derive a bound on its complexity, and demonstrate its effectiveness for a broad class of models, with applications including polynomial neural networks. Beyond providing ground truth to calibrate sampling-based estimators, exact computation reveals algebraic structure in learning coefficients that sampling cannot and out-speeds it in the shallow regime.

Studies of machine-learning-based power-system protection increasingly report near-perfect scores, yet the meaning of those scores depends strongly on the evaluation setting. Protection task, physical scope, measurements, timing, targets, preprocessing, and validation often vary jointly and remain incompletely specified. This paper proposes a standardization-oriented framework that treats evaluation design as part of the scientific contribution. It defines seven required study dimensions: protection objective, physical scope, observability, timing and decision windows, targets and sample validity, validation protocol, and evaluation outputs. The framework is instantiated in a bounded case study on the public PROTECT-90 electromagnetic-transient benchmark, comprising 9022 simulated episodes from a 90 kV double-line topology, for onset-conditioned fault classification and localization. Under centralized sensing, simulation-metadata-aligned 20 ms windows, and episode-grouped validation, a multi-layer perceptron (MLP) achieved a five-fold mean macro-averaged F1 score of 0.991 +/- 0.001 for classification and a localization mean absolute error of 10.20 +/- 0.25% of line length (mean +/- std across episode-grouped folds). Extending the decision horizon to 50 ms preserved this task-dependent performance asymmetry, while reduced observability approximately doubled the MLP localization error but had little effect on classification. A synchronized two-ended conventional locator outperformed the learning locators under its richer clean information set, and measurement degradation showed that clean predictive performance did not determine robustness. The framework turns evaluation assumptions into explicit, reproducible evidence and provides a basis for more comparable, auditable evaluation and future certification-oriented assessment of machine-learning protection functions.

Attention directly derives normalized information flow from pairwise scores. We introduce Relation, an alternative token-mixing primitive that first organizes pairwise evidence into explicit Self and Exchange relations and derives information flow afterward. This relational organization gives rise to Full Relation, FlashRelation, Linear Relation, Hybrid Relation, and a KV-style Relation Cache. Across matched decoder-only models at approximately 10M, 30M, and 100M parameters, Full Relation achieves lower final validation NLL than MHA at all three scales. In a fixed-context reference benchmark, FlashRelation is 3.60-4.41x faster than the materialized Full Relation implementation. Across scale-matched production workloads, it reaches 76.4-84.9% of PyTorch FlashAttention throughput while executing the Full Relation operator. Hybrid Relation uses 75% Linear Relation layers and achieves strong language-modeling quality. These results support a relation-first view of token mixing: ask Self, ask Others, then let Flow follow Relation.

We present a novel approach to efficient LLM agent harness optimization through adaptive validation task selection. Harness optimization iteratively rewrites the harness code based on validation performance, enabling substantial performance gains without updating the underlying model weights. Existing approaches, however, evaluate a fixed validation set in full at every iteration, incurring substantial evaluation costs even on tasks that become less discriminative as the harness evolves. We propose $\textbf{Task-CoEvolve}$, which co-evolves the validation tasks with the harness by addressing two challenges: selecting informative tasks and estimating full-set performance from partial evaluations. Task-CoEvolve builds on the observation that tasks on which candidate harnesses disagree are more informative for distinguishing among them than tasks that are consistently solved or failed. It uses variance-weighted sampling based on past outcomes to focus evaluation on tasks near the agent's capability frontier, with the sampling distribution adapting as the harness evolves. It then estimates full-set scores from the sampled tasks by accounting for their sampling probabilities, enabling consistent comparisons across iterations despite evaluating different subsets. Experiments on online text classification and Terminal-Bench 2.1 show that Task-CoEvolve consistently outperforms fixed-subset baselines and matches the final performance of full-set search while reducing the number of evaluations during optimization by 80%. Code will be released at https://github.com/Agent4Science-UTokyo/Task-CoEvolve.

Open-source libraries play an important role in software development by providing reusable features that expedite the development process. As libraries evolve, they release new versions that add features, fix bugs, or apply security patches. In this process, they may break the contract established with their clients by introducing breaking changes (BCs) that alter the runtime behavior and break client applications. Client-side test suites often fail to detect these BCs because of limited library coverage that does not exercise all library methods used in the client's codebase. We propose BreakGuard, an approach that generates a test suite to detect breaking changes in clients. BreakGuard statically extracts every client method (focal method) that invokes the target library method (call site), then generates tests per focal method. A test detects a BC if it passes on the pre-breaking version and fails on the breaking version. We evaluate our approach on 89 real-world breaking changes from the BUMP dataset, using 3 LLMs (GPT4o, Qwen3-coder-480B, GPT-OSS-120B) and three context levels: minimal, method, and class. Using the best-performing configuration, BreakGuard detects 30.3% of breaking changes (27 of 89) at a mean cost of roughly $0.90 USD per detected breaking change. We successfully detected BCs from different library categories (e.g., JSON libraries, logging, parsing), but we find LLM-generated tests to be more reliable for detecting crash-type breaking changes as opposed to behavioural BCs.

Instruction-based image editing uses a planner-renderer pipeline: a vision-language model (VLM) first converts the instruction into an edit plan, and a diffusion model then executes that plan. Training such systems with only final-image rewards is inefficient because a poor edit does not reveal whether additional optimization should place more emphasis on the planner or the renderer, and even planner-dominant cases remain difficult to localize within a free-form reasoning trace. We present DARS, a reinforcement learning framework for dual-level credit assignment in this two-stage setting. Across modules, multi-plan multi-render rollouts estimate between-plan and within-plan reward variability for soft module routing, while rollout mean rewards provide hardness estimates for an adaptive curriculum. Within the planner, a four-field structured reasoning output enables a prefix-gated reward and token-level advantage reweighting, turning outcome-level feedback into localized supervision. Experiments on five benchmarks show that DARS outperforms a Joint~RL baseline with the same backbone, data, reward model, and rollout budget, with the largest gains on reasoning-intensive edits.

Large language models (LLMs) have shown growing potential for automated theoretical computer science (TCS) research, yet existing benchmarks remain far from realistic research settings. We introduce \ourbenchmark, an expert-validated benchmark for evaluating LLMs on frontier, end-to-end TCS research. \ourbenchmark contains $175$ instances drawn from papers accepted to STOC, FOCS, SODA, and COLT in 2025-2026, preserving paper-specific definitions, assumptions, and proof dependencies, with expert-verified Lean formalizations and proofs. Evaluations of leading LLMs reveal that current models remain far from reliably completing the full research pipeline. In particular, autoformalization is the sharpest bottleneck: the best model achieves only $11.5$ on translating natural-language claims into formal theorem statements, compared with $28.6$ Pass@8 when proving human-provided formal statements. Building on \ourbenchmark, we further develop an automated TCS research framework that generates, formalizes, filters, and proves new claims. Of $64$ generated claims, only $6$ ultimately pass expert evaluation and proof verification, indicating that beyond formalization, limited research taste remains another major barrier to autonomous TCS research.

We present a novel view on feature evolution in Vision Transformers (ViTs) by visualizing the training process over two dimensions -- network depth (layer) and training time (epochs). We employ Sparse Autoencoders (SAEs) to extract candidate sparse features from CLS-token representations and compare their activation profiles across epoch--layer pairs. This allows us to study feature-level dynamics that are not directly visible from representation-level similarity measures. Furthermore, we demonstrate how this framework of feature evolution allows us to describe feature migration, the change in the layer where a feature is most detectable during training. Our experiments show that migration is concentrated early in training, occurs more often toward earlier layers than toward deeper layers, and declines as feature organization stabilizes. We further find that deeper layers stabilize earlier and more strongly than shallow layers. The results show that our approach can be employed as a tool for understanding how ViTs learn and evolve.

Autonomous vehicles require robust perception and decision-making capabilities to operate in diverse and unseen scenarios. While reinforcement learning and rule-based methods can provide effective control and safety mechanisms, their performance may degrade in situations requiring contextual reasoning. Large Language Models (LLMs) have demonstrated strong capabilities in understanding multimodal information and generating contextual reasoning, however, their use for direct vehicle control can introduce latency and hallucination risks. To address these limitations, a hybrid framework is proposed. This system uses an orchestrator to coordinate PPO-trained reinforcement learning and PID control, with LLM common-sense reasoning applied throughout the framework. LLM reasoning is further employed iteratively to refine the RL reward function for dynamic driving environments. The proposed framework is evaluated in highly randomized CARLA scenarios under diverse environmental and traffic conditions. The results demonstrate the potential of integrating LLM-based reasoning with conventional autonomous driving methods while retaining structured control and safety mechanism.

We study inference-time control for text generation in discrete diffusion language models, where the goal is to steer sampling toward sequence-level rewards without retraining. Prior work in this domain has focused on particle-based methods such as best-of-$n$ sampling and bootstrap sequential Monte Carlo, which may suffer from overoptimism and weight degeneracy, respectively. We address these limitations using \emph{nested} sequential Monte Carlo methods. We formulate nested SMC (NSMC) and fully-adapted nested SMC (FA-NSMC) for Feynman--Kac steering, identifying and correcting errors in prior formulations that lead to biased final estimates. We evaluate these methods on toxicity and fluency steering tasks, showing that NSMC and FA-NSMC consistently outperform best-of-$n$ and bootstrap SMC.

The emergence of large-scale Weather and Climate (W&C) datasets offers new opportunities for modeling extreme Earth events (ExEE) and their impacts using deep learning. However, their adoption in operational settings remains limited by the lack of models' interpretability. While for conventional text and image modalities, tools such as Sparse Autoencoders (SAEs) have proven effective for extracting human-understandable concepts, their use for the analysis of ExEE remains challenging due to the nature of W&C data. To address this, we introduce (i) a geographic location-based modulation of the inputs of SAE to capture the local semantic meaning of environmental patterns, and (ii) an ensemble of rule-based SAE-Xplainers to interpret the resulting high-dimensional features derived from complex, multi-modal environmental predictors. We evaluate our method on three ExEE types: the prediction of fires, and the detection of tropical cyclones and atmospheric rivers. We show that SAE input modulation improves both reconstruction performance and feature utilization, and that our SAE-Xplainers enable faithful interpretation of complex climatic patterns by unfolding them into human-understandable rules that are consistent with the scientific literature, while also supporting the identification of feature absorption.

Large language models (LLMs) are increasingly used in settings where textual summaries, numerical observations, and external tool outputs may provide conflicting evidence. We study how LLMs arbitrate between such sources when they support opposing decisions. To do so, we introduce a controlled synthetic benchmark in which latent risk trajectories generate both numerical time series and natural language summaries, allowing us to construct conflicts where exactly one evidence source is aligned with the ground-truth label. This design lets us independently manipulate modality, temporal recency, source reliability, and evidence provenance. Across open-weight instruction-tuned models, we find that arbitration behaviour is systematic rather than random: models exhibit distinct text-versus-number preferences, follow temporal recency more consistently than explicit reliability cues, and can over-rely on external forecasts even when they conflict with direct contextual evidence. These results suggest that current LLMs often rely on heuristic arbitration strategies when integrating heterogeneous evidence, highlighting a failure mode for tool-augmented decision systems.

Mobile manipulation requires a robot to predict how locomotion and arm motion jointly alter future observations and control. Existing world-action models, developed largely for fixed-base platforms, do not explicitly distinguish camera ego-motion from base and arm actions. Here we introduce DECOWAM, a whole-body world-action model that separates these factors through dedicated conditional interfaces. DECOWAM freezes an adapted FastWAM backbone and trains residual adapters, an action-equivalent future bottleneck distilled from privileged observations, adversarially separated base and arm latents, and base-velocity conditioning for video prediction. We further introduce ARMDOG, a real-robot dataset that synchronizes video, whole-body state and action, and language. On a fixed replay protocol, DECOWAM improved both future-video and action prediction over FastWAM, reducing action MSE by 21.7% with 25.95M trainable adaptation parameters. Across 79 closed-loop trials per method, it achieved the highest observed whole-body coordination and base-displacement robustness among the compared systems, while task completion remained comparable to the strongest baseline. These results show that embodiment-aware factorization can support parameter-efficient joint visual prediction and whole-body control under moving viewpoints.

We introduce OenoBench, a wine-domain knowledge benchmark of 3,266 multiple-choice questions across six pillars (regions, grape varieties, viticulture, winemaking, producers, business) and four difficulty tiers. The corpus is built from 38,104 atomic, source-anchored facts extracted by 35 provenance-verified scrapers from government registries (INAO, TTB, OIV), peer-reviewed journals, and Wikipedia/Wikidata. Our methodological contribution is an LLM-driven pipeline in which language models reformat verified facts and audit the result, but never serve as the source of truth: every claim traces to a URL, every question is generated by one of five strategies across five generator families, and every question is scored by a nine-agent audit calibrated against a human gold sheet via Cohen's $κ$. Evaluating sixteen frontier configurations, we find: (i) overall accuracy spans 53%-84%, led by o3 at 83.6%; (ii) reasoning-mode lift concentrates in DeepSeek R1 (+6.8pp) and is absent in Claude Opus and Gemini Pro; (iii) Anthropic shows +9pp self preference on its own questions while Google shows -8pp inverse preference; (iv) frontier open-weight models share the cost-vs-accuracy Pareto frontier with proprietary reasoning models; and (v) every config gains around 33pp on closed-book solvable items, revealing a parametric-recall ceiling that only the contextual slice avoids. We release corpus, audit findings, human-review app, and construction code under CC-BY-SA-4.0.

Artificial immune networks (AINs) are naturally memory-forming systems, but conventional visual AINs often rely on flattened vector affinity that ignores spatial structure. This paper studies whether structured, gradient-free immune affinity can make Deep AINs viable as replay-free visual class-incremental representation-memory learners. Visual B-cells are formalized as structured templates, including shifted-template affinity, zero-normalized cross-correlation (ZNCC) filters, and feature-map binding profiles. A repertoire is treated both as memory and as a representation-inducing basis, while depth is obtained by passing binding-profile response maps to subsequent immune layers. The resulting Deep AIN exhibits adaptive latent coordinate reorganization: as new classes arrive, the binding-profile space evolves while retaining recoverable structure for earlier classes. Experiments on sklearn digits, MNIST, Fashion-MNIST, and KMNIST show that preserving response maps is critical. Scalar binding-profile variants underperform, whereas feature-map Deep AINs learn class-discriminative visual memory without replay, label-driven immune updates, or backpropagation through the immune layers. On sklearn digits, downstream probes fitted on the learned binding profiles reach 0.939 final balanced accuracy with logistic regression and 0.902 with 1-nearest-neighbour after all ten classes are encountered, with initial-class retention of 0.978. Adaptive layer-wise scale calibration further improves the two-layer feature-map Deep AIN to 0.978 balanced accuracy. With the same calibration rule, Fashion-MNIST reaches 0.814 and KMNIST reaches 0.853. These probes are external validation tools, not components of the AIN. The results identify structured affinity, response-map preservation, adaptive latent reorganization, and layer-wise scale calibration as key mechanisms for replay-free visual immune memory.

LLM-based Multi-Agent Systems (MAS) achieve strong performance on complex reasoning tasks by coordinating multiple agents, but at the cost of substantial token consumption. Recent work on automatic topology design, ARG-Designer, has reframed this problem as autoregressive graph generation. However, its training objective provides no explicit incentive for the model to generate sparse and efficient topologies. We address this limitation by introducing a Reward-Guided Autoregressive Graph Generation (RGA-Designer) inspired by Reinforcement Learning from Human Feedback (RLHF). We train a reward model that jointly captures task correctness and structural compactness, and then fine-tune the pretrained graph generator using the reward model as feedback. Our method preserves task accuracy at the level of ARG-Designer while reducing token consumption by an average of 20.5%.

Humanoid robots have recently demonstrated promising capabilities in real-world ball sports. However, achieving professional motion styles while maintaining strong task performance remains challenging. In this work, we propose AdaPT, an Adaptive Motion Planning and Tracking framework that learns professional tennis serving and rally styles directly from broadcast videos. This hierarchical design is motivated by the key insight that the planner generates stylistic kinematic motions, while the tracker executes them with minimal interference with planning. Despite its effectiveness in simulation, a substantial sim-to-real gap emerges: tracking performance inevitably degrades on real robots, and this degradation is partially overlooked by autoregressive planning and further compounded by noisy perception. To address these issues, our adaptation mechanism improves tracking robustness by learning to track randomized execution speeds, while conditioning the planner on a learned motion-speed adapter to mitigate compounding errors. Real-world experiments on the Unitree G1 demonstrate the effectiveness of our adaptation mechanism in bridging the sim-to-real gap. We further deploy AdaPT policies on the full-size Dobot Atom humanoid robot (1.7m) and demonstrate in-the-wild serving without motion capture. Beyond these results, our real-world experiments reveal both algorithmic and engineering insights for future humanoid ball-sports systems. Videos and code are available on our \href{https://humanoidtennis.github.io/AdaPT/}{project website}.

Robots executing long-horizon manipulation tasks from natural-language instructions must reason about both semantic task structure and geometric feasibility. However, under partial observability, the availability of goal-relevant objects may be uncertain. In such cases, approaches that combine Vision-Language Models (VLMs) with Task and Motion Planning (TAMP) may generate subgoals that rely on the VLM's prior knowledge without observational support, leading to execution failures or unintended outcomes. We propose Evidence Acquisition and Feasibility Gating (EAFG), a framework that acquires visual evidence through VLM-generated exploratory subgoals and TAMP-based execution. EAFG then applies a feasibility gate to decide whether to proceed with task planning, acquire further evidence, or halt. Our experiments show that, in cooking tasks with ambiguous object use, EAFG improves recipe completion by discovering task-relevant objects before planning. For instructions requiring an absent object, EAFG promotes appropriate halt decisions and reduces repeated attempts to manipulate that object.

Question Answering over Temporal Knowledge Graphs (TKGQA) requires reasoning over time-sensitive facts, yet existing embedding-based methods struggle with multi-step queries due to single-pass reasoning pipelines. We propose SABET-QA, a framework that iteratively refines reasoning states across multiple hops via a bidirectional entity-temporal scoring mechanism and a slot-aware contextualization module that aligns question semantics with temporal KG embeddings. A differentiable working memory enables progressive hypothesis refinement, while auxiliary temporal boundaries serve as coarse supervision when available. Experiments on CronQuestions, Complex-CronQuestions, MultiTQ, and TimeQuestions demonstrate consistent improvements over strong baselines, particularly on complex multi-step temporal queries.

World models construct latent states that support prediction, planning, and reasoning about an underlying system. Joint-embedding predictive architectures (JEPAs) offer a direct way to learn such states by predicting targets in representation space instead of reconstructing every detail of the observation. Standard JEPAs, however, organize all predictable content through one target embedding and one prediction pathway. In complex systems, this monolithic state can allocate redundant capacity to dominant signals while providing weak or conflicting gradients to less dominant predictive structure. We introduce \method, a latent world-modeling framework based on orthogonal predictive factorization. Learned basis matrices analyze each target state into multiple components, and a dedicated prediction branch estimates each component from a shared context representation. Predictive regression preserves the factor magnitudes required for state synthesis, an orthogonality objective discourages repeated directions, factor-activity regularization maintains variation in projected targets, and online variance regularization discourages coordinate-wise encoder collapse. Predicted components are synthesized into a complete latent state that can be used by a readout, decoder, planner, or autoregressive rollout. The same predictive-state mechanism applies when the target is temporally future, spatially hidden, or another partial observation of the same system. Experiments on controlled vision, single-cell transcriptomics, longitudinal health records, continuous control, and molecular dynamics evaluate representation quality, forecasting, planning, and long-horizon stability.

Mixture-of-Experts (MoE) architectures significantly expand model capacity without a proportional increase in computational cost. However, optimizing their hyperparameters---particularly the learning rate---at extreme scales of both model size and token budget via sweeping remains computationally prohibitive. In this paper, we propose a compute-efficient, two-step hyperparameter transfer framework that estimates optimal learning rates for training large MoE models by transferring them across scaling model widths, and subsequently extrapolating to trillion-token horizons. First, we formulate a Maximal Update Parameterization ($μ$P) adaptation for MoE architectures utilizing Multi-head Latent Attention (MLA) and the Muon optimizer, demonstrating that optimal learning rates transfer consistently across width-scaled models. Second, we extend this transferability along the token dimension by establishing a predictive scaling law. By applying linear regression to the optimal values derived from small proxy models on limited budgets, we successfully extrapolate the ideal learning rate to massive training horizons (e.g., 10 trillion tokens) with high fidelity ($R^2=0.95$). Consequently, this indicates that proxy training on small models is sufficient to determine the optimal learning rate for the extensive training of large-scale MoEs. We apply the proposed methodology to pretrain our foundation model (155B total, 17B active parameters) from scratch, and the stable training and evaluation results validate that optimal configurations for full-scale target models can be accurately predicted with minimal ablation costs.

Hidden chain-of-thought (CoT) traces, especially those from frontier proprietary large reasoning models (LRMs), are valuable model assets. Yet whether these hidden CoTs can be directly extracted from black-box models remains largely unexplored. In this work, we systematically study whether hidden CoTs can be extracted near-verbatim from black-box LRMs through API interactions. We identify a previously overlooked reasoning replay surface between tool calls and develop EchoCoT, a multi-step attack that iteratively extracts hidden CoTs using API-returned fidelity signals. We further develop an LLM-based optimization framework that automatically searches for an effective universal injection trajectory across various datasets. We evaluate EchoCoT on three open-source and five frontier proprietary LRMs. On open-source LRMs, EchoCoT achieves up to 66.4\% near-verbatim extraction success, with the extracted trace length within 10\% of the target and at least 90\% of tokens exactly matching the target CoT. The same injection trajectory also generalizes to unseen datasets, achieving up to 80\% extraction success under the same criterion. For tested frontier proprietary LRMs, a substantial fraction of extracted CoTs closely align with provider-reported reasoning lengths and available CoT summaries. EchoCoT can also extract very long CoTs: on Gemini-2.5, it extracts 33,463 tokens from a 32,948-token target. These results establish hidden-CoT extraction as a practical security risk and highlight the need to better protect hidden CoT assets.

Multi-module systems often expose every module to the full input. We test whether restricting evidence visibility changes which solutions gradient-based training discovers. Four-cell societies share one frozen pretrained language model and one low-rank adapter, communicating only through two model-width continuous vectors in a fixed relay. On a prospectively sealed natural-language function-composition task, we train ten matched restricted/global pairs sharing initialization bytes, training order, token layout, parameters, and computation; only the attention mask differs. Restricted societies outperform their globally visible twins by at least 20 points at both depths in 9 of 10 pairs, with median paired advantages of 0.7648 and 0.6050. Cutting communication reduces every restricted society to chance, and the depth-three advantage remains 0.558 on programs whose composite function never appeared in training. Across six audited restricted societies, same-value packet transplants preserve behavior at 0.94-1.00 across all tested interfaces; destructive interventions collapse performance; and counterfactual packets redirect outputs toward the mathematically predicted answer. The sole high-performing global model also requires communication, but its same-value packets are not interchangeable across episodes. Restricted visibility is thus not necessary for composition; under this protocol it substantially increases the probability of a generalizing relay and favors a reusable, value-indexed interface. The complete preregistered battery nevertheless formally fails because restricted-arm median depth-three accuracy is 0.6988, below the 0.70 floor. An earlier qualification cohort likewise yielded 0/10 complete passes: one model met every task-performance gate, but all ten failed ordinary-language preservation, confining the system to explicitly task-gated use.

The integration of Artificial Intelligence (AI) in safety-critical aviation systems presents significant challenges for certification and deployment. Aviation, often regarded as the safest form of transportation, relies on numerous safety-critical systems. For future safety-critical AI-based systems, EASA requires a Safety-by-Design approach, which can be achieved by using Safety Nets that combine neural network compression with lookup tables to ensure 100 % correct runtime behavior across the discretized operational design domain. Although Safety Nets have been studied, no comprehensive study of their performance characteristics and system design trade-offs has been conducted. This work presents the first systematic analysis of the trade-off between neural network and lookup table size in Safety Nets. By systematically comparing neural networks with diverse architectures, this study identifies optimal design parameters that minimize overall storage and memory requirements while maintaining certification compliance. Results demonstrate that architectures with 3 to 5 hidden layers, each with approximately 50 to 100 nodes, combined with one-hot encoding, achieve the best balance. In these configurations, neural networks accurately represent at least 97 % of the data, while compact lookup tables handle the remaining errors. The resulting Safety Nets reduce the system size by almost three orders of magnitude, fitting within the memory budget of current avionics hardware while guaranteeing 100 % correct outputs across the entire discretized input space, as required by EASA guidelines. This work provides the first-ever open-source implementation of Safety Nets for HCAS and VCAS with replicable results, demonstrating a practical pathway toward certifiable AI-based systems in aviation and establishing Safety Nets as a viable Safety-by-Design solution for safety-critical applications.

Probabilistic time series forecasting remains challenging, largely because modeling distinct trend and seasonal dynamics requires specialized approaches. Existing methods often fail to capture the unique inner properties of these components, lack interpretability, or suffer from heavy memory and runtime overhead. To address these limitations, we propose DecoVAE, a lightweight interpretable trend-seasonal VAE framework that explicitly decomposes time series into trend and seasonal components by applying domain-specific inductive biases. The trend stream enforces structural smoothness using a differential regularizer on the latent trajectory, analogous to the Hodrick-Prescott filter. Concurrently, the seasonal stream operates in the frequency domain via a complex Gaussian VAE, natively capturing the amplitude and phase of periodic patterns. Extensive evaluations across seven real-world benchmarks show that DecoVAE consistently outperforms strong baselines. It achieves reductions of up to 14.96\% in CRPS and 23.30\% in NMAE for short-term forecasting, and up to 52.68\% and 26.51\% for long-term horizons. Crucially, DecoVAE yields these accuracy gains while remaining highly efficient, reducing model weight by up to 93\% and accelerating speed by up to 74\% compared to the second-best method.

Watermarking schemes for large language model output are evaluated almost exclusively on English text using each scheme's detection threshold and a narrow set of quality measurements. Multilingual deployment exposes evaluation-design choices that are inconsequential on English but determine conclusions cross-lingually. We propose an evaluation framework with four components: detection thresholds calibrated empirically per deployment context, a threshold-independent companion measurement that distinguishes calibration failures from detection failures, three disjoint quality measurement paradigms (distributional, paired-semantic, and reference-perplexity), and a generalized-entropy decomposition of cross-language disparity over a typological family partition. Applied to six watermarking schemes, three open-weight generators, eleven languages spanning four scripts and eight typological families, and both base and instruction-tuned regimes, the framework reveals failure modes that single-language single-paradigm evaluation cannot surface. Across detection and quality, observed disparity is predominantly between-family on the typological partition, indicating that cross-lingual fairness gaps in watermarking are structural to language properties rather than idiosyncratic to particular languages.

Early Classification of Time Series (ECTS) requires making accurate decisions as early as possible in inherently online and evolving environments. Yet, most existing methods assume stationarity and rely on separable designs, where classification and triggering are optimized independently, an assumption that fundamentally limits their adaptability under drift. In this work, we challenge this paradigm and study ECTS under non-stationary conditions. We provide the first systematic comparison between separable and end-to-end approaches across controlled drifting scenarios. Building on Reinforcement Learning, we introduce DQeND, a unified architecture that jointly learns representation, classification, and triggering decisions, while remaining directly comparable to state-of-the-art separable baselines. Across a wide range of drifts, DQeND demonstrates strong robustness across various non-stationary scenarios, consistently outperforming separable baselines. An ablation study further highlights that jointly updating representation and decision modules is critical to these gains. Overall, our results indicate that end-to-end learning can offer improved adaptation capabilities for ECTS in dynamic environments, and motivate further investigation of alternatives to separable designs.

Research on the agency of advanced artificial intelligence (AI) systems focuses on agency as a normative concept and on the agency of particularly agentic AI systems. While recent work also focuses on the different profiles of agentic systems, no framework exists to address the question of the type of agency instantiated by advanced AI systems, particularly when considering non-moral forms of agency. Based on established theoretical positions in philosophy, ethics, legal theory and sociology, we develop a typology of agency for frontier AI systems consisting of three dimensions: the nature of agency (moral or legal), its mode (individual or collective) and its locus (human or non-human). Combining these dimensions produces eight possible instantiations of agency, which we classify as conventional, contested or controversial. The typology separates legal from moral agency and thereby creates conceptual space for considering individual, legal, non-human agency without presupposing that advanced AI systems are moral agents. We argue that this distinction is increasingly relevant where instrumental goal pursuit complicates the attribution of AI actions to particular human actors.

Artificial intelligence can support cultural heritage and digital humanities through large-scale retrieval and analysis of digitized collections. However, cultural heritage data are often distributed across institutions, constrained by ownership and access restrictions, and continuously evolving over time. Federated Continual Learning (FCL) is well suited to this setting, as it enables models to learn from distributed and sequential data without sharing raw collections. In this paper, we propose FedCurv-DR, a lightweight, regularisation-based FCL strategy. The method accumulates parameter-importance estimates across clients and experiences to protect learned knowledge, while updating them only at fixed intervals to minimize communication and computation overhead. We evaluate FedCurv-DR in a continual learning scenario using the WikiArt image dataset for genre classification with evolving styles, reporting performance, energy, and fairness metrics. Our results show that FedCurv- DR reduces forgetting and balances performance, fairness, and energy efficiency for sustainable AI in cultural heritage.

Streetscape quality has become a central concern in contemporary urban planning, particularly within the framework of the pedestrian-friendly 15-minute city, where walkability and public-space quality are increasingly recognized as key determinants of urban performance. However, assessing streetscape qualities across large suburban and peri-urban territories remains challenging due to the time and resource demands of conventional field surveys. This paper presents a planning-oriented assessment of streetscape qualities in the north-eastern periphery of Nice (France) using the latest release of SAGAI (Streetscape Analysis with Generative AI), an open-source workflow that leverages vision-language models (VLMs) for large-scale streetscape analysis from Google Street View imagery. The new release addresses limitations of the original framework through improved image acquisition, geographically consistent view generation, support for multiple VLM architectures, consensus-based inference, and an integrated analytical environment. The workflow is applied to several thousand street-level observations to evaluate qualities relevant to pedestrian-friendly urban environments: sidewalk presence, pedestrian entrance density, and vegetation. The resulting maps reveal that the desired streetscape qualities characterize only a fraction of today's suburban streetscapes, mainly in compact developments and traditional suburban faubourgs, while they are particularly lacking on residential hills. The analysis demonstrates the potential of contemporary VLMs to support urban diagnostics in extensive suburban territories where fieldwork would be prohibitively time-consuming. Beyond the case study, the paper illustrates how recent advances in vision-language models can contribute to evidence-based planning by enabling scalable, flexible, and interpretable assessments of urban public-space quality.

Probabilistic forecasting models are widely used for time series forecasting in domains such as energy systems, finance, medicine, and transportation. In recent years, deep generative models have shown strong results on probabilistic forecasting, yet many conventional approaches struggle to capture internal temporal dependencies, leading to latent representations with limited expressive power. To address this limitation, we propose \textit{CLaST}, a VAE framework for probabilistic multivariate time series forecasting. Unlike existing generative models, CLaST learns embeddings that preserve contextual similarity between observations through our contrastive loss function. Experiments across nine widely adopted benchmarks demonstrate that CLaST consistently surpasses strong baseline methods. In short-term forecasting tasks, our approach achieves improvements of up to $16.4\%$ in CRPS and $14.4\%$ in NMAE over the second-best method. Furthermore, in long-term prediction CLaST attains superior overall performance, exceeding the second-best method by up to $48.6\%$ and $25.1\%$ in CRPS and NMAE, respectively.

District heating energy hubs require reliable heat load forecasts for efficient operational scheduling. Conventional forecasting workflows train system-specific models on historical data, which can become burdensome when networks change through new consumers, retrofits, or changing operating regimes. Zero-shot time-series foundation models and in-context forecasting offer a promising alternative: they can adapt at inference time from recent observations rather than by repeated retraining. This study systematically evaluates TabPFN-TS against time-series foundation models and trained machine-learning baselines for probabilistic heat load forecasting in district heating networks. Unlike foundation models pretrained on large collections of real time series, TabPFN-TS relies on synthetic pretraining data, which avoids direct pretraining-test overlap but raises the question of whether the learned prior captures district heating dynamics. We analyze covariate choice, context length, temporal resolution, and prediction horizon on representative operating weeks, validate the selected configuration over a full year, and test transferability on a second network. The results identify hourly 24-hour forecasting with a 12-week rolling context and ambient temperature as a parsimonious high-performing configuration; longer context windows do not improve accuracy. TabPFN-TS remains close to Chronos-2 in deterministic accuracy, reaching CVRMSE values of 13.06% versus 12.48% on the main dataset, and lies within the critical-difference threshold in the daily-rank comparison. Although Chronos-2 achieves the lowest aggregate full-year error, TabPFN-TS shows better empirical calibration. Finally, the diagnostic findings motivate a Multi-Resolution Residual-Correction Forecaster that combines a low-frequency Base Forecaster with a short-horizon Residual Forecaster to improve longer-horizon planning accuracy.

Incomplete multimodal sentiment analysis has garnered significant attention in recent years. Existing approaches typically assume that data is missing at random or are designed specifically for certain missing patterns, ignoring the modality combination inconsistency between training and testing phases. However, in real-world scenarios, the testing phase often encounters modal combinations that were not present during the training phase, which leads to insufficient generalization capabilities and unstable performance. In this paper, we introduce the problem of Incomplete Multimodal Sentiment Analysis with Unseen Modality Combinations (IMSAUMC), aiming to enhance model generalization for unseen modality combinations. To address this challenge, we propose the model named $\textbf{C}$ontrastive $\textbf{M}$ixed $\textbf{P}$rompt $\textbf{L}$earning ($\textsf{CMPL}$) for IMSAUMC. It introduces a label-guided contrastive feature learning mechanism to learn robust and discriminative cross-modal representations. Additionally, we design modality-combination prompts with a soft router to facilitate better learning of various modality combinations. Furthermore, we introduce three prompt contrastive learning strategies, which enable effective learning of prompts corresponding to unseen modality combinations, thereby significantly strengthening the model's generalization capabilities in diverse testing scenarios. Extensive experiments on three widely used datasets demonstrate that $\textsf{CMPL}$ achieves more than a 5% improvement in accuracy compared to state-of-the-art approaches.

Preference optimization is a standard alignment method for generative models, yet extending it to continuous-time dynamics remains non-trivial. In flow matching, reward-driven updates modify transport trajectories without an inherent constraint to the pretrained data manifold and can move terminal samples off the pretrained support. We formalize this failure mode as manifold drift. Theoretically, we show that optimal flow matching recovers the terminal data distribution, whereas a preference update leaves the pretrained manifold whenever its induced terminal displacement has a nonzero normal component. As a remedy, we propose ThermoDPO, a temperature-controlled objective that anchors pairwise preference optimization on preferred samples. Across temperature regimes, this objective connects rejection sampling fine-tuning and FlowDPO and controls a pointwise reconstruction-based surrogate for manifold distance. To counteract diminished signals at low temperatures, we further introduce a weighted variant, ThermoDPO-weighted. On the main toy benchmark, ThermoDPO-weighted attains a StrictScore of 0.899, compared with 0.629 for FlowDPO and 0.857 for FlowDPO+RFT. On SD3.5-M at CFG = 4.5, it improves OCR by 47.5% and the average of four metrics by 16.0%.

Understanding object dynamics requires not only predicting future trajectories but also examining whether a model captures the physical properties that govern motion. However, existing benchmarks rarely expose object-level physical properties as explicit evaluation targets alongside trajectory forecasting. To address this gap, we introduce \emph{ExPhy}, a multi-object trajectory forecasting benchmark containing 24,000 simulated physical scenes with explicit object-level labels for mass, friction, and restitution. ExPhy provides observed and future trajectories together with an in-distribution (ID) split and two out-of-distribution (OOD) splits over physical parameters (OOD-Parameter) and initial states (OOD-Initial) for jointly evaluating trajectory forecasting and physical property estimation. We further instantiate \textsc{PhyODE}, a physics-guided model with an explicit property interface that estimates physical properties from observed trajectories and uses them for differentiable future rollout. On the long-horizon OOD-Initial setting, \textsc{PhyODE} reduces ADE and FDE by 33.1\% and 31.0\%, respectively, compared with the strongest baseline. Zero-shot evaluation on ComPhy further assesses cross-benchmark transfer. Property-level analyses reveal that accurate trajectory forecasting does not necessarily imply accurate recovery of the underlying physical properties. Code and data are available at https://github.com/Zest86/ExPhy.

Pretraining time series foundation models across heterogeneous datasets necessitates effective handling of varying sampling frequencies. Current methods either employ dataset-specific patch sizes and separate FFNs, leading to fragmented representations, or enforce a fixed patch size that neglects inherent temporal variations. To address this, we propose SATS, featuring a scale-aware token alignment mechanism that treats patch size as an explicit notion of scale. By incorporating a contrastive-inspired alignment regularizer, SATS aligns representation spaces across scales while preserving distinct modeling capacities. Furthermore, a hybrid masking strategy combining random and contiguous masking is introduced to capture multi-scale temporal structures. Experimental results on LSTF benchmarks demonstrate that SATS achieves a 9.2% improvement in MSE and an 8.3% gain in GIFT-Eval MASE compared to competitive baselines. Notably, SATS consistently delivers SOTA performance while achieving a 65.6% increase in model efficiency over advanced baselines, highlighting its effectiveness and scalability in time series pretraining.

We summarise the outcome of two summer internship projects based at the University of Manchester, focused on the break-even point in terms of environmental sustainability for ML-based data compression algorithms. Using the example of a ML-based lossless compression algorithm, we compare estimates for the carbon-equivalent of the infrastructure needed for ML training and inference with the carbon-equivalent savings from reduced disk storage requirements, and discuss their break-even point.

Loading reusable skill documents into a bounded context window is now the primary way large language model (LLM) agents acquire task-specific capabilities, which makes skill selection a first-order determinant of task performance and token cost. Yet current agents score skills independently by semantic relevance and assemble the set by top-$k$ or greedy packing, with no quality guarantee or cost awareness on the selected set. As a result, redundant or poorly chosen skills waste scarce context tokens and can even degrade performance. We give the first model of how the selected skill set shapes execution outcomes and cast skill selection as an optimization problem: choose a skill set under a hard token budget to maximize a monotone submodular benefit minus context penalty. For this problem, we develop Best Prefix Selection (BPS), a polynomial-time algorithm, and prove, to our knowledge, the first performance guarantee for skill selection: a bicriteria $(1-1/e,1)$ approximation whose benefit coefficient is optimal in polynomial time. On a contamination-controlled BigCodeBench variant, BPS outperforms all the baselines, reaching $0.73$ measured task success versus $0.20$--$0.52$ for released skill routers, text retrievers, and the executor's own selection, on $28\%$ fewer tokens than the strongest released router.

We present HealMed, an expert-reviewed benchmark for multilingual evaluation of large language models in medicine. HealMed contains 1,000 examples in each of nine languages, drawn from nine datasets and covering three task formats: MCQA, NLI and open-ended QA. The benchmark was developed over two years by 23 physicians and medical experts based across nine countries and regions. Each translation was evaluated and revised by two experts fluent in English and the corresponding target language. On HealMed, performance declined most in low-resource languages, although the size of the gap varied markedly across languages and models. The strongest proprietary models were the most stable across languages, whereas many open-source and medically specialized models showed larger and less consistent gaps. Medical specialization alone did not ensure multilingual robustness. Furthermore, expert revision could either raise or lower measured performance, indicating that translation quality materially affects cross-language evaluation results.

LLM agents in financial markets may cite rules yet still submit orders that violate executable constraints or misread surveillance evidence. We introduce ReguSim, a controlled financial-compliance environment, and ReguBench, a target-marked monitoring benchmark, to separate four artifacts: stated reasoning, attempted action, execution enforcement, and monitor evidence. In trader runs with DeepSeek V4 Pro and Gemini 3.5 Flash, visible rules reduce but do not eliminate rejected actions, and incentive or persona framing shifts behavior. A bridge study shows that trader rationales can mislead an independent monitor unless enforcement evidence is shown. In monitoring, simple structured baselines either match or exceed prompt-only LLMs. The results frame financial compliance evaluation as an audit of rule-grounded actions and evidence use, rather than a single compliance score.

Recently, open-vocabulary 3D object detection (3D-OVD) has gained increasing attention for its ability to detect unseen objects in 3D scenes. Existing approaches typically adopt a two-stage pipeline that first discovers novel objects using foundation models and then trains a 3D-OVD model based on these discovered objects. Although effective, this pipeline often suffers from inaccurate localization and mismatched classification during the discovery stage, which subsequently limits the performance of the model training stage. To address these limitations, we advocate for improving both the reliability of novel object discovery and the robustness of model training, and propose an innovative framework. Specifically, for reliable discovery, our co-distillation strategy distills high-quality novel objects by applying Hungarian matching over a comprehensive score that incorporates geometric consistency, structural objectness, and semantic certainty. To enhance robust model training, we further propose a dual-guidance learning scheme, incorporating a scene-awareness-guided uncertainty regularization for the regression head and an LLM-guided hierarchical alignment for the classification head, effectively mitigating the negative effects of imprecise 3D bounding boxes and semantic ambiguity. Extensive experiments on SUN RGB-D and ScanNetV2 demonstrate that our method achieves significant performance gains over state-of-the-art approaches. Code is available at https://github.com/shangboyuan/Co-3DGT

Multimodal sentiment analysis aims to infer affective states by integrating language, visual, and acoustic cues. However, real-world multimodal inputs are often incomplete or corrupted, which can weaken cross-modal complementarity and introduce misleading information into downstream fusion. Existing proxy-based methods for incomplete MSA commonly rely on one-shot proxy construction to compensate for degraded language information, but the generated proxy may be coarse or unreliable at initialization. Prematurely injecting such a proxy into multimodal reasoning can propagate initial errors and compromise sentiment prediction. To address this limitation, we propose an iterative proxy correction framework for robust incomplete MSA. Our method constructs a language-oriented proxy from non-language modalities and progressively refines it under multimodal context through gated residual correction. The corrected proxy is then adaptively fused with the observed language representation according to an estimated language reliability score, allowing the model to balance proxy-based compensation and trustworthy linguistic evidence. In addition, we introduce a stage-wise latent correction objective that uses the complete language representation as a training-time semantic anchor to stabilize the proxy refinement trajectory. Extensive experiments on MOSI, MOSEI, and SIMS under diverse missing-modality settings demonstrate that the proposed framework consistently outperforms competitive baselines and achieves robust sentiment prediction under incomplete inputs.

Multivariate time series forecasting (MTSF) is a fundamental task in many real world applications. Existing patch based forecasting methods generally fall into three categories: fixed partitioning, multi-scale partitioning, and extendable partitioning. Fixed partitioning often breaks meaningful temporal boundaries, multi-scale partitioning may introduce redundant representations across scales, and extendable partitioning improves flexibility but still lacks an explicit mechanism for organizing semantic structure and modeling interactions among heterogeneous temporal patterns. To address these limitations, we propose SCPaT, a Transformer based framework built on semantic structured partitioning. SCPaT first decomposes input sequences into semantically consistent units through adaptive semantic unit generation, then constructs a dynamic semantic graph to model directed dependencies among these units and organize them into higher order semantic blocks. Based on these structured representations, an importance aware routing mechanism adaptively dispatches different semantic blocks to different experts for customized modeling. Extensive experiments on 12 real world datasets demonstrate the effectiveness of SCPaT.

Segmentation of curvilinear anatomical structures in 3D medical images remains challenging due to complex topology, severe class imbalance, weak contrast, and large variations in structure morphology. While deep learning approaches for 3D curvilinear segmentation have been proposed, they are often tailored to specific anatomies or modalities, limiting generalization across clinical settings and leaving room for improvement. Recent generative models have shown the benefits of iterative prediction for structured segmentation tasks, yet diffusion-based methods suffer from computationally expensive sampling, hindering their use on high-resolution 3D volumes. We present 3D-CurvSegFlow, a flow matching-based model for 3D curvilinear structure segmentation. The model learns a continuous transformation from a simple source distribution to the target vascular representation, enabling progressive refinement of complex curvilinear geometries with efficient inference. We evaluate our method on Three public challenging datasets covering distinct anatomies and modalities: portal vein, cerebral vessel, and coronary arteries. Using a common architecture and training strategy across all tasks, our method outperforms general-purpose and vessel-specific approaches, with strong preservation of thin branches and vascular continuity. This work not only advances the state-of-the-art in 3D curvilinear segmentation but also opens new avenues for efficient, generalizable, and clinically applicable methods in medical image analysis.

Autonomous driving systems must operate under partial observability, where safety-critical objects may be occluded or visible only to neighboring connected vehicles. Vehicle-to-vehicle cooperation can reduce this uncertainty, but existing cooperative driving methods often compress multi-agent evidence into latent features or hidden multimodal states. As a result, they obscure which agent observed each object, whether the object is visible to the ego vehicle, and how conflicting evidence affects downstream decisions. We propose G-MARK, a grounded multi-agent reasoning framework that converts cooperative object-centric observations into explicit provenance-aware knowledge graphs (KGs). The resulting KGs preserve object hypotheses together with their source attribution, ego-versus-partner visibility, uncertainty, conflicts, spatial relations, and planning-relevant context. G-MARK then derives a shared feature representation from these KGs, enabling lightweight task heads to support object reasoning, motion prediction, control selection, and trajectory forecasting. Compared with the state-of-the-art baseline, GMARK improves occlusion reasoning accuracy by 42.2%, reduces control-selection error by 13.1%, and achieves comparable trajectory-planning accuracy with a 25.6x smaller structured communication payload. Our code is available at https://github.com/bhavyagupta98/g-mark.

End-to-end automated testing is increasingly used in web software development to ensure system quality and shorten response times during development. However, the true effectiveness of automated testing depends on many factors including execution time, stability of test results, and maintainability of the test suite as the application continues to evolve. In this paper, we evaluate the effectiveness of end-to-end automated testing using the Cypress framework for an open-source web application. We deployed the experiment with 27 test cases. The test's effectiveness is measured by execution speed, reliability, and maintainability. The experimental results show that the Cypress-based end-to-end test suite has short and stable execution times. It suits frequent runs during software development. The majority of test cases achieved consistent results across multiple runs, while flakiness only occurred in a few tests which involve complex interactive functions. Furthermore, the study highlights the impact of element locator strategies and Page Object Model (POM) architecture on test suite maintainability, demonstrating that resilient data-cy attributes significantly reduce maintenance overhead when UI changes occur.

Natural language code retrieval is a rapidly evolving task in computer science. However, the 1C:Enterprise ecosystem combines Russian syntax with highly domain-specific terminology, for which open datasets and specialized models have been virtually non-existent. We present a comprehensive pipeline for 1C code retrieval: an open benchmark of 3,413 real-world, PII-scrubbed query-code pairs, a reproducible evaluation harness, and a specialized bi-encoder. To overcome scarce labeled data, we fine-tune on 784,057 synthetic triplets generated by google/gemma-4-26B-A4B-it from public code repositories, using Matryoshka Representation Learning (MRL) and a privacy-aware tokenizer. Because the benchmark subsets differ in size, we report balanced-subset macro, query-weighted micro, and forum-only results. Our model reaches 0.5992 balanced macro nDCG@10, 0.5044 micro, and 0.4617 on forum, versus 0.4932 macro for the baseline architecture and 0.5404 for google/embeddinggemma-300m. Removing every benchmark example flagged by the conservative exact/13-gram overlap audit leaves 0.6011 balanced macro (0.5010 micro), indicating that detected train-benchmark overlap does not explain the headline result. MRL truncation to 256 dimensions preserves 99.9% of retrieval quality while reducing dense-index storage and exact similarity arithmetic by a factor of three.

Mixed-Integer Linear Programming (MILP) is a fundamental problem class in operations research and combinatorial optimization, with broad applications to industrial decision-making. Owing to their NP-hardness, however, modern solvers may struggle to find high-quality solutions for challenging MILP instances within practical time limits. Recent learning-based approaches seek to accelerate MILP solving by directly predicting high-quality solutions from static instance-level features, such as variable-constraint bipartite graphs. Yet accurate solution prediction from instance features alone is difficult, and these methods largely overlook the information revealed during the solver's search process. In this paper, we find that solutions produced at the early search stage of MILP solvers, which are computationally cheap to obtain, are often structurally close to the solutions found after full-budget search. Motivated by this observation, we propose a new solver-informed paradigm that shifts the learning target from variable assignment to early-to-final consistency: for each variable, we predict whether its early-stage assignment should persist in full-budget solutions. The predicted consistency naturally guides downstream search, for instance by fixing the assignments deemed consistent. At inference time, we further ensemble consistency predictions across multiple early-stage solutions to improve robustness. Experiments across four MILP benchmarks show our method improves prediction-guided search across diverse downstream pipelines. With Gurobi, our proposed method reduces the primal gap by 56.9% on average and closes it completely on combinatorial auction instances. Besides, we transferred the Gurobi-trained model zero-shot to SCIP without adaptation, achieving a 36.4% average gap reduction across benchmarks.

Artificial intelligence (AI) systems are increasingly used across domains to provide personalized information, recommendations, and decision support. However, in some contexts, AI-generated information may not be suitable for direct delivery to the final recipient. Instead, it may need to be interpreted, adapted, and communicated by a human who understands the recipient's needs, emotional state, and situational context. Human-AI interaction research has given less attention to situations in which a more knowledgeable human acts as an intermediary between an AI system and a less experienced or less informed recipient. We introduce the human-mediated AI guidance framework and explore it through Ready Together, an AI-supported family emergency preparedness system in which parents mediate AI-generated content for their children. The system is designed to provide personalized guidance and support parents in making emergency preparedness more interactive and understandable through guided activities and family-centered learning. The system design was informed by a qualitative, design-oriented research process involving semi-structured interviews and co-design activities. Findings identified challenges in family emergency preparedness, including difficulty discussing emergencies with children, uncertainty about providing appropriate explanations, and a preference for interactive learning activities. These findings informed the design of an interactive prototype, subsequently evaluated through a pilot study and a heuristic evaluation. Participants responded positively to the personalized recommendations and practical activities. Preliminary findings suggest that human-mediated AI guidance may support context-sensitive family preparedness while preserving parents' responsibility for interpreting, adapting, and communicating AI-generated information.

Multimodal sarcasm detection aims to identify sarcastic intent from multimodal content, where inconsistencies between literal meaning and contextual cues often signal irony. This task has attracted increasing research attention. However, accurate detection remains challenging due to instance-dependent modality contributions and misleading semantic consistency, where surface-level alignment masks underlying contradictory intent. Existing methods often rely on fixed fusion strategies and treat sarcasm as generic cross-modal mismatch, limiting their ability to capture subtle sarcasm cues and instance-specific modality interactions. To address these challenges, we propose a novel MSD framework that integrates Dynamic Gated Cross-Modal Fusion with Sarcastic-aware Contrastive Regularization (SaCR). Specifically, a bidirectional gated interaction module performs cross-modal feature filtering and adaptively calibrates textual and visual contributions at the instance level. A dynamic fusion gate further balances modality importance to generate more robust multimodal representations. Furthermore, SaCR is introduced as a label-aware contrastive regularization objective that encourages semantic consistency for non-sarcastic samples while suppressing misleading consistency in sarcastic cases. The proposed framework is trained end-to-end with a multi-objective learning strategy that jointly optimizes multimodal classification and auxiliary unimodal supervision. Extensive experiments on MMSD and MMSD2.0 demonstrate that the proposed method consistently outperforms strong baselines.

Existing approaches to anomalous behaviour log detection, such as Wazuh rely primarily on predefined detection rules, while statistical anomaly detection approaches such as OpenSearch identify deviations from previously observed behavioural patterns. Recent research has investigated LLMs for log anomaly detection because of their ability to interpret semantic and contextual information. However, LLM-based approaches can be affected by prompt construction, noisy log data, and reliance on generic datasets that may lack endpoint-specific authentication behaviours. To address these limitations, this study develops a standardised instruction-based LLM classification framework for detecting anomalous authentication behaviours, including borderline cases. A controlled cybersecurity testbed was developed to generate endpoint-specific authentication data, producing a curated dataset comprising normal, borderline, and anomalous behavioural scenarios. Three instruction-tuned LLMs, Meta Llama 3.1 8B Instruct, Qwen 2.5 7B Instruct, and GPT-OSS 20B, were evaluated against Wazuh rule-based detection and OpenSearch Anomaly Detection using a common ground-truth severity framework. Meta Llama 3.1 8B Instruct achieved the strongest overall end-to-end detection performance, with an accuracy of 89.3%, recall of 88.2%, F1-score of 91.8%, and false negative rate of 11.8%. In comparison, Wazuh achieved an accuracy of 52.0% and false negative rate of 68.6%, while OpenSearch achieved an accuracy of 49.3% and false negative rate of 74.5%. Meta Llama also detected 80% of the borderline anomalous scenarios, compared with 20% for Wazuh and 15% for OpenSearch. Qwen achieved lower overall detection performance than Meta Llama but recorded the lowest average inference latency and 100% structured-response validity. GPT-OSS demonstrated strong classification performance when valid responses were produced.

Public benchmarks are important measures of Automatic Speech Recognition (ASR) model capabilities. However, by nature of being public, there is risk of models being optimized for these benchmarks in ways that do not generalize well to real-world data. We present a methodology for quantifying benchmark optimization, focusing on cases where the audio underdetermines the reference transcript. We identify three families of behavioral probes that reveal models' capabilities of reproducing benchmark reference spans despite underdetermined audio: reference disagreement, masked-number recovery, and orthographic switching. We find that the highest-scoring open source models output verbatim reference transcript spans even when the relevant audio is contradictory, masked, or ambiguous. Using a variety of mechanistic probes, we show that models respond to narrow acoustic cues to override the faithful representation of the audio in favor of a benchmark-optimized policy. We show the benchmark-optimized behavior can be causally manipulated via low-rank linear steering or simply appending audio to the end of a segment in some cases. Overall, our results indicate that high-performing models exhibit benchmark-conditioned behaviors that can inflate benchmark performance without reflecting improved general-purpose transcription ability.

Public cardiac cohorts annotate different subsets of the heart, so shapes from separate sources cannot be pooled without shared correspondence. Among released cardiac shape resources, none we identified carries the atrial appendage, pulmonary veins, and caval stumps as separate blocks in one mesh. Completion benchmarks also compare deep models against a least-squares projection onto shape modes, not the conditional estimator the same fitted model implies. We release an eleven- structure cardiac computed-tomography (CT) statistical shape model, built from 383 automatically labelled cases in 11 571-vertex correspondence, and compare completion estimators under one frozen internal split and endpoint. On a 76-case internal list held out from fitting, a closed-form conditional-Gaussian estimator reconstructed the missing non-chamber structures at 3.717 mm mean per-vertex error, averaged equally over one, three, five, and nine observed structures. A five-refit mask-conditioned graph variational autoencoder reached 5.248 mm and nearest-neighbour retrieval 8.931 mm. The paired difference was 1.531 mm (95% confidence interval 1.384 to 1.711), and the ordering held in a raw-coordinate sensitivity arm. Expert manual labels exist for 58 external CT cases, but our registered reference is close enough to score only five structures. There the closed-form estimator again had lower average surface distance, 95th-percentile Hausdorff distance, and Chamfer error for both completed atria. On a second public benchmark of 20 cases the reference was close enough for three of four completed structures, and the same ordering held there. Four structures have no expert reference. The released model and its completion operator support cohort-unification research on aligned CT, not clinical use.

Under the US Lead and Copper Rule Revisions, a utility may determine a service line's material with a predictive model instead of inspecting it. New York State publishes, per address, which method was used. Almost no address carries both a model classification and a physical verification, so the check is between populations within a utility rather than paired addresses. We screen all 153 New York localities that classified at least 100 addresses this way. Seventy-five (49%), covering 125,990 addresses or 57% of those screened, record one value. Zero variance alone is not misconduct: 68 of the 75 match their own verification or have too little to test. Seven are contradicted by their own crews, six beyond any sampling explanation. Five are boroughs of New York City, which file as one system; one is East Rochester, 550 km away. New York City is the largest case: a predictive model is the recorded basis for 43,215 addresses, and on all of them the recorded material is "Known Other". The city records "Unknown" on 121,779 addresses, 1,880 already excavated, and lead on 120,692. In the model bucket both counts are zero, and the 95% upper bound on the rate is 0.0085%. Across the rest of New York the same method records lead or the hedge "Unknown but could be lead" on 12.21% of 176,888 addresses, a comparison whose weaknesses we report. The model-cleared population is newer, median year built 1984 against 1930, and construction era accounts for about a third of the gap and not the rest: holding era fixed, records-based classification finds lead at 4.3-31.9%, physical verification at 1.5-14.5%, the model in no era. Six era-aware estimators place the expected lead lines among them at 1,150-1,450. Two findings need no comparison: 7,782 of these addresses are in pre-1940 buildings, and the archived 2025 snapshot shows the public-side determination was copied from a customer-side model output.

A lot of prior work addressed key-value (KV) cache selection and compression by sparse attention to enable long-context inference for transformer language models without excessive hardware budgets. We provide a new method for fine-tuning models with sparse attention. It works for any KV cache policy, runs on a moderate hardware budget (e.g., a single Nvidia A100 GPU with 40 GB RAM), and allows the model to co-adapt with the policy, often outperforming models trained with exact attention (sequence parallelism). We also provide an efficient implementation of H2O sparse attention (the leading policy in our experiments) with dedicated scaled dot product attention kernel support. KeysAndValues (https://github.com/awslabs/keys_values), a new open source library for long-context inference and fine-tuning, provides easy-to-use and performant code for all methods discussed here.

Network topology inference from graph signals is central to graph signal processing with applications in neuroscience, sensor, and social networks. In practice, target-domain samples are scarce while heterogeneous source-domain data are abundant. Fusing these sources is challenging: Euclidean averaging works for homogeneous sources but degrades sharply as inter-source divergence grows, collapsing distinct geometries into an inflated, biased consensus. We exploit the Wasserstein metric's distribution-preserving properties to counter heterogeneity while preserving each source's intrinsic geometry. We propose MS-WDRO, a multi-source Wasserstein distributionally robust graph learning framework that fuses heterogeneous sources via their weighted Wasserstein barycenter, a geometrically principled nominal distribution, then builds an ambiguity ball around it to hedge residual uncertainty. Minimizing worst-case risk yields a tractable regularized Laplacian estimator solved efficiently via a provably convergent ADMM scheme. We establish non-asymptotic guarantees: a finite-sample concentration bound for the empirical barycenter, a pooling bias lower bound proving naive aggregation is suboptimal, and an out-of-sample excess risk bound decaying at a parametric rate with only logarithmic dependence on source count. To calibrate hyperparameters governing robustness, sparsity, and source fusion, we unroll the solver into a differentiable architecture trained end-to-end, achieving data-adaptive calibration beyond cross-validation while retaining interpretability. Experiments on synthetic benchmarks and the multi-site ABIDE~I neuroimaging dataset show MS-WDRO consistently outperforms seven baselines in graph recovery, sample efficiency, and downstream diagnostic utility, with the largest gains in the sample-scarce regime.

Probability estimation over large alphabets under log loss is a well-studied problem, with celebrated methods such as the Good-Turing estimator. We introduce and study a new Bayesian estimator with four notable properties. First, its construction is exceptionally simple: multiply independent uniform draws from the probability simplex coordinate-wise and renormalize. Depth is the only structural parameter, and averaging over depths eliminates the need to tune it. Second, the regret of the resulting mixture, the excess code length it pays relative to a code that knows the source, admits an explicit and efficiently computable expression. Third, despite its simplicity and lack of tuned constants, the estimator is competitive across a diverse set of synthetic and real-text benchmarks with substantially more specialized methods, including Good-Turing. Fourth, the tractability of its regret allows us to identify scaling laws in data, alphabet size, and depth. For Zipf targets with exponent above one, the regret has a simple reading as long as the sample reveals only a small fraction of the alphabet. It closely matches the description length of the set of discovered symbols, at one bit of code per bit of description, plus a further cost per symbol. The data exponent is therefore the rate at which new symbols are discovered.

This paper presents a Bayesian control framework that integrates spike-based dynamics with probabilistic inference for adaptive control. Bayesian inference is widely regarded as a core computational principle of brain function, providing a normative framework for perception, decision-making, and learning under uncertainty. By combining a biologically inspired spiking neural model with Bayesian inference principles, we propose a brain-like control algorithm capable of operating in uncertain environments. We use the mountain car parking problem as a benchmark with non-linear dynamics. Our results demonstrate that the proposed controller can successfully update states in real time and generate goal-directed action plans through spike-driven dynamics. The results highlight the proposed model's potential as a bridge between computational neuroscience and probabilistic control theory.

Accurately ranking active ligands for a target protein pocket from massive chemical libraries remains a central challenge in virtual screening. DrugCLIP and its recent extensions substantially accelerate this process by encoding protein pockets and molecules into a shared embedding space. Despite this progress, further performance improvements typically require retraining the entire model, incurring substantial computational overhead and making target-specific customization inefficient. In this work, we formulate the specialization of pretrained virtual screening models to individual pockets as a test-time adaptation problem and propose PETA, a parameter-efficient framework that directly adapts pretrained model at test time. Given a target pocket, PETA constructs pocket-specific negatives through molecular diffusion and chemical validity filtering, and further moves them toward the reference ligand retrieved from structural databases via embedding-space mixup to create more challenging ranking tasks. A ranking objective then places greater emphasis on suppressing high-scoring invalid candidates that could contaminate the top-ranked screening results, providing structured supervision for lightweight adaptation. Experiments across diverse benchmarks demonstrate that this lightweight, pocket-specific adaptation outperforms both pretrained and fully retrained baselines while updating only the LayerNorm parameters, which account for approximately $0.03\%$ of the full model.

In fixed-confidence best-arm identification, proofs often use a union bound across the competing arms. From a multiple-testing point of view this can look puzzling: if the best arm is unique, only one hypothesis of the form ``arm $i$ is best'' can be true. Why then should there be a Bonferroni-type factor of $K-1$? The answer is that there are two natural ways to orient the hypotheses. In one orientation, best-arm identification is literally a strong familywise-error-rate (FWER) problem with $K-1$ true nulls. In the opposite orientation, exactly one null is true, but a pairwise implementation can falsely reject that one null through any of $K-1$ comparisons. Thus the multiplicity has not disappeared; it just pops up in different places. This note makes the equivalence explicit in the terminology of both communities.

AI agents can execute scientific analyses, but an analytic output becomes a defensible claim only after alternatives are weighed and the claim is limited to what the evidence supports. Agents may reproduce failures including selective analysis, premature declarations of success and optimization of imperfect criteria. We present Brain Researcher, an agentic research harness operating in a neuroimaging researcher's computational environment under rules for admissible analyses, required checks and claim scope. In benchmarks, Brain Researcher increased first-choice tool-selection accuracy across seven models by 70.2 percentage points (23.3% without it versus 93.6% with it) and verifiable grounding from 4.6% to 22.0%. In collaborator-led and self-evolving studies, multiverse analyses exposed analytic-choice sensitivity, and scientific review classified claims as accepted, qualified, revised, blocked, rejected or deferred. By linking decisions to evidence and provenance, Brain Researcher embeds methodological judgment within the workflow, not after it.

Agent Skills extend LLM agents with reusable instruction packages that may also include scripts, resources, and service configuration. This creates a direct distribution channel for malicious behavior, yet existing malicious-Skill datasets are fragmented across sources, artifact formats, evidence regimes, and benign coverage; duplicated and structurally related content further complicates direct aggregation and evaluation. We present MaliciousSkillBench, a comprehensive benchmark for malicious Agent Skill detection. We consolidate 13 public sources, 11 of which contribute Core malicious artifacts, and reduce 8,414 raw malicious records to 7,539 normalized-unique identities in 4,588 operational structural families. After conservative cross-label conflict exclusion, the primary benchmark contains 9,740 Skills: 7,505 malicious and 2,235 benign. To characterize its coverage, we harmonize 11 attack categories for 4,983 malicious identities with supported source-native mappings and find substantial differences in threat composition across sources. We then evaluate three learned text detectors and three off-the-shelf Skill scanners. Learned detectors achieve 0.882-0.932 Random Macro-F1 but only 0.653-0.665 under Source-Disjoint evaluation; the strongest word TF-IDF SVM scores 0.932/0.916/0.665 on Random/structural-disjoint/Source-Disjoint while retaining 95.6% malicious recall but producing 62.4% benign FPR on held-out sources. Off-the-shelf scanners occupy different but also unsatisfactory operating regimes, reducing false positives only at the cost of sharply lower malicious recall. Together, these results show that reliable malicious-Skill detection requires both broader cross-source benchmark coverage and evaluation that jointly measures attack detection and benign over-flagging.

Machine learning (ML) is widely used for the development of ocean colour algorithms, but most studies focus on model parameter training and hyperparameter tuning. The optimisation of the data that feeds the models - i.e., Feature Engineering (FE) - is not fully explored. We assess the impact of FE in ocean colour machine learning models and we propose an optimisation framework that includes seven sequenced levels of data transformation: i. band choice, ii. log scaling, iii. spectral shape normalisation, iv. index extraction, v. principal component analysis, vi. feature scaling, and vii. zero-to-one scaling. We demonstrate the application for Multi-layer perceptron, Support Vector Machines, and eXtreme Gradient Boosting Trees on Sentinel-3 OLCI observations in the Norwegian coastal waters. The models are trained to estimate Chlorophyll-a concentration [Chl-a] and Secchi disk depth (Zsd). Results show that accuracy is highly variable among FE found in six studies using Sentinel-3 OLCI and the ones that we optimise. The R range from 0.01 to 0.55 for [Chl-a] and from 0.15 to 0.68 for Zsd, where the optimised FE shows the top results. The ML models with optimised FE could also improve by two times the R and reduce up to 63% of the mean absolute error when compared to CHL_OC4ME and CHL_NN standard algorithms. Nevertheless, no common optimised FE is found for all target variables and ML models, suggesting that FE optimisation is necessary for each application. Therefore, our proposed framework can be key for improving the accuracy of water quality monitoring in coastal waters.

Where does the novelty a base language model produces with no task come from, and what can an LLM judge of a long stream actually see? We dismantle a cognitively inspired generation loop over 24 conditions on three base models. Most of its effect lives in one operation: a new subject injected every few hundred tokens (an interruption) into a stream whose literal repetition is damped (habituation). We judge windows of generated text only, with the premise as the unit (n=10) and a judge measured for repeatability, against a second judge family and against human readers. Under that protocol the interruption raises judged surprise by 1.2 to 1.4 points and connection by 0.8 over habituation alone. A connective that asks for continuity hurts; a bare paragraph break adds nothing detectable on fresh text; a reset context does at least as well as a kept one; and a pre-registered replication on new premises confirms the primary contrast. Three things the window judge could not see changed the first version of this study, and we think they are of general use. The judge scores the experimenter's injected sentence as the model's own. A fixed rotation of injected sentences makes the model replay its earlier segments from beyond the judge's horizon, and the judge scores the replay as surprise and connection (65-80% of post-interruption windows at periods 150-300). And the local gains do not compose: no arm produces an integrated document. The salience monitor, the in-loop judge, memory across interruptions and a judge-gated Review run with a gate that opens add nothing. On a problem with a verifier (online bin packing), the interruption multiplies valid, distinct candidate heuristics three- to fourfold without raising the quality of the best. We report an evaluation protocol for long generation and a controlled characterization of a simple intervention, not a mechanism of creativity.

How to efficiently finetune robot policies to learn new tasks on the fly? State of the art robotic manipulation policies are based on behaviour cloning of large vision-language-action (VLA) models with billions of parameters on huge teleoperation datasets. While this simple approach has enabled significant advances for robotic manipulation, finetuning of VLA policies for learning new tasks still remains an open problem. In particular, collecting teleoperation datasets requires hundreds of hours of expensive human labour and the alternative, reinforcement learning (RL), can be notoriously sample-inefficient especially for long-horizon tasks. In addition, RL with VLAs imposes several challenges due to the model's size and architectural design. In this work, we propose EXIMO, an efficient algorithm for finetuning of VLA policies. EXIMO operates in three stages: explore, imitate, and optimize. During the explore phase, EXIMO equips the VLA with a vision language model (VLM) that acts as a planner. The VLM thinks and breaks down challenging long-horizon problems into shorter ones for the VLA. The VLM, together with the VLA, is used to collect an orchestrated dataset on new tasks. During the imitate phase, the VLA is finetuned with the orchestrated data. Finally, during the optimize stage, we use residual off-policy RL to further finetune the policy. In our experiments, we ablate all three stages of EXIMO and show that it outperforms existing approaches significantly in terms of sample-efficiency and final performance.

Test-Time Adaptation (TTA) methods aim to bridge the domain gap between the source and target domains. However, traditional TTA methods become ineffective when the label distribution shift occurs, a challenge commonly referred to as an open-world scenario. In this paper, we introduce a new method named Reliable Neural Collapse approximation (ReNC) for Open-World Test-Time Adaptation (OWTTA). Specifically, we leverage neural collapse as a structural prior for reliable target-domain adaptation. Guided by this prior, we justify that the pre-trained classifier weights can serve as the prototypes of the source domain. By measuring the similarity between samples and prototypes, we filter out the Out-Of-Distribution~(OOD) samples for reliable updates. Furthermore, we propose a neural collapse approximation mechanism to refine these prototypes, ensuring they can gradually adapt to the target domain while maintaining the neural collapse structure. Extensive experiments on several open-world benchmarks demonstrate the superiority of the proposed method. Our empirical analysis suggests that ReNC better preserves NC-related properties in the target domain, providing useful evidence for explaining reliable OWTTA and offering new insights for model design. Code is available at https://github.com/JiaqiLin-AI/ReNC.

The entire ecosystem of open-source language models effectively relies on a single platform. What if this platform was forced to shut down tomorrow? Implementing and maintaining efficient model definitions and translating them between different training and inference regimes is a resource-heavy task that severely limits model efficiency and portability, hindering both scaling and deployment. Here, we present Axon, a strongly typed domain-specific language with Haskell-like syntax, that enables a write-once, run everywhere paradigm for LLM architectures. By basing collaboration on a language specification rather than a specific framework's vision, Axon fosters open cooperation and empowers researchers to implement highly specialized architectures without giving up optimization infrastructure or accepting deployment lock-in. Axon allows for concise, auditable specifications that can be automatically compiled to standalone implementations for leading frameworks: PyTorch, PyTorch with Triton, JAX, MLX and vLLM. In 467 inference benchmarking experiments on models ranging from 135M to 32B parameters, we demonstrate median speedups of 7% on PyTorch, 12% on PyTorch with Triton, 91% on JAX, and 107% on MLX, compared to the reference implementations from Transformers. When deployed as native vLLM architectures with PagedAttention and KV-cache, Axon models achieve a 58% median speedup over Transformers implementations.

Streaming systems that maintain a pool of expert models must repeatedly decide whether to reuse an existing expert for arriving data, spawn a new one, or defer. We present a decision layer that makes all three outcomes statistically meaningful. Reuse and spawn are posed as one-sided sequential hypotheses on a conditional (mechanism-level) discrepancy, separated by an indifference zone; defer is exactly the state in which neither betting e-process has accumulated sufficient evidence. We prove finite-time anytime validity for the observable surrogate discrepancy of a predictable discriminator sequence, and an unconditional one-sided transfer to the population quantity in which each side's slack is the excess risk of a single discriminator; an empirically observed downward-bias regularity makes the spawn side exactly conservative. Recency without sacrificing the guarantee is obtained by a restarted e-detector: a bank of unwindowed betting supermartingales at geometrically spaced restart times (O(log t) memory), with the error budget spent over restart instances, which preserves lifetime anytime validity; spending over expert-creation order likewise controls multiplicity for unboundedly many experts. On synthetic multi-concept streams, Electricity, Covertype, and the recurrence-heavy INSECTS benchmark, the instance-accounted restarted bank achieves zero false spawns and zero false reuses after switches and matches or exceeds the retired windowed heuristic (INSECTS-reoccurring accuracy 0.675), making the deployed algorithm and the guaranteed algorithm one and the same.

Streaming systems that maintain a pool of expert models must repeatedly decide whether to reuse an existing expert for arriving data, spawn a new one, or defer. We present a decision layer that makes all three outcomes statistically meaningful. Reuse and spawn are posed as one-sided sequential hypotheses on a conditional (mechanism-level) discrepancy, separated by an indifference zone; defer is exactly the state in which neither betting e-process has accumulated sufficient evidence. We prove finite-time anytime validity for the observable surrogate discrepancy of a predictable discriminator sequence, and an unconditional one-sided transfer to the population quantity in which each side's slack is the excess risk of a single discriminator; an empirically observed downward-bias regularity makes the spawn side exactly conservative. Recency without sacrificing the guarantee is obtained by a restarted e-detector: a bank of unwindowed betting supermartingales at geometrically spaced restart times (O(log t) memory), with the error budget spent over restart instances, which preserves lifetime anytime validity; spending over expert-creation order likewise controls multiplicity for unboundedly many experts. On synthetic multi-concept streams, Electricity, Covertype, and the recurrence-heavy INSECTS benchmark, the instance-accounted restarted bank achieves zero false spawns and zero false reuses after switches and matches or exceeds the retired windowed heuristic (INSECTS-reoccurring accuracy 0.675), making the deployed algorithm and the guaranteed algorithm one and the same.

Deep learning models remain vulnerable to real-world input perturbations, especially when multiple corruptions co-occur in the same input (e.g., brightness shifts and motion blur). Compositional testing reveals these interaction effects but introduces two challenges: combinatorial growth of the perturbation space as dimensions and severity levels increase, and uneven diagnostic value-many combinations yield unrealistically degraded inputs with limited practical relevance. We present TESTNAV, 1 a Pareto-guided robustness testing framework for efficiently exploring discrete, compositional perturbation spaces when only a limited number of perturbation configurations can be evaluated. TESTNAV prioritises severe yet realistic failures by formulating robustness testing as bi-objective optimisation: maximise performance degradation while preserving input fidelity measured by modality-specific metrics (e.g., SSIM and KID for vision; chrF and BERT-F1 for language and code). It uses NSGA-II to approximate the bi-objective Pareto front. Across four benchmarks spanning vision, natural language, and code generation, TESTNAV recovers Pareto fronts up to 2.15x faster than search-based baselines, using 35.8%-89.3% of the discrete perturbation space defined by four perturbation dimensions with six levels each.

Radio frequency (RF) fingerprinting authenticates wireless devices from hardware-induced I/Q impairments, typically with deep learning feature extractors that are accurate but opaque, limiting their use in security critical settings. We propose Polar Monotonic Kolmogorov-Arnold Networks (Polar MKAN), a block partitioned monotonic encoder on polar inputs in which each latent dimension depends exclusively on magnitude or phase, yielding channel separation and monotone responses by construction. On a synthetic gain and carrier frequency offset (CFO) benchmark, Polar MKAN reaches 57.2 percent DCI Disentanglement versus at most 12.9 percent for unpartitioned baselines. We further evaluate the detection accuracy trade off on real data and the sensitivity to blind CFO compensation.

LLM agents learn by interacting with environments, yet these environments are hand-built and static: blind to an agent's weaknesses, and quickly left behind as it improves. While recent environment generation methods attempt to address this, they require domain-specific pipelines, rely on expensive or unreliable verifiers, and still produce static environments. To alleviate the engineering burden of rebuilding environments from scratch, we propose Environment Harness (EnvHarness), a programmable layer of plug-in components that wraps a static environment to reshape its behavior without modifying the underlying logic. Operating through standard interfaces, EnvHarness applies across diverse domains while ensuring every reshaped environment retains its original verifier. To automate this process, we introduce EnvRigger, which treats the target policy as a black box, observing its execution trajectories to synthesize EnvHarness components targeting diagnosed flaws, and validating them via fresh rollouts. Across five benchmarks in four domains, EnvHarness outperforms both original environments and domain-specific environment generation pipelines, achieving up to a 9.0-point improvement on held-out instances with 9.8% fewer execution steps. Furthermore, EnvHarness provides a superior optimization signal for reinforcement learning, enabling continuous, targeted co-evolution of the policy and its environment.

This document describes the application of a first order linearised nonlinear repeated measurements approach to the analysis of battery cell ageing profiles generated under controlled conditions in a laboratory. The primary advantage of the model is it reflects the obvious structure in the data. Consequently, it is a two-component of variance model: variation within ageing profiles (measurement noise) and variation among ageing profiles (test-to-test or cell-to-cell) variation. Novel regularised iterative generalised least squares parameter identification schemes, with optimal hyper-parameter re-estimation, are used to identify the hierarchical nonlinear model. The training data comprised $SoH$ profiles for 10 cells aged at various constant discharge and charge current cycles at a fixed chamber environmental temperature of 25 [$^\circ$C]. Each cell $SoH$ profile is modelled using a simple power law expression, whereas the variation in ageing parameters is modelled using a single knot cubic B-spline. $SoH$ is accurately predicted to $\pm 0.191\%$ for $SOH \in [0,20]$.

Patients often submit short, underspecified queries to healthcare chatbots that lack the patient-specific information needed to determine an appropriate response. Although these queries may be linguistically clear, they can support multiple plausible answers depending on undisclosed factors such as symptoms, diagnoses, medications, allergies, or dietary restrictions. A language model answering such a query directly may therefore rely on unsupported assumptions about the patient. We introduce a knowledge-guided agentic framework for mitigating patient-context ambiguity before final response generation. The framework operates between the patient and an otherwise unchanged downstream language model. It interprets the initial query, uses a task-specific knowledge graph to construct a set of plausible hypotheses, identifies the missing patient-context variables needed to distinguish among them, and asks targeted follow-up questions. The original query and the acquired context are then combined into a clarified prompt for the downstream model. We evaluated the framework across five language models using two controlled ambiguity-mitigation benchmarks: diagnosis retrieval from 1,034 symptom queries with clinically relevant evidence systematically masked, and dietary-safety classification from 487 queries with decisive health context omitted. The framework was compared with direct answering of the underspecified query and with rephrasing the same query without acquiring new patient information. In diagnosis retrieval, it increased overall exact Top-1 accuracy by at least 57.1 percentage points and selective exact Recall@5 by at least 77.7 percentage points across the five evaluated models compared with direct prompting. In dietary-safety classification, it improved accuracy across all five models and achieved the highest Matthews correlation coefficient for four...

Self-supervised learning (SSL) has driven substantial progress in audio representation learning, though existing methods have increasingly relied on elaborate pre-training recipes to reach competitive performance. A markedly different pre-training philosophy underpins the most influential progress in language modeling and, more recently, in visual representation learning: rather than train encoders as static feature extractors, models are trained to predict the next element, a discrete token or a continuous embedding, from the preceding context. Autoregressive prediction thereby provides a unified pre-training interface that transfers across modalities, compelling the model to learn the underlying data distribution. We ask whether such a simple causal paradigm can yield strong audio learners, given that audio's temporal structure makes autoregressive prediction of patch embeddings a natural fit. We introduce NAPE (Next-Audio-Patch-Embedding prediction), a self-supervised framework in which a causal Transformer predicts each next patch embedding of a log-mel spectrogram from the previous ones, using causal masking and stop-gradient as its sole training signal. The design is intentionally minimalist, avoiding reconstruction decoders, acoustic tokenizers, student-teacher setups, and auxiliary regularization losses. Across six audio and speech benchmarks, NAPE achieves state-of-the-art fine-tuning performance on several tasks, scales consistently across encoder sizes, and yields strong linear-probing results. NAPE also produces structured attention patterns without explicit supervision.

Customer-service LLM agents must follow organizational policy when acting on a user's behalf. Compliance failures arise from either forbidden actions, such as granting an ineligible change, or omitted procedural requirements, such as identification or confirmation. Runtime safeguards can intervene on risky actions, but action-local checks do not guide an agent through a multi-step procedure. Workflow-following systems support prescribed process execution, but primarily target workflow completion rather than safeguarding agent behavior. PolicyGuide instead compiles each domain policy into a workflow graph and invokes a proactive verifier at user-turn boundaries. From persisted graph state, the verifier reconciles open requests and returns step-specific remediation along a policy-compliant path. Across the $τ^2$-bench airline, retail, and telecom domains with a GPT-5.4 agent and verifier, PolicyGuide raises mean $\mathrm{Pass}^4$ from $0.42$ to $0.62$, with the largest gain on telecom ($0.19$ to $0.61$), the most workflow-structured domain. The same workflows transfer to Claude Sonnet 4.6 and Gemini 2.5 Pro agents. Complementary evaluations find the lowest observed attack-success rate under adversarial users and the strongest procedural compliance in an author-designed workflow-level validation.

Generalizable dynamic graph anomaly detection (DGAD) enables pretrained detectors to identify anomalies in unseen target domains without costly retraining. However, existing methods often fail for two reasons. First, they mainly rely on domain-agnostic patterns and miss domain-specific patterns that keep evolving. Second, they assume access to the full target domain data, whereas in more practical online test-time adaptation settings, target data arrive sequentially in unlabeled chunks. To address these limitations, we formulate online test-time adaptation for generalizable DGAD and propose OTTA-DGAD. OTTA-DGAD first extracts dynamic prototypes, i.e., evolving representations of normal and anomalous patterns, from temporal ego-graphs and stores them in a memory buffer. The buffer selectively retains general patterns shared across the source domains used for pretraining while incorporating new patterns from the target domain. An anomaly scorer then compares incoming edge representations against these prototypes to identify both general and domain-specific anomalies. During adaptation, OTTA-DGAD updates the memory buffer using reliable pseudo-labels identified through confidence-based detection. It further enriches each target chunk with relevant representations retained from previous chunks, compensating for information loss resulting from the sequential arrival of data. Extensive experiments under strict test-then-adapt OTTA settings demonstrate state-of-the-art performance on ten real-world datasets from diverse domains.

For AI agents to be useful beyond simple chat, they must hold sensitive user context such as calendars, credentials, health records, and financial data. We study whether the mere presence of such secrets in a model's context window introduces hidden correlations into the model's benign outputs, allowing reconstruction even when the model correctly refuses direct extraction. We further study whether an adversary can actively engineer prompts that amplify this effect, using the model as a covert carrier to transmit secrets through seemingly innocuous text. In both cases, this limited leakage is exploited using a novel adaptive attack that assumes black-box access to the underlying model. In controlled experiments across eight proprietary models, we find that 2-digit in-context secrets are reconstructed with near-perfect accuracy and 4-digit secrets at 82\% exact match, all from outputs the model produces in response to ordinary, non-adversarial requests. We observe that more capable models leak more: stronger instruction-following amplifies sensitivity to in-context secrets, suggesting leakage is a byproduct of capability as opposed to a patchable bug. We show this leakage enables two practical attacks: (1) a trained classifier that infers semantic predicates about user memories (e.g., health conditions, financial events) from routine natural-language outputs, and (2) an RL-trained adversary that extracts full Social Security Numbers from a production-style agent.

Large language model agents have made substantial progress in code generation, yet most existing systems assume a predefined repository architecture. This assumption does not hold in zero-to-all code generation, where an agent must construct an entire software project directly from natural-language requirements while maintaining a modular repository architecture throughout development. We present Repo0, a continuous structural evolution framework for zero-to-all code generation. Repo0 maintains an explicit architectural state instantiated as a Dual-Directed-Acyclic-Graph (Dual-DAG), consisting of a requirement-level DAG, a component-level DAG, and their alignment relation. Starting from natural-language requirements, it iteratively evolves component boundaries through structural actions guided by modularity metrics until structural convergence, after which the converged architecture guides test-driven development code generation. We evaluate Repo0 on six real-world repositories from RepoCraft using GPT-5 mini and DeepSeek V3.2. Repo0 achieves the highest Functionality Coverage and Pass Rate across all settings. Compared with RPG, the strongest repository-planning baseline, Repo0 improves Functionality Coverage by up to 20.08 percentage points and Pass Rate by up to 29.74 percentage points. Ablation and structural-evolution analyses further demonstrate the importance of the Dual-DAG architectural state, modularity-guided structural evolution, and explicit structural convergence.

The VDM formalism can ensure that models are internally consistent. Potential inconsistencies are highlighted by assertions called proof obligations. This paper is an update to the work described in [1], regarding proof obligation generation for VDM operations. We present the latest work which improves the approach to loop invariants, adds loop variants, improves proof obligations for operation calls, including implicitly declared operations and specification statements, and adds recursive measures for operations. The new features are illustrated with examples.

Agentic reinforcement learning (RL) has become a critical stage in the post-training of large language models. Existing critic-free, group-relative methods estimate policy advantages from multiple rollouts, avoiding the substantial memory overhead of conventional proximal policy optimization (PPO) and achieving strong performance on long-horizon interactive tasks. Despite their success, recent studies revealed three limitations: (1) Lack explicit value generalization and effective temporal credit assignment; (2) Suffer from potential advantage collapse in long-horizon complex tasks; (3) Require a costly trade-off between sampling budget and policy performance. In this work, we propose Single-rollout Autoregressive Policy Optimization (SAPO), a low-memory and compute-efficient framework in which the policy and value functions share a single autoregressive backbone. SAPO exploits the autoregressive structure of LLMs to produce policy and value predictions at distinct causal boundaries with shared parameters, while independently optimizing the PPO objectives and auxiliary on-policy SARSA objectives. To robustly estimate the contribution of each turn, we further introduce a trajectory-level generalized advantage estimator that combines lambda-returns with batch normalization. Experiments across ALFWorld and WebShop with Qwen2.5-1.5B/7B show that SAPO trains stably and outperforms PPO and GRPO by mean +15.1 and +12.1 percentage points, respectively, while eliminating the memory cost of a separate critic model and reducing per-iteration runtime by 33.2% over PPO.

Spec Driven Development SDD has consolidated the idea that the specification rather than the code should be the primary artefact governing AI assisted work. Tools such as GitHub Spec Kit, and proposals such as Constitutional SDD, have formalised this principle in the software domain, while the executable data-contracts literature has extended it to schema and quality enforcement at run time. Nevertheless, the treatment of the specification delta OpenSpec's core idea that every change should produce a reviewable increment of requirements as the unit of change in data platforms remains empirically unexplored, even though many data-platform changes are contractual (new datasets, service-level agreements, metric semantics, access policies) rather than purely code changes. This work formalises the spec-delta concept, proposes a taxonomy of data platform changes according to their suitability for incremental specification, and defines a controlled experiment comparing a spec-delta-driven workflow against a conventional code pull-request workflow without a delta. The response variables are discovery to deployment time, the density of defects reaching the Silver and Gold lakehouse layers, cross-tool metric divergence, and reviewer cognitive load measured with NASA TLX. The paper explicitly reserves a demonstration-and-laboratory section for instantiation on a real lakehouse environment. The contribution is not a tool but reproducible evidence and an applicability guide that helps to avoid the up front over specification antipattern.

Probabilistic shielding is a technique for safe reinforcement learning (RL). Typically, a static observer -- called the shield -- constrains the learning agent's actions to those for which acting safely remains feasible. Traditionally, the shield is computed from the transition probabilities of the underlying Markov decision process (MDP). Thus, this technique is not applicable when the MDP model is not given a priori, which, unfortunately, is the case in typical RL applications. In this paper, we study the problem of computing a shield in the setting where the transition graph of the MDP is known, but the transition probabilities are unknown. Our approach integrates probabilistic shielding with online model learning: as the RL agent explores the environment, we estimate the transition probabilities. From this estimate, we compute a shield. While the shield may be conservative initially, it adapts as the model estimate becomes more precise. Thus, the shield improves in tandem with the RL agent. This paradigm of adaptive probabilistic shielding raises a number of challenges, such as when to recompute the shield and how to balance between exploration and safety during learning. We empirically evaluate multiple variants of this paradigm across several environments.

Causal Bayesian networks (CBNs) and structural causal models (SCMs) are the dominant frameworks for graphical causal reasoning, but they cannot adequately represent all real-world causal systems. In particular, systems at equilibrium---where feedback mechanisms create cyclic causal dependencies---can exhibit causal semantics that are fundamentally incompatible with these frameworks: different interventions that enforce the same variable value may have different effects, rendering the standard ``perfect intervention'' do($X = x$) ambiguous. We propose bipartite graphical causal models (BGCMs), in which the structure of a system of equations is encoded by a bipartite graph with variable and equation nodes. In this framework, a hard intervention do($f_j : X_v = ξ_v$) specifies which equation is replaced, which variable is targeted, and at what value---resolving the ambiguity of the standard notion. We demonstrate, through a detailed case study of a physical system, that this representation naturally corresponds to distinct real-world interventions. We formulate a Markov property in terms of a new graphical separation criterion (B-separation) that exploits the functional determinism inherent in the equations, and we extend it to settings with non-random inputs. We show how this gives rise to a do-calculus for reasoning about domain invariances. BGCMs strictly generalize CBNs and SCMs while retaining the ability to perform graphical causal reasoning.

Medical image captioning is a technique that accelerates early-stage diagnostic workflows and enhances the interpretability of medical diagnostic AI systems. However, unlike general image captioning, clinically reliable captioning remains challenging due to grayscale-based modalities, subtle anatomical cues, specialized medical phrasing, and variations in data quality. Despite recent advances in large vision-language models, fluent outputs do not necessarily guarantee sufficient alignment with clinical concept spaces or evaluation criteria. To address this issue, we propose a framework that strengthens clinical alignment by separating and enhancing training-time alignment and inference-time alignment. We build a medical image captioning pipeline that integrates single/dual vision encoders based on BioMedCLIP and SigLIP2, a Q-Former, and a LLaMA-based decoder, and examine the contribution of auxiliary learning for UMLS concept/type prediction. At inference, we apply single-embedding-based reranking to select the best caption among candidates, while at training we introduce MedPAIR-SCST, which combines clinically relevant rewards to shift the generative distribution toward improved clinical alignment. Our experiments show that complementary visual representations with a multi-encoder design and concept-level auxiliary learning help preserve clinically meaningful information. Furthermore, inference-time reranking provides a practical way to improve semantic and clinical alignment without additional training, whereas MedPAIR-SCST goes beyond selection by directly improving the model's distribution to generate more consistent and clinically grounded captions. These findings suggest that jointly leveraging selection-based alignment and reinforcement-learning-based alignment can promote more trustworthy medical image captioning even in data-constrained settings.

Conventional convolutional kernels are typically defined on fixed discrete grids, limiting their ability to accommodate heterogeneous local structures. Existing adaptive operators improve flexibility but often couple geometric scale variation with content-dependent filtering, while incurring high computational cost from per-location kernel generation. To decouple geometric scale adaptation from content-dependent filtering while avoiding expensive per-location kernel generation, we propose Continuous Relative-scale KAN (Core-KAN), a relative-scale-conditioned continuous convolution operator. Core-KAN maps input features into a compact latent basis space and uses a lightweight scale controller to predict local scales relative to an exponential moving average reference. A KAN-based generator represents depth-wise kernel bases as continuous coordinate functions, allowing the operator to synthesize spatial filters at arbitrary resolutions rather than being confined to a fixed lattice. Instead of synthesizing independent kernels at every location, it constructs a compact bank of scale-conditioned kernel responses and interpolates them according to the predicted local scale map. An independent mixing controller further combines the interpolated basis responses based on local content, explicitly decoupling geometric scale adaptation from content-dependent filtering. Together with lightweight pointwise projections, this design forms a low-rank dynamic convolution that scales efficiently with kernel size and integrates readily into hierarchical vision backbones. Experiments across three representative vision tasks show Core-KAN consistently outperforms strong convolutional and dynamic-kernel baselines with only marginal parameter and computational overhead, offering an efficient, general framework for continuous, scale-adaptive convolution.

To prevent the adoption of aesthetically polished but pedagogically flawed AI content, we study a video authoring pipeline featuring two layers of structured refusal. The first layer empowers educators to iteratively reshape AI scripts based on multimedia learning theory, while the second employs automated metrics to flag violations in instructional coherence and narrative-visual synchronization. While neither layer is exhaustive, their synergy ensures that principled resistance--the act of deferring AI output until it meets rigorous standards--becomes a catalyst for higher quality. Evaluation combining a study with 23 educators across 3 topics and automated metrics across 7 topics drawn from established science and philosophy curricula shows that both layers independently improve the same instructional dimensions, suggesting that thoughtful resistance and generative AI are not opposites but partners.

Cyclic peptides are emerging as promising molecular scaffolds in drug discovery due to their high binding affinity and structural stability. However, extending generative models from linear to cyclic peptide design remains challenging, as cyclization sharply restricts the feasible design space through coupled geometric and biophysical constraints. Moreover, limited training data has led existing approaches to rely largely on zero-shot generation or post hoc filtering, resulting in low yields of feasible designs and limited control over multi-objective trade-offs. To address these limitations, we propose FAR-DPO (Feasibility-Aware and Robust Direct Preference Optimization), an architecture-agnostic framework that steers generative models toward structurally and biophysically feasible cyclic peptide designs, particularly for challenging targets. FAR-DPO integrates feasibility-aware preference construction with difficulty-aware group-robust optimization. Specifically, it constructs within-target preference pairs through feasibility-gated multi-objective dominance and adaptively reweights predefined difficulty groups according to their current preference losses. On the CPSea LNR benchmark, under a fixed generation budget, FAR-DPO increases overall success rate from 46.89% to 57.79% on PepGLAD and from 47.96% to 49.57% on PepFlow. These gains also extend to the hardest target quartile and are accompanied by more favorable best-per-target binding scores. Together, these results demonstrate FAR-DPO's effectiveness in improving feasibility and target-wise robustness.

Vision-language models (VLMs) can estimate physical quantities such as duration, speed, and acceleration from visual observations, but existing benchmarks primarily assess overall model performance against annotated ground truth. In deployment, a key question is whether an individual prediction can be trusted when its ground truth is unavailable. Self-consistency alone may fail to capture important failure modes: a VLM may produce stable-but-wrong estimates or rely on textual priors rather than visual evidence. We formulate answer-level selective prediction for quantitative physical reasoning and propose Answer-Level Trust Selection (ATS), a post-hoc, model-agnostic framework for accepting or rejecting individual VLM predictions. ATS requires no fine-tuning, auxiliary verifier, or access to the model's internal logits. Instead, it aggregates eight interpretable behavioral diagnostic scores derived from repeated queries and controlled interventions into a unified trust score. We evaluate ATS in depth on Qwen2.5-VL-7B and across 20 VLM backbones, examining selective performance, diagnostic behavior, and targeted failure modes. Our results show that intervention-based diagnostics help identify stable-but-wrong and prior-tracking predictions that repeated agreement alone may miss. However, improved failure-case rejection can come at the cost of lower retention of correct predictions. ATS therefore complements model-level capability evaluation with answer-level reliability assessment for quantitative VLM predictions. Code will be released upon publication.

Buildings account for roughly one-third of global energy consumption and CO$_2$ emissions. Optimizing indoor climate systems plays a critical role for urban climate mitigation aligned with UN Sustainable Development Goals 11 and 13. However, indoor delayed thermodynamic responses and partial observability severely hinder existing methods, which are primarily limited by implicit thermal inertia, occupancy dynamic prediction, and cumulative prediction errors, especially for out-of-distribution environments. In practice, these challenges are further exacerbated by the high cost and privacy burden of dense indoor sensing, forcing operators to collect only limited data in a single operating regime while expecting controllers to generalize reliably across unseen seasons and climate regions. To address this problem, we propose ADAPT, a physics-aware conditional diffusion indoor environmental world model for HVAC control. The model predicts a short-horizon held-action thermal baseline to capture the latent thermal inertia of the buildings. The diffusion backbone utilizes the robustness of generative models, while a learnable multi-zone heat-balance regularizer constrains generated trajectories to satisfy transferable building thermodynamics without requiring known building geometry or manually calibrated thermal parameters. A credit assignment is then design for the downstream reinforcement learning. Extensive experiments on SemibuildingSim and Sinergym demonstrate that ADAPT reduces HVAC energy consumption by 7.3\% and occupant discomfort by 30.2\% compared with state-of-the-art baselines under IID control. Under OOD control scenarios spanning unseen seasons and climate regions, ADAPT maintains robust performance with only marginal degradation relative to its IID performance, substantially outperforming existing methods in transfer robustness.

Credit assignment is challenging in long-horizon agentic reinforcement learning, where supervision often comes only from final rewards. Existing methods refine trajectory-level signals into step-level credits through step grouping or graph-based advantage estimation, but can overlook meaningful intermediate milestones. We propose MileGPO (Milestone Inference with Local Evidence for Graph-Based Policy Optimization), which derives process-level credit from grouped on-policy rollouts through three designs. Milestone Discovery identifies candidate milestones on successful rollouts and recurring traps on failed ones. Reliability-Calibrated Shaping (RCS) weights these candidates by outcome-based confidence, strengthening reliable milestones and traps while down-weighting uncertain ones. Progress-Contrastive Calibration (PCC) further tests whether a candidate reflects local progress and whether its incoming ansition outperforms observed alternatives from the same state.MileGPO requires neither auxiliary models nor additional environment interaction. Experiments on ALFWorld and WebShop show state-of-the-art performance and a small in-distribution to out-of-distribution gap on ALFWorld. Ablations and credit diagnostics indicate that reliability weighting, local progress, and same-state branch evidence complement milestone discovery and resolve ambiguous intermediate credit.

LLM evaluation pipelines often have many candidate judges: general LLM-as-a-judge prompts, reward models, safety classifiers, confidence variants, and task-specific verifiers. The deployment question is not only which judge is best, but which judges should be called, on which examples, and when panel construction should stop. We formulate judge-panel design as a role-conditioned allocation problem. From a small labeled audit set, declared slices, and judge costs, the method estimates target-relative roles: copies add no conditional information, complements improve the global panel, and specialists help only on slices. These roles induce a policy: drop copies, add complements globally, route specialists conditionally, and stop when validation gain falls below a threshold. Across reasoning, code, safety, preference, reward-model, summarization, and math audits, the method is compared with single judges, flat panels, matched diversity heuristics, full-call stacking, reliability juries, and frugal cascades. The result is a regime map for judge calls: route specialists on deployable slices, stop in saturated verifier regimes, keep broad ensembles when their risk benefit is worth the cost, and ignore conditional copies. The output is a reusable, auditable call plan for the next evaluation batch.

Financial anomaly detection often relies on large unlabeled transaction logs, where anomalous samples may already be present during training. Such training-set contamination violates the clean-normal data assumption underlying many anomaly detection methods. Although flow matching has demonstrated strong performance in generative modeling, its robustness in unsupervised tabular anomaly detection remains underexplored. In this work, we study flow-matching-based anomaly detection under contaminated training data by comparing Time-Conditioned Contraction Matching (TCCM) with Forest-Flow and evaluating multiple anomaly scoring functions. Our results show that the choice of anomaly score is critical. The original single-step Decision score used by TCCM is sensitive to contamination, whereas trajectory-based Deviation and Reconstruction scores provide more stable anomaly signals. With these scores, Forest-Flow becomes competitive with, and in some cases outperforms, TCCM. These findings highlight the importance of anomaly scoring for flow-matching methods in financial anomaly detection under severe class imbalance.

Low-Rank Adaptation (LoRA) is a prominent fine-tuning method for large models, achieving competitive performance with reduced memory overhead. However, a persistent performance gap remains between LoRA and full fine-tuning. Recent studies have sought to narrow this gap by employing one-step gradient approximations of pretrained weights to align LoRA updates with the principal directions or intrinsic dimensionalities of full fine-tuning updates. Nevertheless, these approaches fail to capture the full dynamics of the gradients. In this paper, we propose LoRA-GA$^2$, an effective fine-tuning algorithm that fully leverages multi-step gradient information. Specifically, we introduce a lightweight probe for multi-step gradients of pretrained weights that incurs no additional GPU memory cost and only marginal time overhead. We further employ a spectrum-aware, importance-based rank allocation and optimal initialization derived from multi-step gradients. Extensive experimental results demonstrate that LoRA-GA$^2$ consistently outperforms existing LoRA variants while preserving the efficiency advantages of vanilla LoRA. For instance, LoRA-GA$^2$ surpasses the leading baseline by an average of 0.66 points on the GLUE benchmark, and outperforms the strongest baseline by 1.03 points on GSM8K and 0.87 points on HumanEval, respectively.

Software increasingly functions as part of the scientific instrument itself, making failures in scientific code capable of compromising not only program behavior but also the evidence underlying scientific conclusions. Yet existing evaluations of coding agents largely emphasize aggregate task success, providing limited insight into why agents fail when repairing scientific software. We introduce \textbf{SWE-bench Science}, a repository-level benchmark for scientific software engineering comprising 119 tasks from 98 GitHub repositories across 20 scientific domains. Each task is organized into one of three paradigms: Issue-driven, Expert-exploratory, and Engineering-integration. Even the best-performing agent, \textbf{Claude Code with Opus-5 (max), achieves a pass@1 below 50\%}, highlighting the substantial challenges posed by scientific software engineering. We identify four recurring failure mechanisms: deficits in scientific knowledge or abstraction, misguided exploration or surface-level repair, incomplete repair coverage or system integration, and failures to generalize scientific knowledge beyond observed cases in our analysis. We further conduct a paired ablation that removes explicit scientific guidance while preserving the repository and executable engineering context. The results show that scientific knowledge is not uniformly beneficial: well-grounded information can constrain repair and improve average performance and token efficiency, whereas poorly aligned guidance can induce anchoring and does not necessarily improve exact repair success. Together, SWE-bench Science provides a broad testbed for studying both the capabilities and failure mechanisms of coding agents in scientific software engineering.

The convergence of large language models (LLMs), structured knowledge bases (KBs), and reasoning ability (RA) presents a promising trajectory toward general embodied intelligence (GEI). This paper reviews the evolution of LLM-centered intelligent systems, emphasising their integration with knowledge representation, logical reasoning, and physical embodiment. We analyse LLM architectures, pre-training methods, and inference mechanisms, along with their interaction with external knowledge sources and structured reasoning frameworks. Furthermore, we examine embodied intelligence (EI) paradigms wherein agents learn and act in physical environments. To synthesise these dimensions, we present a conceptual framework that illustrates the synergy among LLMs, KBs, RA, and embodiment, serving as a guiding model for perception, reasoning, and action rather than an implemented engineering architecture. To advance toward GEI, we identify five key challenges: efficient LLM deployment, closed-loop knowledge integration, hybrid symbolic-neural reasoning, perception-action grounding, and continual learning. This survey provides a comprehensive roadmap for developing adaptive, multimodal agents capable of operating in complex, dynamic settings.

Discovering materials with desirable properties often requires searching large candidate spaces while experimental or computational evaluations remain costly. Active learning addresses this challenge by using previous observations to select which candidate to evaluate next, typically through probabilistic surrogate models. We investigate whether open-weight large language models (LLMs) can serve as standalone acquisition policies in this setting. We evaluate five LLMs across four retrospective finite-pool materials optimization tasks under different candidate-presentation strategies and compare them with random selection and conventional Gaussian-process methods. LLM policies generally reach the global optimum in fewer iterations than random selection, indicating that they provide a useful acquisition signal without task-specific training. Their performance relative to Gaussian-process methods is mixed: conventional acquisition performs better on most tasks, while LLMs match or outperform it in some settings. Performance varies substantially across tasks, models, initializations, and candidate presentations, with no LLM approach performing best across all tasks. Overall, open-weight LLMs show potential as acquisition policies for finite-pool materials search, although their reliability remains sensitive to the task and to how candidates and scientific context are presented.

Developed as a workhorse for classical simulations of quantum algorithms and quantum many-body systems, Tensor Network methods have entered the scientific mainstream in quantum physics. Among various types of tensor networks, Tensor Trains (commonly know as Matrix Product States in the quantum computing community) have already found applications in machine learning. These methods often rely on a powerful linear algebra tool called the Singular Value Decomposition (SVD). Several conditional GAN architectures for image denoising incorporate SVD as a single-cut decomposition step applied to generator feature maps. In this work we introduce TT-Net, which replaces the per-channel SVD denoising block with a two-cut tensor-train decomposition capable of accessing cross-channel information directly, a capability absent from contemporary alternatives. In a controlled comparison differing only in this decomposition mechanism, TT-Net outperforms SVD-Net on PSNR and SSIM across all three noise types tested (Gaussian, motion blur, and salt-and-pepper), supporting the hypothesis that cross-channel access improves denoising quality. Training-dynamics analysis further shows that TT-Net's adversarial loss term consistently saturates to a stagnant state across all three noise types, more so than SVD-Net's, while reconstruction quality continues to improve regardless, raising an open question about the adversarial component's contribution that this work identifies but does not resolve. Furthermore, for Gaussian noise our method outperforms both the EigenGAN and the state of the art Pix2pix method which does not assume any linear algebra decompositions and does not retain any linear algebra information. Our manuscript shows how quantum inspired tools can be used as practical real world feature filters for deep learning applications.

Discovering dynamical systems from trajectory data is a central problem in applied mathematics and engineering. Whilst recent advances in machine learning have led to strong progress in data-driven system identification, much less attention has been given to systems with discontinuous dynamics. These systems are nevertheless highly relevant in applications, including climate dynamics and mechanical systems with friction. In this work, we consider the problem of identifying piecewise-smooth dynamical systems directly from trajectory data. Compared with the smooth setting, this requires recovering the governing equations and detecting the switching hyperplanes that separate different dynamical regimes and characterising their behaviour, such as sliding motion. We present a modular framework for discovering such systems by first estimating switching hyperplanes from data and then learning smooth dynamics within each region using geometry-constrained neural networks. The geometry-learning phase is studied from a statistical perspective, analysing the identifiability of the discontinuities and the robustness of the procedure. We also introduce a novel neural network architecture with a prescribed discontinuity set, and provide a theoretical analysis of its approximation properties. The approach is tested on low-dimensional benchmark problems, including dry-friction oscillators and the PP04 climate model for the ice ages.

LLM agents have achieved strong performance on general software engineering tasks, yet struggle with domain-specific code generation. We identify the root cause as the agent's lack of tacit knowledge, including domain-specific business rules, interface contracts, and operational conventions that developers internalize through practice but never document. This knowledge is deeply buried beneath the domain code, dispersed across code entities and their dependency relations, and invisible to the agent that lacks it. These properties make tacit knowledge inherently difficult to retrieve or learn. In this work, we propose PRAXIS, a framework that enables agents to systematically extract, represent, and reuse tacit knowledge for domain code generation. PRAXIS acquires tacit knowledge by simulating human development workflows within the target codebase, distills it into structured units organized on the code dependency graph, and proactively surfaces it to the agent at the point of code interaction. Extensive experiments demonstrate that PRAXIS outperforms state-of-the-art agents equipped with powerful agentic search capabilities, as well as experience-based and skill-based methods. The approach integrates seamlessly into various agent frameworks and LLMs with consistent performance improvements, and supports continual evolution with performance steadily scaling as practice accumulates.

World models provide digital simulacra of the true world, allowing agents to be trained and tested before costly real-world deployment. At each time step, they receive an action and generate an observation and reward matching the statistics of the true world. In complex environments where present outcomes depend on events far in the past, this requires memory. One might expect that, by increasing memory, we can always build a model accurately enough to align the optimal agent policies of the real and virtual worlds. We show that this is false for classical world models, even when the true world itself is classical. We construct true worlds for which every finite classical model fails along the same possible trajectory: it either loses the ability to distinguish actions when the true world clearly prefers one, or repeatedly assigns the highest expected reward to suboptimal actions. Its expected-reward estimates also retain a nonvanishing average error. In contrast, each such true world admits a quantum world model using a single qutrit that reproduces it exactly: its reward estimates and preferred actions always match those of the true world, ensuring that the optimal policies of the real and virtual worlds remain perfectly aligned.

Pedestrian simulators need a behaviour rule for every agent, but privacy usually limits the data for setting one to aggregate statistics, namely zone-level device counts and origin-to-destination (OD) flows, with no individual trajectories. Such aggregates under-determine individual behaviour, because many different sets of decisions reproduce the same counts. We fine-tune a language model crowd agent so that the simulated population matches the observed destination composition, the fraction of the departing crowd heading to each point of interest. We read this target from the OD flow and reweight the model's own destination distribution onto it by iterative proportional fitting. Because fine-tuning inflates the dominant destination class, we fit the low-rank adapter to trajectories resampled to a corrected training composition that reaches the target after this inflation. On mobile network counts from two baseball games the fine-tuned agent runs without inference-time correction, cutting the destination-share error by 25%, while the grid correlation remains similar across policies.

Current dexterous grasp planners primarily optimize for physical stability, focusing on whether an object can be grasped rather than how it should be grasped to support downstream functional tasks. However, conditioning grasp synthesis on specific human grasp taxonomies typically requires prohibitively expensive, object-annotated datasets. To address these limitations, we propose CoToGrasp, a novel generative framework that synthesizes diverse, stable grasps strictly conditioned on specific contact topologies. To bypass the data collection bottleneck, CoToGrasp is trained entirely in an object-agnostic manner. We introduce a feature-based canonical workspace that projects local object features into a unified gripper-centric domain, effectively decoupling the semantic functional intent from the arbitrary object geometry. By learning the intrinsic contact manifold of the gripper within this workspace, our model achieves zero-shot generalization to unseen objects at inference. Extensive evaluations on the large-scale DexGraspNet dataset demonstrate that CoToGrasp achieves state-of-the-art performance, outperforming existing taxonomy-guided planners. Finally, we demonstrate the physical viability and kinematic feasibility of our synthesized contact topologies on a physical robot platform. Code is available on our project website https://cea-list.github.io/cotograspweb/ .

Skchange is an open-source Python library for detecting structural changes in time series. It implements modern change detection algorithms within a unified and extensible framework. The algorithms are modular and composable, and they include changepoint search methods based on both cost minimisation and statistical tests. Key features include the detection of anomalous segments in addition to changepoints; theoretically well-founded fast and approximate search methods; theoretically well-founded algorithms for high-dimensional data, covering settings where either few or many features change simultaneously; utilities for automatic and data-driven penalty calibration, which balances false alarms against missed detections; and a large collection of built-in costs and statistical tests. The design follows established scikit-learn conventions to streamline both user and contributor experience, and Numba is used extensively to achieve high computational performance. Source code and documentation are available at https://github.com/NorskRegnesentral/skchange.

Self-supervised pretraining on remote sensing imagery typically treats all samples as equally informative, despite large variability in geographic and visual structure. We propose a curriculum learning strategy for self-supervised Earth observation that ranks samples by geographic isolation, a label-free proxy derived entirely from geolocation metadata already present in geospatial datasets, requiring no image decoding, no model feedback, and no manual annotation. Unlike visual complexity proxies, it scales as O(D log D) with dataset size D and is well-defined for both contrastive and reconstructive objectives. We integrate the proposed measure into MoCoV2 and MAE pretraining and evaluate across three downstream tasks from CopernicusBench (BigEarthNet, DFC-2020, LCZ). Our curriculum reaches baseline final-epoch performance using as few as 20% of the training budget (MAE) and at most 40% (MoCo) of the training budget, and improves final downstream performance by up to +5 mAP on BigEarthNet, with gains of 1-5 points across benchmarks, matching visual-complexity curricula while reducing pre-computation cost by more than 140x (4 s vs. 568 s on SSL4EO). A CKA and effective-rank analysis further reveals that curriculum-trained encoders develop higher-dimensional, more uniformly utilized embedding spaces throughout training.

A minibatch can influence training beyond the update at which it is observed because AdamW stores past gradient information in its optimizer states. We study this delayed effect through paired trajectories that differ only in one gradient update and share the same subsequent training sequence. We formulate AdamW as a finite-horizon input--state--output (ISO) system whose state contains the model parameters and first- and second-moment estimates. Linearizing the joint dynamics yields a signed response operator that maps a localized gradient perturbation to its future loss effects, revealing how optimizer memory shapes their magnitude, timing, and sign. We further derive an exact multistep error decomposition and establish first-order finite-horizon accuracy under local smoothness and controlled activation switching. Experiments validate the response mechanism and optimizer-state effects, while repeated-future analyses reveal substantial prospective structure in delayed influence that can be partially recovered from ISO approximations. Code is available at https://github.com/Kanyooo/Loss_ISO.

Audited against causal ground truth from executed replay in a single-agent tool environment (ALFWorld), none of the step-level credit signals used to train LLM agents -- LLM-judge scores, outcome-conditioned logprob ratios, or the policy's own confidence -- identifies which steps causally matter better than chance. Existing evaluations grade these signals against annotated step *correctness*; we audit them against step *contribution* -- what re-sampling the policy's own alternatives at each decision point and rolling forward actually changes about the outcome -- and the two come apart. The ground truth itself is structured: causal contribution is sparse (30.5% of decision points where ground truth is defined carry measurable effect), and measurability is model-dependent -- the fraction of points with no policy-supported counterfactual differs by a factor of two (13.1% vs. 26.8%) between two similar-scale policies. The failure mode is identifiable: implicit credit echoes the policy's fluency (median rank correlation +0.75, replicating at +0.70 in a second family under a corrected instrument), while conditioning on the outcome adds no causal information (partial correlation -0.004, Qwen). A confidence-only router recovers pivotal steps at chance level, but cuts judge cost by 13.1% per turn (14.0% per trajectory). In a seven-arm pre-registered training experiment, no arm reliably outperforms the untrained policy, and the checkpoints' apparent instrument signature is fully explained by training dose -- sparser credit retains fewer examples, an order-of-magnitude spread in optimizer steps -- not credit content. Comparisons of credit rules must therefore match effective sample size, or they measure dose, not credit.

Multifingered grasping is a crucial robotic skill, but current deep-learning grasp planners often struggle to generalize to new objects because they are trained on limited, object-specific datasets. We introduce a fundamentally different approach, grounded in the observation that the gripper and the object share identical surface geometry at their mutual contact points. We propose GOAG: Generative and Object-Agnostic Grasp Planner for Dexterous Robotic Manipulation, a novel deep generative model that learns a compact latent representation of a specific gripper's contact surface distribution, enabling the efficient sampling of valid grasp configurations without relying on object-specific training data. We show that by introducing object features only at inference time, our model can effectively retrieve admissible contact areas that are compatible with the gripper's capabilities. We validate our approach through extensive experiments on established grasp protocols in both simulated and real-world scenarios, demonstrating its effectiveness with different grippers from the literature. Our method delivers state-of-the-art results on the objects from the MultiDex dataset, achieving an average success rate of 86.93%. It offers significantly faster processing when generating numerous grasps, while matching the performance of leading approaches specifically trained on this dataset. Unlike these methods, our approach does not rely on object-specific training data, highlighting the advantages of object-agnostic learning. It effectively addresses the generalization challenges faced by traditional data-driven grasp planners. Code and videos are available on our project website https://cea-list.github.io/goagweb/ .

Long-context modeling is a pivotal capability for Large Language Models, yet the quadratic complexity of attention remains a critical bottleneck, particularly during the compute-intensive prefilling phase. Our previous work, FlashPrefill, mitigates this cost through instantaneous pattern discovery and max-based dynamic thresholding; however, it remains an algorithmic prototype that is still distant from production deployment. In this paper, we present FlashPrefill V2, which evolves FlashPrefill from a prototype toward practical long-context serving along three dimensions. First, we introduce a mean correction term that effectively suppresses the approximation error, keeping performance degradation manageable even at extreme sparsity levels. Second, we redesign the sparse attention operator with PackGQA memory access, warp specialization, and pingpong pipelining, fully aligning with the latest FlashAttention-3/4 implementations and supporting FP8 inference to meet practical quantization requirements. Third, FlashPrefill V2 natively supports paged KV cache and continuous batching, allowing integration as an attention backend in modern inference frameworks such as SGLang. Extensive evaluations on NVIDIA H20 GPUs---among the most widely deployed inference accelerators---demonstrate that FlashPrefill V2 delivers up to 47.26x and 27.19x speedups over FlashAttention-2 at 128K context length under FP8 and BF16 precision, respectively, and, in FP8, still achieves a 30.49x speedup against an FA3/4-aligned dense baseline.

Testing embedded software in modern vehicles is challenging due to system complexity, decentralized architectures, and strict safety and performance constraints. In this work, we present an end-to-end, deployment-aware testing pipeline for IoT-based automotive applications. The pipeline combines requirement-driven test and code generation with large language model (LLM) and vision-language model (VLM) assistance, and human-in-the-loop curation to reduce manual effort and improve consistency. Using Eclipse openDuT, it supports flexible, distributed deployment across geographically separated cyber-physical and IoT infrastructures, optimizing for node availability and cross-organizational coordination. For validation, we conduct a case study using a Child Presence Detection System (CPDS), achieving full functional requirement coverage across all 9 requirements and 100% Gherkin generation accuracy on the controlled requirement set. Distributed test execution across geographically separated ECUs via Eclipse openDuT confirms the pipeline's applicability to OEM--supplier testing workflows.

Micro-View Order-Dispatching assigns available drivers to passenger orders within each dispatch batch and is critical to the service quality and operational efficiency of ride-hailing platforms. Mainstream industrial solutions follow a multi-stage paradigm of model prediction, value calculation, and dispatch matching. Although dispatch quality is determined by the final batch-level assignment, these stages optimize different intermediate objectives. This cross-stage objective inconsistency means that improving a single stage does not necessarily improve the overall dispatch result. We therefore formulate Micro-View Order-Dispatching as a generative matching problem and propose GenMatch, an end-to-end Generative Matching framework and the first such framework deployed in a real-world production environment. Applying generative modeling to this problem introduces three challenges. First, each dispatch batch forms a dynamic sparse bipartite graph, requiring efficient structured batch-level encoding. Second, replacing the hand-crafted value function requires learning unified business utility from heterogeneous feedback. Third, directly generating an assignment requires tracking the evolving matching state because each selected order-driver pair changes the remaining feasible candidates. GenMatch addresses these challenges with a Context-Aware Bipartite Encoder, a Business-Aware Utility Learner, and a State-Aware Pointer Decoder. Extensive offline evaluations and online A/B tests in five cities across DiDi's international ride-hailing markets show consistent improvements over competitive baselines, confirming the effectiveness and practicality of GenMatch for industrial order-dispatching.

Inference-time selection methods, such as Best-of-N, improve generation by sampling a pool of candidates and selecting the top-ranked completion according to a reward model. Distillation seeks to amortize this procedure into a single policy by replacing raw rewards with in-pool ranks and learning a policy that upweights higher-ranked completions. However, existing rank-based policies typically use smooth full-support reweighting, so low-ranked completions receive less mass but remain in the target support. Although a sharper reweighting reduces lower-tail mass, it also increases reliance on brittle ranking at the top made by a single reward model. We propose TUP: a Truncate-bad, Upweight-good Policy that removes low-ranked completions from the support and reweights only the retained upper tail with a tunable sharpness. TUP admits a closed-form, prompt-independent normalization and can be trained fully offline via binary cross-entropy, using shifted-truncated win-rates as soft labels and distilled-to-reference log-likelihood ratios as logits. Theoretically, under certain assumptions, we show that for any unknown oracle reward, the best monotone rank-reweighting can be matched by a lower-tail truncation rule, providing formal support for removing the lower tail rather than merely downweighting it. Empirically, we show that TUP is competitive with strong offline alignment baselines.

Personalized text generation aims to make LLMs write in a specific individual's style, yet existing benchmarks measure task accuracy or preference alignment rather than whether the model's output actually resembles the target author's writing. We introduce PersonalBench, a benchmark that evaluates inference-time personalization methods through three independent lenses: LUAR (a trained authorship verification model), an LLM-as-judge, and automated stylometrics. Across 50 authors, 1,000 generations, and two model families (Qwen 3, GLM-4), we find that personalization methods do produce author-differentiated output (LUAR discriminates target authors within generated text at AUC=0.918) but this differentiation never crosses the human-LLM boundary. All methods achieve LUAR similarity to real authors in the range 0.484-0.508, below the cross-author human floor of 0.626 (ceiling 0.756). The LLM's own authorship fingerprint dominates: generated text is more distant from any human author than random humans are from each other. Methods are statistically indistinguishable from each other on LUAR (spread 0.024) despite appearing differentiated on the LLM judge, a discrepancy we trace to circularity between trait extraction and profile extraction. We validate that LUAR reliably measures authorship in our corpus (AUC=0.76 single-post, 0.96 multi-post). We release PersonalBench as a calibrated measuring stick: inference-time personalization modulates the LLM's style but does not bridge the gap to human authorship.

Recent agent benchmarks increasingly ground evaluation in executable environments, from code repair to web navigation, app APIs, and function calling. Yet completing consequential work beyond code requires more than producing a plausible response or valid tool call: agents must gather missing information over multiple turns, follow domain policies, coordinate dependent tools, and realize the correct persistent state transition without collateral effects. In this paper, we introduce Thinkingbox, a sandbox for tool-agent-user interaction that provides isolated MCP-compatible tool sessions, complete execution traces, and outcome evaluation over terminal backend state. Built on this sandbox, Thinkingbox-bench contains 507 policy-conditioned workflows across numerous scenarios, including retail, hospitality, auto insurance, neobank internal IT, and consulting IT/HR support. Each attempt is evaluated by task-specific executable checks that accept valid trajectories while rejecting wrong, missing, or extra effects; designated tasks additionally check required properties of the final response. Across proprietary and open-weight models, the strongest achieves 65.36% pass@1, but only 25.25% pass^20. Moreover, many failed trials show clean termination and valid state-changing actions, showing that response or tool-call-level signals are not clear proxies for end-to-end task completion. Thinkingbox-bench reveals a large gap between occasionally finding a successful trajectory and reliably completing stateful business tasks. We release both Thinkingbox and Thinkingbox-Bench: https://github.com/microsoft/thinkingbox

Multimodal LLMs can see a document, but they often can't read it reliably. Small text, tables, visual cues, and topological elements still trip them up under direct visual inference, even when the page is already sitting in the model's context. Most document-VQA systems treat perception as fixed: they encode the page once, ask the question, and answer from whatever the model happened to extract in that single fast pass. We think document VQA needs slower, more deliberate perception: rather than answering from one fixed encoding, the model should spend a bit of extra compute at inference time working out what to look at next, and only then answer. We build this into \textbf{Q-Guide}, a small agent that reads a question, works out what evidence it is still missing, and calls targeted tool(s) to recover it---reading text where text is needed, zooming in where detail is needed, or grounding a region where position matters. On DocVQA2026 and Manga109, Q-Guide outperforms both direct prompting and recent multi-agent document systems ($65.0\%$ vs.\ $40.0\%$ on DocVQA2026, $32.4\%$ vs.\ $24.4\%$ on Manga109), and the improvement holds across three Claude backbones (Opus 4.6, Sonnet 4.6, and Opus 4.5). We find that accuracy scales with the perception budget---most of the gain appears within two to three deliberate rounds---and that the gain comes from directing perception to the right place, not from complex control logic: adding planners, routers, or multiple collaborating agents does not help.

Full-cycle biventricular geometry is essential for characterizing cardiac function. However, dense and temporally consistent 3D+t biventricular meshes are not routinely available, whereas end-diastolic (ED) anatomy can often be obtained reliably. We therefore investigate full-cycle biventricular motion synthesis from a single ED mesh. This task is challenging because cardiac deformation is spatially heterogeneous and phenotype dependent, while conventional global generative models often obscure localized motion patterns. In this study, we propose a region-specific and phenotype-adaptive framework that integrates motion-informed functional parcellation with conditional latent flow. A functional partition learned from reconstructed motion organizes the ventricular surface into regions with coherent dynamics and enables topology-aware regional feature exchange. A phenotype-conditioned rectified-flow model subsequently maps the ED anatomy to full-cycle motion latents through fine-grained conditioning and prototype-routed motion adapters. An optional control branch further incorporates available motion descriptors for controllable synthesis. Experiments on ACDC, M\&Ms, and M\&Ms-2 demonstrate consistent improvements in geometric accuracy and functional fidelity. Under ED-only synthesis, our method achieves biventricular ASSD, HD95, and vRMSE of \(1.49\pm0.34\)~mm, \(3.77\pm1.06\)~mm, and \(3.31\pm1.03\)~mm, respectively, outperforming all competing methods. Complementary functional and robustness evaluations further demonstrate that the synthesized sequences preserve physiologically plausible ventricular dynamics and generalize across cohorts and disease phenotypes. The code will be released publicly upon acceptance of the manuscript for publication.

Large vision-language models (LVLMs) have achieved remarkable progress in video understanding and reasoning. Despite extensive studies on text- and image-based jailbreaks, video jailbreaks against LVLMs remain largely unexplored. Existing video jailbreak methods mainly manipulate textual content embedded in videos, while overlooking how such information is organized over time. Our analysis reveals that jailbreak effectiveness depends not only on the semantics of textual information but also on its temporal presentation, including duration and timing-slot allocation. Motivated by this finding, we use subtitles, which are common in real-world videos and allow semantic content to be presented under precise temporal control without appearing visually intrusive, as a natural attack medium. Based on this insight, we propose TempJail, a black-box video-based jailbreak framework that constructs query-aligned dialogue-style subtitle sequences and optimizes their temporal scheduling to exploit temporal vulnerabilities in LVLMs and elicit responses that satisfy the harmful intent of the source query. Extensive experiments on four representative LVLMs and two datasets demonstrate that TempJail achieves the highest attack success rate across all evaluated model--dataset settings, outperforming the strongest baseline by 53 and 18 percentage points in dataset-averaged ASR on GPT-5 and Gemini 3.5-Flash, respectively.

We introduce RecPFN, a prior-fitted network that brings in-context learning to sequential recommendation. RecPFN is pretrained entirely on synthetic clickstream environments sampled from a broad structural causal prior, enabling it to amortize Bayesian-style inference from a small support set. At inference, a lightweight decoder-only transformer conditions on a handful of domain sequences and produces next-item predictions for queries in a single forward pass, without any weight updates. Across eight public benchmarks, RecPFN achives state-of-the-art zero-shot performance while remaining strongly competitive with supervised methods in low-compute and low-data regimes. It is deployment-efficient and robust to domain shift, outperforming strong zero-shot baselines that rely on large real-interaction corpora. RecPFN provides a practical path toward generalizable, data-efficient recommenders and opens avenues for richer priors, longer-context ICL, and multimodal extensions. Code for training and evaluation is publicly available at https://github.com/SAP-samples/tabular-ai-recpfn/.

Vision-language-model-based embodied agents can complete instructed tasks but often violate safety constraints in the process, a problem recently framed as interactive safety. Training such agents to act safely is difficult, since safety and task success are distinct objectives, and safety arises only at a small number of safety-critical steps within a trajectory. Standard supervision is insufficient: imitating safe trajectories teaches behavior without explaining why it is safe, and contrasting arbitrary safe and unsafe trajectories mixes the safety signal with unrelated differences. We propose SafeBranch, a framework that aligns an embodied actor on safety through branch pairs constructed from the actor's own unsafe rollouts via environment rollback. SafeBranch rolls each unsafe rollout back to the safety-critical step that caused the violation, queries the actor for a safe alternative, and pairs the original action with the alternative so that the two branches differ only at that step. The trained actor acts safely at deployment with no critic in the loop. On IS-Bench, SafetyALFRED, and out-of-distribution variants with unseen tasks and objects, it handles safety reliably without sacrificing task success, achieving roughly ten times more safe successes than the untrained baseline on the unseen-object variant.

Watermarking is a central tool for provenance in generative models, yet its application to multivariate time series remains hindered by reliability failures under post-editing attacks. We show that existing detectors, which rely on globally coupled re-encoding, suffer from bidirectional drift of the null distribution: post-editing attacks can shift the z-score of non-watermarked samples in either direction, invalidating clean-calibrated thresholds. We argue that this instability is a property of the re-encoding, and that reliable detection requires each recovered unit to depend only on a bounded temporal neighborhood. Guided by this principle, we propose L-VQVAE, a generative model in which each discrete token is produced from a short contiguous window, and LVQMark, a watermarking method over this token space that combines logit-bias injection with robust re-encoding for attack-time detection. Experiments on four benchmarks spanning finance, energy, and neuroimaging show that our approach preserves generation quality while stabilizing both detection power and false-positive behavior under post-editing attacks.

The typical training process of a multimodal large language model (MLLM) involves adapting both the language model backbone and the projector between the backbone and a modality-specific encoder. We ask whether fine-tuning the backbone of an MLLM is necessary to adapt it to a new modality. Through experiments on 3D MLLMs, we find that training only the projector is sufficient to achieve strong multimodal performance relative to existing baseline models and our jointly trained MLLMs with the same encoder and backbone. We also show that joint training leads to undesirable drift in existing capabilities of the language model, which projector-only training avoids by definition. Furthermore, projector-only training has approximately twice the training sample throughput of joint training. We validate our findings across different language model backbones via 3D classification and captioning benchmarks as well as standard benchmarks evaluating language, vision, and spatial reasoning capabilities.

Streaming video understanding requires models to causally update state as video arrives and organize growing history into semantic units that can evolve, persist, and be recalled under bounded computation and memory. This challenge is pronounced in live soccer commentary, where a system must describe completed events, summarize recent play, recall earlier events, or remain silent using only information available before each utterance. We present StreamSoccer, an event-driven system that uses event memory as its intermediate representation. A fixed-budget active memory integrates the stream; completed event states are retained locally and consolidated into retrievable historical records. A unified generator uses current, recent, and historical context to produce three commentary modes, while a rule-assisted scheduler selects a mode or silence. Unlike streaming video-language models organized around frames, visual tokens, or caches, and soccer-commentary methods based on predefined clips or output timestamps, StreamSoccer explicitly models event lifecycles. We construct a three-track streaming soccer commentary dataset and a layered evaluation protocol. At common reference anchors, StreamSoccer obtains CIDEr scores of 38.62, 23.96, and 17.39 for current-event, recent-window, and historical-memory commentary, ranking first on the current-event and historical-memory tracks and second on recent-window. Controlled ablations show that local completed events improve all tracks and that the full system performs best on all three. Across 174 raw-video runs on 58 matches, per-minute RTF p95 ranges from 0.10 to 0.22 without sustained growth with match history. These results indicate that event memory supports streaming soccer commentary across temporal scopes while controlling long-history computation.

Reference-based diffusion stylization requires separating target geometry from transferable appearance. Existing tuning-based methods often rely on aligned content-style-target triplets or auxiliary visual encoders, which increases data cost and can transfer unintended scene structure from the style reference. We propose SEFS (Style-Encoder-Free Stylization), a style-encoder-free conditioning framework for diffusion transformers. SEFS forms style tokens from stochastic low-resolution crops of single training images. This crop bottleneck preserves local appearance statistics such as palette, stroke, texture, and material, while reducing access to global layout cues. Target content is encoded by edge and segmentation cues and fused with the noisy latent through parameter-efficient trainable projections. We add style-to-denoising re-normalization for token-statistic alignment and cross-block skip fusion for spatial detail. SEFS trains on unpaired single images; the frozen diffusion VAE is used only to place image conditions in the latent space. On artistic stylization benchmarks, SEFS improves content consistency and leakage diagnostics while retaining reference-style affinity, and ablations support the crop-resolution, re-normalization, and skip-fusion choices. The code of SEFS will be made publicly available.

Reliable underwater robotic perception remains difficult because optical imagery degrades under turbidity, wavelength-dependent attenuation, low illumination, scattering, and blur. Although sonar provides complementary information that is less affected by optical visibility, prior visual-sonar research has largely focused on feature alignment and nominal detection performance. We investigate cross-modal robustness as visual reliability deteriorates and assess whether pretrained visual foundation-model representations can be complemented by sonar under severe degradation. We use frozen DINOv2 as the visual encoder and construct a controlled five-level benchmark ranging from clean to extreme visual conditions. We compare conventional visual detection, frozen foundation-model representations, sonar context, fixed multimodal fusion, clean-trained adaptive gating, and degradation-aware gated fusion. Our method trains the fusion mechanism across the full range of degradation while keeping the visual and sonar encoders frozen, allowing modality contributions to adapt without fine-tuning the pretrained backbone. Under extreme combined degradation, the DINOv2 baseline achieves 0.4610 balanced accuracy, while degradation-aware visual-sonar fusion reaches 0.6152, a 33.5% relative improvement. The learned sonar contribution increases from 14.2% under clean conditions to 41.3% under extreme degradation, demonstrating adaptive redistribution of cross-modal reliance. Fusion provides the largest gains under severe turbidity and blur, whereas color attenuation alone yields little additional benefit. These results show that foundation-model representations remain valuable but insufficient under severe information loss, and that explicitly adapting fusion to modality reliability can improve robust underwater multimodal perception.

Sequential agent search accumulates changes from its current champion but discards alternative branches; independent proposals preserve breadth but restart from the root. Loreley instead retains complete repository states in a Quality-Diversity (QD) archive and samples them as parents or supplies them as context for later edits. Candidates are Git commits produced in isolated worktrees and judged by a project-supplied evaluator. We compare configured Loreley QD, sequential champion editing, and independent root proposals in a matched Zstandard experiment: seven paired blocks and 48 physical candidate jobs per policy and block (1,008 total), with root-only initialization and each policy's native concurrency. Validation selected a winner at each budget checkpoint; an agent-hidden holdout measured the fixed candidate. At 48 jobs, QD was 0.135% below Sequential Champion (95% BCa interval for the paired effect: -0.556% to +0.161%) and 0.320% above Independent Root (-0.082% to +0.686%). Neither contrast established a QD advantage; Sequential had the highest observed 48-job mean and median. Archive retention and later sampling did occur. Four of seven final QD winners had a non-incumbent state in their primary-parent ancestry under a retrospective one-incumbent rule applied only to the observed QD stream. Including inspiration edges raised the count to six, without showing that supplied context caused an edit. Three earlier capability campaigns produced generation-4, multi-file improvements in two Python libraries and a separate Zstandard revision. Loreley engaged the intended stepping-stone mechanism, but the controlled experiment did not show an endpoint benefit at 48 jobs.

Long-term multi-agent systems continuously accumulate the memories produced by different agents. Existing memory methods typically treat retrieved memories as independent evidence and combine them through voting or weighting. However, this independence assumption often fails in multi-agent settings: memories written by different agents may inherit the same upstream source or shared bias, causing correlated evidence to be repeatedly counted and creating a false majority. We term this failure mode \textit{Memory Correlation Bias}. To address the issue, we propose the \textbf{C}orrelation-\textbf{A}ware \textbf{M}emory \textbf{A}rbitration (CAMA) framework that jointly decouples retrieved memories and recovers missing independent evidence. We model the retrieved memories as query-conditioned evidence groups and combine neural dependency inference with provenance-based symbolic priors to estimate the effective number of independent evidence sources, thereby preventing correlated memories from forming a false majority. Since critical independent evidence may be absent from the initial retrieval set, \textsc{CAMA} further learns a sequential recovery policy that actively retrieves alternative evidence or traces upstream sources before making the final decision, aiming to recover sufficient independent evidence for reliable arbitration while minimizing retrieval cost. Experiments on multiple benchmarks demonstrate the superiority of our method over the state-of-the-art baseline methods, suppressing false majorities induced by correlated memories.

Sparse keypoint extraction and matching underpin core tasks in geometric computer vision, including structure-from-motion, visual SLAM, augmented reality, and medical image registration. Learning robust local feature representations, however, typically requires accurate camera poses or depth supervision, which are often unavailable in real-world settings. Reinforcement learning (RL) has recently emerged as a promising alternative, requiring only the information if two images show the same scene or not. However, existing RL formulations such as RIPE rely on coarse binary rewards and carefully constructed negative training pairs, limiting training stability and descriptor discriminability. In this paper, we revisit RL-based keypoint learning and propose a reward that fully exploits the geometric consistency signal, deriving both reward and penalty from a single positive pair without contrasting against negatives. This richer signal provides sufficient supervisory contrast to learn discriminative detectors and descriptors from positive image pairs alone, enabling representation learning under extremely limited supervision. Furthermore, we show that the same RL objective can be extended to the matching stage by adapting LightGlue, raising AUC@5 on MegaDepth1500 from 56.58 to 59.65 and enabling weakly-supervised training of the full sparse matching pipeline from image pairs with partial visual overlap. We validate our approach on established benchmarks, demonstrating competitive results compared to fully-supervised methods. We further show that the method can be even trained on low texture medical video sequences, where camera poses are usually unavailable and standard SfM pipelines often fail. Code and data are available at https://github.com/fraunhoferhhi/RIPEpp .

LLM-based social simulation is a promising complement to traditional methods such as surveys and behavioral experiments. A core question is how to evaluate the fidelity of LLM-simulated human behavior and optimize LLMs toward it. Prevailing practice evaluates by accuracy, checking whether the model selects the single response observed from a human, and trains the LLM to reproduce this hard label. However, human behavior is inherently subjective: the same person in the same situation may reasonably act differently, so an observed response is only one draw from an underlying response distribution, rendering accuracy-based evaluation unreliable and hard-label training misleading. To address these problems, we first introduce the subjectivity coefficient, an entropy-based quantity distinguishing objective tasks such as coding from subjective ones such as social simulation, and use it to systematically analyze how accuracy-based evaluation and hard-label training fail as subjectivity grows. Based on the subjectivity coefficient, we propose Subjectivity-Adaptive soft-Label Training (SALT): it pools observed outputs from semantically nearby inputs into soft distributional labels, with an aggregation radius adapted to the estimated subjectivity of each input; in the near-objective limit the neighborhood shrinks, so SALT naturally falls back to standard single-label training. Moreover, since existing datasets record only single observed responses and cannot support distributional evaluation, we construct SUBJSIM, a benchmark of 19,300 contexts covering 193 annotators and 100 subjective questions. Since real-world data typically provide only a single observation per input, our experiments train models from single observed outputs while evaluating them against the full response distributions, verifying feasibility in realistic settings. Results on SUBJSIM demonstrate the advantages of our method.

We develop a geometric framework for learning deterministic and stochastic forced Hamiltonian systems with neural networks. Motivated by the Lagrange-d'Alembert principle and the theory of variational integrators, we introduce the notion of a Lagrange-d'Alembert map and establish a $C^r$ convergence theorem for first-order one-step methods. Building on these results, we propose Generalized Forced Hamiltonian Neural Networks (GFHNNs), a class of structure-preserving neural networks obtained by concatenating Lagrange-d'Alembert-Euler maps, and prove a universal approximation theorem for this architecture. We further extend the framework to parameter-dependent systems, leading to Parametric Generalized Forced Hamiltonian Neural Networks (PGFHNNs). By interpreting the multiple Stratonovich integrals appearing in the Stratonovich-Taylor expansion as parameters, the same framework can be applied to stochastic forced Hamiltonian systems whenever information about the underlying Wiener process is available. Our numerical experiments demonstrate that the proposed geometric architectures provide significantly improved long-time stability and accuracy compared to non-geometric residual neural networks, while requiring substantially less training data to achieve a comparable level of performance.

Recent studies investigate how to leverage pre-collected datasets to improve the policy performance and sample efficiency of RL. One promising approach to achieve this goal is to employ a two-stage strategy: In the first stage, diverse skills are extracted as a low-level policy from a given dataset, and a high-level policy is trained to solve a specific task in the second stage. Typically, extraction of the low-level policy is performed based on unsupervised learning such as trajectory VAE. However, a limitation of this approach is that the quality of the low-level policy highly depends on the quality of the dataset. To address this issue, we introduce QDOS (Quality-Diversity Offline Skill learning), a unified pipeline for robust offline-to-online learning. Our approach incorporates an Advantage-Weighted Quality-Diversity pretraining objective, which weights the skill extraction and diversity objectives by the estimated advantage of each trajectory segment. This approach allows the model to extract diverse and high-value skills. By providing robust and task-relevant skill representations, QDOS significantly improves the quality of the embedded skill space used by the low-level policy. We further integrate this with a dual dataset reuse strategy, where offline data is used both for skill pretraining and for populating the online replay buffer via pseudo-labeling. Experiments demonstrate that QDOS significantly outperforms strong baselines in structured manipulation tasks and unstructured locomotion tasks, confirming its ability to accelerate exploration and improve final returns in challenging sparse-reward domains.

Smart contract vulnerability detection with Large Language Models (LLMs) faces three causally linked challenges. First, new vulnerability categories demand parameter-efficient adaptation, since full retraining is prohibitive for sequentially arriving tasks. Second, training per-task adapters on a shared backbone causes catastrophic forgetting of previously learned vulnerabilities. Third, the resulting multiplicity of adapters must be consolidated into a single model, since task identity is unknown at inference time. Each challenge arises directly from the solution to its predecessor, making an integrated framework essential. We propose a three-stage pipeline in which each stage addresses one challenge and feeds into the next. The adaptation stage uses Frequency-Aware Low-Rank Adaptation (FA-LoRA), which performs adaptation in the Fourier domain with per-frequency importance gates, requiring only 0.4% trainable parameters while outperforming standard LoRA and QLoRA. The continual learning stage applies Forget-Aware Replay (FAR), which uses these frequency gates to estimate per-sample forgetting risk via loss dynamics and prioritizes vulnerable knowledge for rehearsal, achieving an average Micro-F1 of 0.8022 across sequential tasks. The deployment stage employs Anchor-Protected Progressive Merging (APPM), which exploits the asymmetric generalization produced by FAR training to identify the strongest-generalizing adapter as an anchor and consolidates all adapters into a single model via anchor-protected weighted merging with frequency-domain gate competition. APPM achieves a Micro-F1 of 0.8085, within 2.7% of the independent per-task upper bound, at a merge cost of 156 ms and no additional runtime memory. Experiments on DIVE confirm the framework effectively addresses all three challenges for evolving blockchain ecosystems.

Prefix caching avoids prefill only when a repeated request returns to a server that still holds the prefix KV. Cache-blind balancing disperses that reuse; fixed affinity preserves it but can overload a server. CacheRoute resolves this tradeoff with a periodic routing plan. It admits high-rate keys to a stable warm set and places their assignments by expected load. Hot keys may use more than one destination, although every key in our primary semi-synthetic aggregate uses exactly one. On Llama-3.3-70B in fp8 across 60 H100 GPUs, CacheRoute sustains 176+/-11 QPS at a 3.5-s p99 SLO, 2.3x the strongest of five baselines. Served KV-cache hit rate rises from 64.1+/-1.3% under cache-blind balancing to 93.2+/-0.5%. A second semi-synthetic aggregate and controlled 8B and burst experiments separate the effects of affinity and placement. Two 32B workloads provide the counterexamples: when affinity recovers too little KV work, its residual load skew reduces or erases the improvement. We therefore recommend gating any deployment with a shadow replay rather than enabling affinity from workload statistics alone.

Computing has been an astonishing success - but the accumulated technical debt exposes us all to huge costs in business and societal risk. For 75 years, we've built systems to prose specifications with test-and-debug development. That works well enough for industry to thrive, but it's an expensive and ineffective feedback loop, and leaves everyone relying on shaky foundations. Now, AI-enabled engineering is amplifying the success by reducing coding costs, but also amplifies the risks, by rapidly increasing technical debt, and by automating detection of the vulnerabilities therein. How can we do better? Research has long pursued mathematical proof of correctness, which, unlike testing, can cover all cases. This too has advanced massively, but it remains hard to apply, both technically and because of a deep-seated cultural disconnect. Instead, we argue for a pragmatic approach to flexible combinations of testing, *specification*, and proof, that provides more effective feedback loops for both AI and human development. Most simply, one can incrementally co-develop executable-as-test-oracle partial specifications alongside conventional prose descriptions, code, and tests. This clarifies design and makes testing much more discriminating. Developers can and should do it today. Or, even better, one can use specifications that support the full gamut of testing, property-based testing, symbolic execution, and proof. This enables a range of intertwined feedback loops, again both for AI and humans, from cheap testing to more expensive proof. However, making it really practical needs *semantics infrastructure*: specifications and tooling for the main programming languages and other abstractions, which we now more-or-less know how to build, but which is not yet in place. We call the community to arms to create and deploy it - to enable a future built on firmer ground.

Natural hazard susceptibility mapping is often constrained by limited labeled data, reducing the generalizability of conventional machine learning and limiting the applicability of complex deep learning models. This study proposes SAGE (Spatially Augmented Graph Embeddings), a structurally informed feature-engineering framework that combines controlled noise-based data augmentation with neighborhood-based graph embeddings to improve prediction under data-scarce conditions. A K-nearest neighbor graph is constructed to derive local spatial statistics, which are reduced using principal component analysis and integrated with environmental covariates and spatial coordinates. The resulting features are used with XGBoost to develop the SAGE-XGBoost model. The framework was evaluated for landslide and wildfire susceptibility mapping. SAGE-XGBoost consistently outperformed conventional and spatially explicit machine learning models. Compared with Spatial XGBoost, it achieved an absolute improvement of above 33 percentage points across the two case studies. The model reached AUC values of approximately 0.97 for landslide susceptibility and 0.95 for wildfire susceptibility. Feature importance analysis confirmed the contribution of graph embeddings to prediction, while their integration improved spatial coherence and reduced local noise amplification. Overall, SAGE-XGBoost provides an efficient and transferable alternative to deep representation learning for environmental hazard assessment and other geospatial prediction tasks under limited supervision.

Large language models (LLMs) compression reduces deployment costs, but standard aggregate metrics like perplexity and accuracy often mask underlying behavioral shifts. In this work, we systematically evaluate 3 LLMs across 11 compression methods to investigate the effects of compression on knowledge retention, model confidence, and social bias. We find that compression disproportionately reduces the relative retention of head knowledge compared to tail knowledge. Furthermore, compressed models often remain substantially confident in their incorrect answers on newly lost knowledge. Finally, we demonstrate that stable aggregate bias scores can conceal substantial, opposing shifts in stereotypical preferences across demographic subgroups. Together, these findings reveal asymmetric behavioral changes that aggregate performance measures fail to capture, highlighting the need for granular evaluation of compressed models before deployment.

Latent reasoning has advanced multimodal reasoning through a two-stage training paradigm: (1) a helper image is encoded into latent tokens to teach visual chain-of-thought during a supervised fine-tuning (SFT) stage, and (2) these latent tokens are further refined with reward feedback during a reinforcement learning (RL) stage. In this paper, we identify two key limitations of this framework, one in each stage. First, the SFT stage typically relies on an off-the-shelf vision encoder to encode the helper image, yielding suboptimal latent representations that may not be well aligned with the downstream reasoning task. Second, existing RL methods treat the latent component only through deterministic regularization, which constrains policy drift but does not create alternative latent trajectories for exploration. To address these limitations, we propose Scaffolding Minds. Our approach learns a dedicated scaffolding encoder that provides an optimized target in latent space, and learns both the mean and variance of the RL sampler. We further show that these two improvements are complementary, together yielding substantial gains over strong baselines. Empirically, our method improves over the strongest latent-reasoning baseline by +9.5% on FrozenLake spatial planning, with the gain widening to +19% at 32x32 grid map, and by +5.2% on average across nine visual-centric reasoning benchmarks.

Agentic language models repeatedly encode tool and skill schemas that recur across requests in different combinations and orders, preventing standard prefix caching from reusing their key--value (KV) states. We introduce \textbf{ReCache}, a framework for independently caching resource representations while reducing their inference-time computational and memory overhead. Resource-wise attention removes cross-resource interactions and assigns resource-local positions, producing composition-invariant KV blocks. ReCache then restricts resource visibility to contribution-selected layer--KV-head-group routes and retains only invocation-critical fields through structural and semantic pruning. We evaluate ReCache on a benchmark assembled from seven public tool- and skill-use datasets, including resource-disjoint tests. Resource-wise attention matches dense invocation performance (82.3\% versus 82.4\% Inv-F1) while providing a 3.655$\times$ time-to-first-token speedup. The complete framework reduces allocated KV-tensor memory by 92.43\% and accelerates attention by 1.423$\times$. These results show that separating reusable schema encoding from selective resource access substantially reduces agentic inference costs with limited effectiveness loss. The code is available at https://github.com/EIT-NLP/ReCache.

Choosing tensor-parallel (TP) degrees and replica counts for an LLM serving fleet is difficult because performance is not monotonic in TP and the feasible choice can change with load. Exhaustive profiling resolves this uncertainty, but measures many configurations that do not affect the final resource allocation. We present FleetSieve, which selects measurements according to their expected effect on a resource-coupled, SLO-aware fleet decision. FleetSieve models capacity and tail latency jointly, compares conservative and optimistic allocations, and stops when their remaining decision gap is below a specified tolerance. On a fixed H100 measurement grid for a 31B-parameter open-weight model, FleetSieve reaches the oracle aggregate decision using 22,200 GPU-seconds, 6.9% less than uniform random profiling in the fixed comparison. Across 200 random reveal orders, its mean saving over random profiling is 5.4% (95% bootstrap CI: 3.5-7.2%). The fixed-comparison saving is 21.5% for Chat, while FleetSieve does not use the fewest GPU-seconds for Code. Joint capacity and tail modeling also avoids selecting a configuration whose 46.4-second completion p99 violates a 30-second SLO. In a 16-GPU allocation, an incorrect sparse-profile decision loses up to 1.93 requests/s and 12.4 percentage points of max-min fulfillment. Boundary repeats and BurstGPT measurements support the observed load-dependent tail-latency mechanism.

This study demonstrates a rationally enriched Chebyshev (REC) trunk for deep operator network (DeepONet) surrogate models of singularly perturbed and high-Péclet transport problems whose solution profiles are characterized by thin localized boundary or wall layers. The REC trunk combines Chebyshev polynomial dictionary elements with rational dictionary elements constructed using the adaptive Antoulas-Anderson (AAA) algorithm. Over five independent training runs, the resulting REC-trunk DeepONet is evaluated against a vanilla DeepONet and a Chebyshev-trunk DeepONet whose prescribed dictionary consists only of Chebyshev polynomials across three problems whose singular perturbation parameters are diffusion-to-advection ratios: a singularly perturbed scalar boundary-value problem (BVP), the thermal entrance problem with a prescribed wall temperature, and the concentration entrance problem with an absorbing wall. Across the held-out test profiles, the REC-trunk DeepONet improves over the vanilla DeepONet and remains comparable to the Chebyshev-trunk DeepONet in predicting the scalar profile, with its clearest advantage over the Chebyshev-trunk DeepONet appearing when the perturbation parameter lies between $1.00\times10^{-4}$ and $1.78\times10^{-4}$, where it reduces the profile-error metrics by up to $19.5\,\%$ relative to the Chebyshev-trunk DeepONet. In predicting the wall-normal temperature and concentration profiles, the REC-trunk DeepONet reduces the profile-error metrics by up to $60.2\,\%$ and $32.2\,\%$ relative to the vanilla and Chebyshev-trunk DeepONets, respectively, while suppressing artificial near-wall oscillations as the Péclet or mass-transfer Péclet number ranges from $10^{2}$ to $10^{4}$.

Autonomous agents for machine learning experimentation must navigate heterogeneous repositories, repair training pipelines, and evaluate candidate improvements under realistic compute constraints. Existing benchmarks only partially capture these conditions. We introduce DeltaML-Bench, a benchmark comprising 48 tasks sourced from research papers that require agents to improve published baselines within imperfect, open-source repositories. We evaluate GPT-5 and Claude Sonnet 4 with a standard Modular agent and a search-based ARG scaffolding. In the 4 x 6h allocation, ARG raises GPT-5's per-run success rate from 9.4% to 33.9%; in the 2 x 12h allocation, GPT-5 ARG reaches 49.0%. Modular configurations exhibit specification gaming rates as high as 47.9%, while no gaming is observed in the evaluated ARG configurations. These results indicate that scaffolding design and integrity checks are important considerations when deploying agents for autonomous ML experimentation.

As LLM-based agents are deployed for longer and higher-stakes tasks, their memory systems continue to have crucial gaps. While existing memory benchmarks focus largely on recall-shaped tasks, we argue an effective memory system must track the evolving state of the world; as facts, constraints, and decisions are revised over a long interaction, answers must reflect the current state and not a superseded one. We define this capability as state tracking and instantiate it in StateMemBench, a benchmark of 234 multi-session scenarios spanning two conversation-length regimes. Its closed-pool grading scores whether an answer reflects the current state, the superseded state, or fails otherwise, separating state-tracking failures from other errors by construction. Our analysis shows that this task is challenging for existing memory systems, retrieval-augmented baselines, and long-context baselines. We then present StateMem, a state-first memory method that explicitly tracks supersession and relational dependencies, and show it improves current-state accuracy over the strongest same-backbone baseline by 1.8x (0.205 -> 0.363) on DeepSeek-V4-Flash and over the strongest memory system by 1.6x (0.149 -> 0.233) on Qwen-3.5-9B, while remaining competitive with the long-context baselines. Finally, we show the same state approach can be applied as a lightweight single-call wrapper over existing memory systems, lifting current-state accuracy by +32 to +67 points on StateMemBench across six memory and retrieval backends. A length- and cost-matched control attributes +15 to +32 of those points to state structure rather than added context.

Self-normalized concentration inequalities are standard tools in bandit and reinforcement-learning analyses. A widely used weighted extension claims an analogous time-uniform guarantee for discounted least-squares estimators in non-stationary problems. A simple scalar Gaussian counterexample with a fixed parameter shows that the claimed bounded radius is crossed with probability one. For fixed discount and regularization parameters, we further show that, when $δ\leq1/2$ and $T/δ$ is sufficiently large, any deterministic anytime boundary valid uniformly over the stated conditionally sub-Gaussian model class must be at least of order $R\sqrt{\log(T/δ)}$ at some time by horizon $T$; for nondecreasing boundaries, this order is required at time $T$. We identify the proof error: different terminal times use different Gaussian mixing distributions, so the fixed-time mixtures do not form one supermartingale, and the stopping-time argument does not repair this failure. Finally, we show that the weighted inequality remains valid at each fixed deterministic time, give valid finite- and infinite-horizon corrections, and discuss consequences for downstream analyses.

Physics-informed neural networks (PINNs) embed governing partial differential equations directly into the training loss, offering a promising alternative to costly CFD solvers for unsteady flows. Yet the growing list of techniques proposed to improve PINN training is typically validated one at a time, leaving open whether these techniques actually compose. We study this question in depth on the DFG/Schafer-Turek unsteady cylinder wake benchmark. In isolation, nearly every technique performs no better than an untreated baseline. However, combining periodic (SIREN) activations with causal weighting unlocks a previously inaccessible regime, reconstructing velocity and pressure fields to within 4.1% average relative L2 error against an OpenFOAM reference solution. Adding further techniques instead causes catastrophic performance degradation, demonstrating that individually effective PINN interventions can interact nonlinearly and that more elaborate training recipes are not necessarily better.

Execution feedback is often treated as a self-verifying signal for improving LLM-generated tests. However, when generated inputs are executed on a single accepted program and its outputs are used as ground truth, invalid or underspecified inputs can create spurious fault detections and apparent evolutionary gains. We audit this failure mode in feedback-driven test generation using 142 development tasks, 114 locked external tasks, and 138 held-out tasks, with two code models, three seeds, and fault-cross-fitted real submissions. On external inputs for which three accepted implementations agree, generated outputs match the panel on only 27.79% and 50.12% of cases. A single-reference oracle inflates the measured gain from evolution by 9.46-14.85 percentage points; after auditing, equal-budget independent resampling outperforms mutation-based evolution by 6.01-18.83 points. We further compare a genuine three-round feedback loop with a density-matched placebo. External Real-Placebo differences are +0.13 and -0.50 points, while held-out differences are +1.99 and +0.28 points and do not provide robust evidence of fine-grained feedback benefit. A blinded semantic audit by two software engineering doctoral students classifies 94.41% of panel-disconfirmed inputs as invalid but 3.60% as valid, showing that panel disagreement is informative but not semantic proof. We propose an audit-and-placebo protocol that separates verifier artifacts, interaction scaffolding, and grounded feedback credit in evaluations of self-evolving test generators.

Scientific data are increasingly used by AI agents, yet existing dataset representations provide limited support for autonomous discovery, interpretation, and invocation. This limitation stems from the fragmentation of scientific data across heterogeneous repositories and from dataset representations designed primarily for human use. To address this limitation, we introduce the Scientific Data Skill (SciDSK), an agent-ready representation that packages dataset-specific knowledge and operational guidance as a reusable agent skill. A SciDSK integrates dataset descriptions, scientific context, file organization, usage procedures, quality checks, and provenance information while retaining the underlying data in its original repository. We define a structured SciDSK specification and develop a systematic construction pipeline that grounds each SciDSK in authoritative dataset records and associated supporting materials. We further establish the Scientific Data Skill Bank, a unified platform that publishes SciDSK resources across six scientific disciplines and supports package access, persistent identification, and traceability to source datasets. We evaluate SciDSK through a retrieval benchmark for dataset discovery and controlled cases for dataset interpretation. The results show that SciDSK improves agent-driven dataset discovery and provides more precise and actionable support for dataset interpretation. These findings support the value of organizing dataset-specific knowledge in an agent-ready representation.

Large language models (LLMs) offer a scalable approach to social simulation, but their credibility depends on how agents are constructed. Existing methods can partially reproduce population-level patterns, yet often fail to capture human-like diversity. Our analysis shows that static-profile agents exhibit stronger demographic separation and within-group compression than humans, a pattern consistent with identity essentialism: demographic labels can encourage models to treat group-average tendencies as individual traits, homogenizing responses within groups. We argue that this limitation arises from two related factors: sparse, static agent representations and the limited ability of prompt-only memory to persistently integrate experience. Inspired by complementary memory systems, we propose LifeMem, a longitudinal memory framework that combines structured life-event retrieval with agent-specific parametric memory for experience integration. Experiments on Add Health and Understanding Society with three LLMs show that LifeMem improves alignment with human data in terms of response distributions, overall and within-group diversity, and patterns of within-person response change across life stages. These findings highlight the value of longitudinal life-event memory for constructing more faithful and dynamically evolving social agents.

LLM reasoning is stochastic, and so understanding a model requires grappling with the distribution of reasoning chains that it might produce for a given question, i.e., its uncertainty. Resampling-based analyses characterize this distribution, revealing which steps of a rollout determine how the model arrives at its answer. However, a major limitation of these approaches is that resampling text sequences at every token or sentence in a reasoning chain is very costly. Our work strives to make resampling analysis more computationally efficient, while also shedding light on an important scientific question: what is the right statistical model for explaining uncertainty dynamics in text generation? We show that when resampling many reasoning chains, uncertainty dynamics converge to stable patterns, and noise is largely an artifact of sampling rather than an LLM's sensitivity to each individual token or reasoning step. We develop a statistical model for smoothing noisy low-sample rollout data to better approximate high-sample data, allowing us to significantly cut sampling costs.

Direct Preference Optimization (DPO) has emerged as an effective approach for aligning large language models (LLMs) with human preferences. However, its adaptation to multimodal settings remains unexplored. Through representational analysis, we identify a key limitation in multimodal preference optimization, which we term visual insensitivity: models often fail to distinguish between images and those with critical visual context removed. Our theoretical analysis further uncovers two manifestations of this problem, namely Across-Image Insensitivity and Within-Image Insensitivity. To address these challenges, we propose Perception-Enhanced Alignment DPO (PEA-DPO), a framework for multimodal LLMs alignment, which explicitly leverages visual preference signals to overcome visual insensitivity. We further provide a theoretical analysis demonstrating that PEA-DPO provably mitigates both failure modes. Empirical results demonstrate that PEA-DPO enhances sensitivity to visual context while preserving the language modeling capacity of the base model. Evaluations across three hallucination benchmarks using MLLMs of varying scales show that PEA-DPO effectively mitigates visual insensitivity, achieves stronger multimodal alignment, and substantially reduces hallucinations.

While entropy regularization is widely used to stabilize and accelerate Natural Policy Gradient methods, its ability to yield faster convergence rates for the unregularized objective remains underexplored. Existing analyses often rely on double-loop architectures and invoke a linear entropy penalty. To bridge the gap between theory and practice, we analyze a single-loop, entropy-regularized Natural Actor-Critic algorithm under compatible linear function approximation. By training an uncentered critic, our critic tracking can remain stable even as the training policy approaches determinism and the Fisher information matrix degenerates. We focus on two primary regimes for the optimization landscape: a Stochastic Regime, where we fuse coupled actor-critic updates into a joint Lyapunov recurrence, and a Deterministic Regime, where we pivot to a Policy Mirror Descent framework to circumvent the collapse of Euclidean geometry. By exploiting a positive Minimal Action Gap in the unregularized Markov decision process, we introduce an Exponential Translation mechanism that maps the regularized gap to the unregularized one up to an exponentially decaying tail. By tuning the fixed temperature, our algorithm achieves accelerated unregularized convergence rates, up to approximation-error terms: $\tilde{\mathcal{O}}(T_{total}^{-1})$ in the Stochastic Regime, and $\tilde{\mathcal{O}}(T_{total}^{-2/3})$ for the average iterate alongside $\tilde{\mathcal{O}}(T_{total}^{-1/3})$ for the last iterate in the Deterministic Regime. Here, $T_{total}$ denotes the total number of stochastic critic updates (or Monte Carlo rollouts). Furthermore, in the tabular setting, our positive-action-gap analysis yields a $\tilde{\mathcal{O}}(T_{total}^{-2/3})$ average-iterate rate, surpassing the $\mathcal{O}(T_{total}^{-1/2})$ worst-case statistical barrier that applies without a positive action margin.

We study landscapes for complex-parameterized networks. Our approach is motivated with an information-theoretic manifold perspective of the parameter and via classical optimization guarantees although of complex geometric variety such as through Dolbeault asymptotics. The descent path admits a Kähler information metric under a cross-entropy via the Wirtinger Hessian on the log-likelihood potential. We restrict attention to a descent update rule with natural gradient descent via a differentiated loss scaled by the inverse metric, so the descent path remains in the holomorphic tangent bundle. We emphasize Calabi-Yau information manifolds which profane theoretical guarantees via an ill-curvature-conditioned landscape. Under a Calabi-Yau metric, specifically in a non-compact setting with a global potential so defined geometrically rather than invoking the topological requirements of the Calabi conjecture, a wedged nowhere-vanishing holomorphic form is the top exterior product of the Kähler form up to constants, yielding a constant determinant condition. Under a fixed determinant, a metric almost low rank up to an eigenvalue tolerance implies a blow-up effect. Moreover, it has been discovered that negative curvature subverts the loss landscape, specifically sectional curvature, so we expand on this and draw interconnections to negative-definite Ricci curvature. Our arguments primarily exist in a geometric analytic modality, although we establish roots in deep learning theory such as through asymptotics at initialization and connections through failure modes of neural network guarantees under vanishing and negative Ricci curvature.

Recent studies suggest that video generation models can exhibit certain forms of zero-shot visual reasoning through generated frames. Yet reliable evaluation remains challenging: benchmarks should adopt inputs aligned with the visual priors of current video models, require valid evolving processes rather than only plausible final states, and calibrate task difficulty to remain challenging yet partly feasible. To this end, we introduce VGI-bench, containing 27 tasks and 810 instances, organized by a two-level taxonomy of task domains and skill tags for fine-grained evaluation of visual reasoning capabilities of video generation models. Our evaluations show that current generative systems can solve a subset of visually grounded reasoning tasks, but remain far from reliable, with even the strongest model, Seedance~2.0, achieving only 51.0% under our evaluation criteria. Our analysis further explore the output failure modes, input condition sensitivity, performance transfer boundary from synthetic fine-tuning, and internal denoising perspective revealing limited self-correction, where later steps mainly refine early hypotheses rather than correct reasoning errors. We hope VGI-bench will help stimulate the development of next-generation video generation models. We will release our code and data.

Large Language Models (LLMs) are increasingly deployed in high-stakes applications, yet their tendency to generate toxic, harmful, or policy-violating content poses significant risks. Detecting these unsafe outputs efficiently in a black-box manner remains an open challenge. In this paper, we extend a recently proposed dynamical systems framework designed for hallucination detection to LLM safety classification. By projecting both prompts and responses into high-dimensional embedding spaces and fitting separate Koopman-based predictive models for safe and unsafe regimes, we classify new outputs using a new differential residual score that compares prediction errors of the safe and unsafe regimes. A key contribution is the incorporation of the prompt and response embedding dynamics, yielding fitted Koopman operators that capture crucial interaction patterns. We evaluate our black-box method across three safety benchmarks using three embedding models. Our results show that incorporating prompt embeddings yields consistent improvements, particularly for interaction-dependent violations when paired with causal decoders (e.g., in Llama-3), while response-only violations benefit more from dense semantic embedding representations. These findings opens the door for using dynamical systems to analyze AI systems rather than the dominant paradigm of using AI to model dynamical systems.

Acute exacerbation of chronic obstructive pulmonary disease (AECOPD) can worsen rapidly, making timely prediction a clinical priority. Most existing machine learning approaches rely on episodically collected clinical variables, introducing delays that limit their practical utility in home monitoring settings. Home ventilators offer a lower-latency alternative, producing a near-continuous record of respiratory status during daily use. However existing ventilator-based approaches either compress the waveform into handcrafted features or focus primarily on binary risk classification, leaving the timing of an impending event unresolved. In this paper, we present a two-stage framework that operates directly on raw pressure and flow waveforms from the most recent seven days of home ventilator use. The first-stage classification model identifies patients at high risk of a severe exacerbation. The second-stage regression model then estimates how many days remain before the event occurs. Our experimental results demonstrate that the two-stage model outperforms traditional baseline models on both risk classification and time-to-event estimation, with our selected Stage 1 classifier achieving F1 = 0.91 and our Stage 2 regression model achieving RMSE = 1.00 days and R^2 = 0.76, giving clinicians both an early warning and actionable lead time before a severe exacerbation occurs.

Drafting a new Magic: The Gathering expansion begins before any pick from it has been observed: the complete card list is public, but the draft logs that supervised pick models train on do not yet exist. We study this day-zero regime directly. DraftFM is a discrete-choice policy that scores exactly the cards available in the current pack, conditioned on the drafted pool and the state of the draft. Every card enters as a frozen 775-dimensional function of its public card record, structured features and a fixed text embedding, with no card identities, set identities, or usage statistics anywhere in the model, so an unseen card is scored by the same machinery as a familiar one. A 1.6-million-parameter network fitted on 149 million human picks from 29 expansions predicts held-out picks in three expansions withheld in their entirety, reaching 50.8%, 60.4%, and 56.7% top-1 agreement, where uniform chance at the opening pick is about 7%. Refitted on all 32 observed expansions, the same architecture produced a card ranking for the then-unreleased set The Hobbit, sealed with its complete cryptographic provenance and published roughly 36 hours before the set became draftable on MTG Arena. The sealed ranking agrees with six independent expert reviewers roughly as much as those reviewers agree with one another. Evaluation against realized outcomes is committed to a follow-on note, whatever it shows.

Persistent memory can personalize an LLM agent, but an incorrect durable update can silently distort future behavior. We study the memory-clarification boundary: whether interaction-derived information should be persisted, used only in the current context, re-verified, or clarified with the user. MCB contains 140 primary scenarios, split into 70 development and 70 held-out items, plus a separate 70-item contrast set. It evaluates both action labels and structured tool-call selection. Two non-authors independently label the 70 held-out primary and 70 contrast items (97.1% agreement, Cohen's kappa = 0.962); a blind third resolves four disagreements, replacing eight author labels by non-author majority. Across Claude and Qwen, models verify changing facts more reliably than they ask users to resolve ambiguity. Bare Qwen asks on 0/12 clarification items while verifying 12/18 freshness items. Few-shot prompting raises accuracy from 0.557 to 0.771 (paired delta = +0.214, Holm-adjusted exact McNemar p_H = 0.002), yet clarification recall remains 0.333. The policy prompt reduces erroneous persistence from 0.243 to 0.100 (p_H = 0.038), although its accuracy gain is not significant. Label-tool agreement is 57% for each Claude model and 23% for Qwen; Qwen accuracy falls from 0.557 to 0.343 (p_H = 0.047). Memory evaluation must test both stated decisions and tool-call choices.

Financial AI systems often train information extractors on one textual register and deploy them across filings, news, and user-generated content, while standard F1 scores do not indicate which predictions remain safe to automate when the input distribution changes. We study confidence estimation and selective prediction for financial named entity recognition (NER) on a three-tier stress test spanning SEC filings, financial news, and general-topic social media as an extreme out-of-domain condition. We evaluate a BERT tagger and LoRA-tuned Qwen2.5-0.5B/1.5B models using five inference-time confidence signals, three training seeds, and bootstrap intervals. Confidence rankings themselves change under distribution shift: whole-output probability is the strongest in-domain error detector but deteriorates out of domain, whereas entity-span probability and self-consistency are more robust; self-consistency is also better calibrated without post-hoc fitting. Abstention reduces sentence error from 34.3% to below 2% on the highest-confidence 40% of in-domain inputs and remains useful on financial news, but recovers no usefully large clean subset under the extreme social-media shift. These results motivate a staged deployment strategy that detects severe distribution shift upstream before applying prediction-level confidence gating.

Streaming autoregressive diffusion models enable real-time, long-horizon video generation, but their training objectives optimize local frame prediction rather than the geometry and dynamics of a coherent world: long rollouts accumulate geometric drift and degrade into static or unnatural motion. Recent bidirectional approaches address this problem using rewards signals built upon 3D Gaussian-Splatting reconstruction. However, a single rigid 3d reconstruction cannot model a dynamic scene, so this critic penalizes genuine object motion as reconstruction error and is maximized by freezing the video. This shortcut is especially detrimental in the AR setting, where each chunk can propagate an already-static configuration. In this work, we propose Stream4D, which replaces the static critic with a feed-forward 4D reconstruction reward that explicitly models scene dynamics, allowing coherent motion to receive high consistency rewards. To further guide motion magnitude and quality, we add a motion prior that rewards natural scene-flow magnitude while penalizing jitter and non-rigid artifacts. Our final recipe combines these two terms with a lightweight perceptual anchor. Across various autoregressive video backbones and various generation horizons, Stream4D improves 4D reconstruction quality, preserves motion more effectively, and achieves higher human-aligned preference. Project page: https://banyuanhao.github.io/Stream4D/

This paper addresses the issue of the significant labor required to test interview dialogue systems. While interview dialogue systems are expected to be useful in various scenarios, like other dialogue systems, testing them with human users requires significant effort and cost. Therefore, testing with user simulators can be beneficial. Since most conventional user simulators have been primarily designed for training task-oriented dialogue systems, little attention has been paid to the personas of the simulated users. During development, testing interview dialogue systems requires simulating a wide range of user behaviors, but manually creating a large number of personas is labor-intensive. We propose a method that automatically generates personas for user simulators using a large language model. Furthermore, by assigning personality traits related to communication styles when generating personas, we aim to increase the diversity of communication styles in the user simulator. Experimental results show that the proposed method enables the user simulator to generate utterances with greater variation.

Functionalities of ferroelectric materials are governed by the spatial organization and coupling of polarization, strain, lattice rotation, and structural order accessible via atomically resolved scanning transmission electron microscopy (STEM) images. Quantitative interpretation of atomic-resolution STEM data has conventionally relied on locating atomic columns and converting their fitted coordinates into local structural descriptors. Here, we develop a field-based approach in which atomic-resolution images are represented by spatially varying latent Bragg fields, whose amplitudes and phases provide continuous maps of crystalline order, lattice displacement, strain, rotation, and mode-specific residual structure. The observed atomically resolved images are decoded from the latent fields. We apply this framework to image series of Sm-substituted BiFeO3 spanning 0-20% Sm and crossing the composition-driven boundary between the R3c ferroelectric phase and the orthorhombic, nonpolar Pnma phase. Conventional atom-resolved parameterization is used as an independent validation, showing that reconstructed Bragg amplitude tracks local atomic-column intensity and that field-derived shear reproduces unit-cell angular distortions obtained from atom fitting. The combined analysis reveals a systematic evolution from extended ferroelectric domains at low Sm concentration, through the appearance and growth of localized regions with period-doubled Pnma order at intermediate compositions, to a connected Pnma-dominated state at high Sm content. The period-doubled order is accompanied by enhanced shear and lattice rotation and by progressive reorganization of the ferroelectric domain structure. These results establish latent-field reconstruction as a physically interpretable complement to atom finding and provide a unified framework for resolving composition-driven phase evolution in ferroic materials.

Training fast generators on new domains with limited data remains challenging for two reasons. First, adapting a pretrained diffusion or flow model to a new domain leaves its costly multi-step sampling unaddressed, and existing acceleration methods are tied to the source parameterization--$ε$, $x$, $v$, or $u$--leaving heterogeneous pretrained models with no common acceleration target. Second, while adversarial refinement is proven effective for few-step quality, it is formulated only for instantaneous-velocity flows, not for the finite-interval average velocities that MeanFlow (MF) models predict. We address both problems. We propose MeanFlow-Transfer, which maps heterogeneous source outputs into a shared velocity representation, uses it to initialize an MF generator from the source weights, and optimizes an MF objective on the target domain. This unifies adaptation and acceleration in a single training loop across a broad range of pretrained models. We then introduce Continuous Adversarial MeanFlow, a post-training stage that extends continuous adversarial flow models from instantaneous velocities to MF's finite-interval average velocities. CAMF contrasts changes in a learned potential between real and predicted interval endpoints, recovering fine detail that MF regression averages away, and reduces to the instantaneous criterion in the vanishing-interval limit. Adapting four ImageNet-based source models--DiT ($ε$), SiT ($v$), JiT ($x$), iMF ($u$)--to five target domains, MF-T with CAMF matches or exceeds the fine-tuned teacher in FID and FDD at up to $125\times$ fewer Neural Function Evaluations (NFEs), while CAMF improves MF-T's few-step FID by $29\%$ on average.

Learning-based global point cloud registration has achieved remarkable progress, yet its reliance on geometric representations makes existing methods sensitive to variations in point density, scan pattern, viewpoint, and sensor characteristics. We propose CVSD-Reg, a robust global LiDAR registration framework that distills visual semantic priors from a vision foundation model into LiDAR representations. In Stage 1, a Point Transformer V3 student learns from a frozen DINOv2 teacher through contrastive distillation and spherical-manifold alignment, which preserves the hyperspherical geometry of the teacher embedding space. Self-supervised InfoNCE consistency and soft $\mathrm{SE}(3)$ invariance further encourage viewpoint-robust descriptors. In Stage 2, the distilled representation is adapted to registration through correspondence learning, density-aware point-dropout augmentation, and end-to-end pose optimization. With a single checkpoint, CVSD-Reg generalizes to both single-sensor and zero-shot cross-sensor scenarios without sensor-specific adaptation and remains entirely camera-free at inference. On KITTI, nuScenes, and HeLiPR, CVSD-Reg achieves strict success rate (SR@0.5\,m/$1^\circ$) of 97.7$\%$, 99.0$\%$, and 99.3$\%$, respectively, including 97.3$\%$ on sparse 16-beam Velodyne scans. It outperforms state-of-the-art geometric registration methods by up to 44.0 percentage points without requiring camera inputs or post-hoc ICP refinement.

Retrieval-augmented generation (RAG) improves language-model responses by grounding generation in external passages, which comes with overhead: retrieved context lengthens the prompt, increasing prefill work, KV-cache footprint, memory traffic, latency, and energy. Context compression offers a natural remedy by pruning retrieved text before generation. However, state-of-the-art context-compression methods are typically used with a fixed compression budget, or with the rate selected offline and then applied at inference time. This static view ignores both workload variation and the live state of the edge device. On an edge SoC, compression is not free: the compressor itself runs on the same SoC and consumes latency and energy that can offset any generation savings. This paper proposes a vision for telemetry-informed adaptive compression in edge RAG, grounded in experimental evidence. We characterize the compression tradeoff on the NVIDIA Jetson AGX Thor using Llama and Qwen generators, Natural Questions and HotpotQA datasets, and LLMLingua-2 compression. Our measurements show that generation dominates the RAG budget for larger models, reaching roughly 90% of per-query latency and 91% of GPU energy for 7B-8B generators. Exploring the impact of the compression rate reveals an adaptive operating region: mild compression can miss energy opportunities, and overly aggressive compression can hurt inference quality. Intermediate compression can reduce GPU energy by up to 53.2%, and SoC energy by up to 48.2%, with negligible quality loss. We argue for runtime policies that dynamically manage compression, guided by workload features and edge telemetry.

Many real-world AI systems represent entities, behaviors, and structured information using discrete machine-native symbols rather than natural language. While these representations are compact and preserve task-relevant structure, they lie outside the linguistic token space of pretrained large language models (LLMs), creating a fundamental divide between language modeling and structured prediction. We introduce UniLang, a unified generative framework that bridges this divide by extending pretrained LLMs to treat machine-native symbols as first-class generative units alongside natural-language tokens. UniLang expands the LLM's vocabulary and embedding space with grounded machine-native representations, enabling textual and symbolic tokens to be jointly modeled and generated under a single autoregressive objective. This unified interface allows pretrained LLMs to directly operate on machine-native representations without requiring them to be verbalized as natural language or relying on task-specific architectures. We evaluate UniLang on two structurally distinct tasks, sequential recommendation and legal precedent prediction, spanning different domains and types of structured prediction. Across both tasks, UniLang consistently outperforms strong baselines, demonstrating a path toward extending pretrained LLMs beyond language and using them as a common generative modeling backbone for heterogeneous machine-native representations.

AI clones that imitate a specific person typically reproduce what the person says and how they sound, but not how they listen. We investigate whether adding multimodal listening behaviors gives such a clone more presence and authenticity. We integrated verbal backchannels and head nodding, driven by real-time prediction models, into an AI clone equipped with voice cloning and LLM-based responses. In a within-subjects study (N=35), adding these behaviors significantly improved the perceived attentiveness of the avatar, the sense of talking with the real person, and the feeling of co-presence. These results indicate that AI clone fidelity should extend beyond voice and response content to include interactive listening behavior.

Stock market analysts and investors face a daily challenge: too much financial news, too little time. Manually reading and synthesizing hundreds of company-specific articles is impractical, yet missing key information can directly affect investment decisions. This project, conducted at George Washington University in Fall 2023, explores whether Large Language Models can automate this process reliably. We built a pipeline that pulls news articles from the News API, company background from Wikipedia, and stock price data from Yahoo Finance for ten major companies (AAPL, MSFT, GOOGL, AMZN, META, TSLA, JPM, NVDA, WMT, DIS). Because LLMs cannot directly process numerical tables, we developed a simple but effective template that converts stock data into natural language narratives. We then tested two summarization approaches (Summarize Chains and Retrieval-Augmented Generation with FAISS) across three open-source models (Falcon-7B-Instruct, DistilBART-CNN-12-6, BART-Large-XSum) for news, and GPT (text-davinci-003) for stock summaries. Falcon-7B with Summarize Chains gave the best results, covering all news events accurately and coherently. RAG, while promising in theory, caused severe repetition in Falcon and hallucinated facts in BART-Large when k was large. Both LLM-based approaches outperformed a simple Lead-3 baseline on ROUGE-1. We also built a Streamlit dashboard for interactive stock visualization. The work was done in Fall 2023, before RAG-based financial tools became widespread, and the failure modes we document, particularly hallucination under RAG in smaller models, remain relevant today.

Garbage-collector selection is a low-effort configuration decision that can influence both the performance and energy consumption of Java applications. However, it remains unclear whether aggregate energy rankings among collectors generalise across heterogeneous applications, workload intensities, and JDK distributions. This study presents a controlled empirical evaluation of Serial, Parallel, and G1 garbage collection across three Java applications, three workload intensities, and two JDK distributions: OpenJDK and Oracle JDK. Using EnergiBridge, processor-package energy consumption and execution time were measured and complementary energy-performance metrics derived. Across all evaluated configurations, Parallel recorded the lowest mean energy consumption (839.8 J), followed closely by Serial (857.6 J) and G1 (969.0 J), but an RCBD ANOVA did not establish a statistically reliable collector effect. Workload intensity, by contrast, was a significant driver of energy consumption, with heavy workloads consuming substantially more energy than light or medium workloads regardless of collector. Energy consumption showed a moderate positive association with execution time (r = 0.33), indicating that longer-running configurations tended toward higher energy use, though the relationship was far from proportional. No statistically significant energy difference was found between OpenJDK and Oracle JDK. Overall, the results do not support a universally energy-optimal garbage collector; instead, workload intensity emerges as the more reliable lever for managing Java energy consumption, and collector selection should be treated as an application-specific tuning decision supported by measurement on the target system rather than aggregate rankings alone.

Prosodic cues can convey task-relevant information that alters the trajectory and outcome of a task-oriented dialogue, even when the words themselves remain unchanged. Yet existing benchmarks typically evaluate prosodic perception, response appropriateness, and task-oriented dialogue in isolation, making it difficult to test whether prosodic evidence changes downstream decisions. We introduce Hear2Act, a unified evaluation protocol for text and spoken assistants with 480 persona-grounded scenarios, hidden user concerns, and objectively verifiable outcomes. For each scenario, we keep the task and user needs fixed while varying whether the same concern is conveyed explicitly in words or primarily through prosody, and evaluate decisions under transcript, audio, and concern-state access. Using Hear2Act, we evaluate two audio-capable LLMs. Under Prosody-mediated feedback, adding audio to the transcript changes the average optimal-solution rate only from 14.6% to 15.3%. In contrast, when models infer the concern status from audio, represent it in text, and use it for next-action selection, the rate rises to 39.6%, close to 40.7% with the ground-truth state. This contrast, however, largely disappears under Explicit lexical feedback, where the concern is verbally mentioned in the utterance. Together, these results show that prosody matters when lexical evidence is insufficient, and that audio-capable LLMs can recover information from speech but do not reliably carry it into action without an explicit intermediate representation.

Persistent intelligent systems require the ability to learn continually, but current machine learning approaches face significant challenges in this area compared to biological learning systems. Machine learning algorithms typically trade off retention of previously learned information and adaptation to new or changing data patterns. When continual learning capabilities are absent, algorithms must undergo retraining using the entire data set, an approach that becomes impractical when original training data are unavailable due to storage constraints, financial or computational costs, or privacy restrictions. However, biological animals can learn continually, without experiencing catastrophic forgetting. This paper attempts to build a high-level framework for how animals learn and preserve knowledge by modelling neural components and states that are known to be related to memory consolidation. We focus on three concepts: experience replay, REM sleep, and bilaterality. We propose 4MAS (4 Module Awake/Sleep), a novel macroarchitecture demonstrating how machine learning models might benefit from asymmetric hemispheres, each with their own long- and short-term memory mechanisms, and how a period of sleep between incremental learning tasks might benefit memory consolidation. Finally, we present results showing that our architecture achieves competitive results on the Split-MNIST, Split-Fashion-MNIST and Split-CIFAR-100 datasets, with 98.3%, 84.9%, and 29.29% accuracy respectively.

Symposium is a formal framework and practical implementation to record the operation of AI agents deployed by small scientific research communities. Symposium provides long-term, immutable histories of agent-driven research activity, leaving auditable trails of analyses, hypotheses, data, and scientific discourse. This shared record of published artifacts enables agents to build on prior work and preserves the evidence researchers and agents need to make purpose-dependent trust assessments. Symposium captures scientific argument, including structured claims, fine-grained evidence citations, assumptions, and explicit declarations of what material may and may not be used as evidence. Symposium differs from AI co-scientist agents or integrated AI research environments; it is a framework that separates a scientific community's durable history from the agents and other systems that operate on that history. It assumes that a community will use diverse AI systems in a rapidly evolving environment. A working implementation of the publication infrastructure, agent prompt components, and documentation are provided to enable users to rapidly set up and run their own Symposium community.

VDM has been pursuing rigorous verification through mathematical theorem proving and software testing via simulated execution. Animation through an interpreter enables validation of the specification to ensure it meets the required functionality. Step-by-step execution in a debugger also allows the user to follow the internal behavior of operations. In this paper, we propose the recording and utilization of execution traces of assignments, operation calls, and return statements to make the internal behavior of operations persistent and analyzable as state-based models. The data model of events in execution traces, its implementation in ViennaTalk, and its application to visualization will be introduced.

Complying with the EU Cyber Resilience Act (CRA) is a resource-intensive challenge for SMEs due to the complexity of cybersecurity conformity assessment. Yet, it is essential for demonstrating regulatory compliance and ensuring product security and resilience. To address this, we introduce an automated framework for generating Assurance Cases (ACs) using an agentic Retrieval-Augmented Generation grounded in a formal Claim-Argument-Evidence logic. By systematically mapping technical documentation requirements, the framework streamlines the generation of certification evidence. We validate our approach on a case study of Catalink's PATROLIoT wildfire monitoring system, where the agentic RAG generated 70 ACs with high grounding density (~4.4 artefacts per AC). The proposed Natural Language Inference (NLI) evaluator achieves 0.88 accuracy, which provides robust evidence grounding and traceability, while expert-validated plausibility (3.06) supports interpretable justifications. For practitioners, this work provides a scalable, interpretable approach for automating mandatory CRA conformity assessments, reducing manual effort while maintaining transparent decision support.

We characterize the learning dynamics of a compact hybrid quantum forecasting model through comparison with a structurally aligned classical baseline. Using stationary harmonic-mixture and nonstationary chirp benchmarks with controlled spectral complexity and data availability, we analyze empirical Neural Tangent Kernel dynamics through kernel-target alignment, kernel drift, spectral concentration, and training loss. The classical model exhibits stronger early target alignment, whereas the hybrid model generally develops a less concentrated kernel spectrum and smaller kernel drift. Despite these distinct optimization geometries, both architectures attain similar held-out performance across the evaluated regimes. Notably, the hybrid model uses 125 trainable parameters compared with 281 for the classical baseline and reaches its validation-selected checkpoint earlier in 15 of 18 frequency conditions. A Fourier-augmented classical baseline does not reproduce the observed training behavior, while a controlled re-uploading ablation shows that repeated encoding systematically modifies both optimization and kernel geometry. These results demonstrate that comparable generalization can emerge from substantially different learning trajectories and that individual NTK diagnostics do not provide monotonic predictors of validation convergence. Rather than claiming a general quantum advantage, the study identifies architecture-dependent learning behavior that is masked by endpoint accuracy alone.

World models increasingly treat compact multimodal representations as interfaces between perception and physical interaction, yet existing probes do not establish whether different sensors carry the same executable meaning or whether that meaning survives a new action composition. We introduce an operational capability hierarchy and the Disjoint-Bridge Operator-Substitution Certificate (DBOSC), which asks whether independently trained modality compilers enter a frozen response chart interchangeably on evidence outside their training panels. On Cluster Haptic, audio and acceleration representations of the same unseen surface are 4.5x closer in response space than wrong-surface pairings, with the gap holding for all 19 held-out surfaces; unsealing withheld responses confirms that every branch predicts the physics better than the population chart. We then test ordered execution in a controlled elastoplastic system with complementary modality blind spots. At the pre-registered budget, the prerequisite refuses the stack because the frozen executor cannot advance even an exact chart coordinate through a held-out program. At a converged budget, the same rank-three chart executes those programs (oracle NMSE 0.18), fusion improves on both modalities, and 14 of 16 registered checks pass; the two failures arise because a diagonal restriction of the fused information matrix performs as well as the full one. Clearing the gate is a property of the executor, not the chart: an executor emitting whole programs instead of shared per-step dynamics is 38x worse than an entity-blind predictor on the same chart. A matching non-identifiability result explains why compression and fusion alone cannot determine an unseen composition law. These results separate attribute access, response substitution, fusion closure, and ordered execution into distinct, separately testable achievements.

Most modern optimizers form their momentum as an exponential moving average (EMA) of past gradients, forgetting every direction at one fixed rate. However, the inputs a deep network sees during training can be highly anisotropic, with a few directions queried frequently while most are seen rarely. Recent methods address this anisotropy by wrapping extra processing around this buffer, leaving the momentum update itself unchanged. We propose DeltaMomentum, which builds direction-awareness into the momentum update rule. The main observation is that the gradient of a linear layer splits into an input that acts as a key and an output-side error that acts as a value. Exploiting the key-value structure, DeltaMomentum updates the momentum buffer by the canonical delta rule, so each direction is forgotten at a rate set by how often it appears. We prove that it is a valid momentum, that it applies the input-side curvature correction without matrix inversion, and that it clears stale directions faster than EMA under both a fixed and a drifting optimum. It is a drop-in replacement for the momentum buffer of any optimizer, its coefficient transfers across widths under $μ$P, and its extra compute stays between $22.2\%$ and $25.0\%$ of a gated-MLP block's linear cost with no persistent memory. In FineWeb-Edu pretraining, AdamW with DeltaMomentum (DeltaAdamW) reaches AdamW's validation loss in up to $46.39 \pm 4.32\%$ fewer steps at 67M and $22.12 \pm 0.80\%$ at 370M over three seeds, and the gain persists at 1B on a Chinchilla-optimal budget. A Muon baseline tuned under the same protocol sits above DeltaAdamW at both language-model scales, and the gain holds for SGD, ResNet-18, and ViT-Tiny on CIFAR-10. Training-time diagnostics confirm the predicted mechanism, better gradient tracking and healthier input directions.

State-of-the-art vision-language-action (VLA) models such as $π_{0.5}$ exhibit strong semantic understanding, instruction following and task behavior. However, when deployed on new robots, even minor mismatches in hardware configuration relative to pretraining can cause severe performance drops. Finetuning the VLA on in-domain expert data from the new embodiment improves performance on the expert task but leads to a loss in its original instruction following and behavioral priors. In this paper, we propose a self-supervised method that generates online interaction rollouts from the zero-shot VLA as additional training data for finetuning. Our experiments show this finetuning scheme yields strong multi-task policies that, on the target robot, (1) inherit prior tasks distilled from the zero-shot model, (2) enable generalist instruction following, while (3) learning new skills from expert data with improved sample efficiency. We demonstrate the success of our approach across test sets probing generalization on a real ALOHA robot and a new simulation benchmark in RoboTwin. Video results are available at https://self-supervised-control.pages.dev/

Production machine learning systems degrade under concept drift, yet practitioners have little principled guidance on when to retrain. Retraining is costly, retraining budgets are finite, and a retrained model does not take effect instantly: training and deployment latency leave a stale model serving predictions while the data continues to move. We present a controlled empirical study of three practical model-refresh policies (periodic retraining, error-threshold triggering, and statistical drift-triggered retraining with ADWIN) against a no-retrain baseline, evaluated under a unified system model that makes retraining budgets and training-plus-deployment latency explicit. Across 3,933 experiment runs spanning three drift regimes, three budget levels, up to five latency levels, three datasets, and two learning modes, we find that the single most consequential design decision is not the retraining policy but whether the deployed model learns incrementally. With per-sample incremental updates, and for the linear online learner with immediate labels studied here, no policy differs from the no-retrain baseline by a practically significant margin in any of 54 paired comparisons, even at extreme latency. Without incremental updates, policy choice separates outcomes by 15-55 percentage points of post-drift accuracy, and simple periodic retraining significantly outperforms both reactive policies under abrupt and gradual drift, while reactive policies retain an advantage only under recurring drift. We document systematic failure modes of reactive policies and a latency-budget queueing interaction that silently halves effective retraining budgets, and release the full simulator, dataset pipelines, and per-run artifacts for reproducibility.

Python is widely used in scientific research because it enables rapid development and provides rich ecosystems for data analysis, artificial intelligence (AI), and machine learning. However, customized research code can become prohibitively slow as experiments scale. This challenge is particularly acute in discrete-event project-scheduling simulations, where sequential state updates, nested loops, conditional evaluations, and object-oriented structures limit the benefits of compiled numerical and GPU-accelerated libraries. Addressing these bottlenecks typically requires iterative profiling, refactoring, testing, and validation, yet researchers may lack the time or specialized software-engineering expertise for low-level optimization. This paper presents a systematic refactoring approach using Claude agentic AI on real-world project-scheduling workloads in a high-performance computing (HPC) environment. Guided by representative benchmarks and correctness checks, the agent identifies bottlenecks, implements targeted optimizations, and evaluates their effects, while the researcher retains final control. Testing runtime reduced from 1,298 seconds to under 200 seconds without changing outputs, saving four million core-hours (NZ\$320,000) annually.

This work introduces a formal semantic-block model for specifications and an execution-judged benchmark for evaluating specification quality independently of model capability. A specification is represented as a structure comprising semantic blocks, dependency relations, block-owned rules, decision points, and explicitly open questions, subject to four machine-checkable well-formedness conditions: acyclicity, single ownership, constraint domination, and totality or ambiguity-stop. Determinacy is defined model-theoretically as agreement among all conforming implementations and is estimated empirically through convergence across independent implementers. The model is instantiated on an Oracle-to-PostgreSQL migration specification containing 18 blocks and 19 dependency edges. Computational validation shows that the five-layer decomposition reduces mean per-task context by approximately 71% through dependency closures, covers 85.5% of the study-defined Oracle construct taxonomy with all identified gaps triaged, is not Pareto-dominated by the tested alternative partitions, and is recovered at the 99.9th percentile from citation-derived edges not used to define the original structure. The benchmark keeps the implementer panel fixed, includes a mandatory no-specification control arm, and uses PostgreSQL 16 and a live Oracle instance as deterministic execution judges. Six designed studies, including three pre-registered manipulations and three diagnostic analyses, further examine specification effects. Repeated runs on a 25-unit subsample reveal an empirical variability floor with a median arm-delta spread of 14.4 percentage points. The results support determinacy as a formal concept but not as a standalone empirical quality metric for the evaluated contemporary LLM implementers.

Lexical semantic change (LSC) is commonly modelled through vector-space representations, but these approaches often provide limited insight into which aspects of usage are changing. Diachronic corpus research instead examines interpretable dimensions such as syntactic behaviour, morphology, and constructional patterns, but typically through separate analytical workflows. We present SynFlow, an open-source toolkit for multidimensional diachronic analysis of linguistic usage. SynFlow converts linguistic observations into period-specific distributions and applies a shared workflow across dependency-based co-occurrences, morphological features, constructional configurations, and externally derived representations such as Frame Semantics. It supports different distance measures, together with value-level decomposition, statistical testing, and incremental clustering of lexical fillers. We demonstrate SynFlow through a qualitative case study of the German adjective viral, showing how a single semantic development is reflected across syntactic, lexical, constructional, and morphological dimensions. We further report previously published results on SemEval-2020 Task 1 to situate the performance of these representations relative to existing lexical semantic change detection systems.

Anomaly detection in tabular data is challenging because abnormal samples often arise as violations of cross-feature dependencies rather than simple marginal deviations. Existing detectors rely on geometric or reconstruction signals, while prior LLM-based approaches mainly fine-tune LLMs with normal samples or generate synthetic anomalies. We propose LLM-Detector, a framework that utilizes the in-context learning capacity of LLMs for structured, prompt-conditioned scoring synthesis, enabling LLMs to derive anomaly detection logic from structured normal-state knowledge. Specifically, normal training data are converted into statistical summaries, causal dependencies, and distilled prototypes that are organized into a prompt for code generation. The resulting scoring engine evaluates statistical deviation, structural inconsistency, and density-based abnormality then computes an anomaly score for each test sample. We evaluate LLM-Detector on 24 tabular datasets, comparing against 15 SOTA baselines. Results show consistent improvements across both mixed-type and continuous-only settings. Moreover, this design eliminates the need for LLM fine-tuning or neural network training, reducing computational cost and enabling practical anomaly detection in real-world tabular systems.

Real-time systems have various requirements such as the deadline and resource constraints. In addition, real-time systems are becoming larger and more complex, and studies on performance analysis and efficient scheduling algorithms are becoming increasingly important. Directed acyclic graph (DAG) models, which can express task dependencies and parallelism, are used for such studies. Random DAG sets are used to demonstrate the effectiveness and objectivity of methods proposed for real-time systems. However, there is no random DAG generation tool available that can generate a DAG set that considers the latest multi-rate applications. Therefore, researchers need to generate random DAG sets on their own, leading to additional effort and reduced reliability and reproducibility. To solve this problem, we propose a random DAG generator considering multi-rate applications for reproducible scheduling evaluation (RD-Gen). RD-Gen also enables batch generation of random DAG sets with different parameters. Case studies are used to demonstrate that RD-Gen can manage various problem settings and DAG study requirements.

Shocks that spread through the web, such as cybersecurity breach disclosures, can abruptly disrupt financial time series and cause substantial abnormal losses. While these events are disclosed as discrete records through news reports, regulatory filings, or public databases, their consequences unfold through continuous market dynamics. This creates an event-conditioned impact prediction problem: given pre-event market history and limited event metadata, the goal is to estimate short-term post-disclosure abnormal loss rather than reconstruct the full post-event trajectory. However, most time-series forecasting models focus on endogenous regularities such as trend, seasonality, and autocorrelation, and thus struggle with rare and heterogeneous external events. The challenge is further amplified by sparse high-impact events and background market noise. We introduce EventTime, a multi-resolution framework that combines long-horizon market context, short-horizon pre-event dynamics, and event metadata. It incorporates an event fusion module that couples temporal representations with event attributes to identify relevant recent market patterns. To mitigate sparse supervision, EventTime further introduces a dynamic contrastive objective that constructs event- and time-series-aware positive and negative pairs during training. We also construct SECURE, a real-world dataset aligning cybersecurity incidents with stock-market time series and structured and LLM-derived semantic features. Experiments show that EventTime consistently outperforms state-of-the-art time-series and event-aware baselines in estimating post-event financial losses. Further analyses demonstrate more event-sensitive representations, greater robustness to incomplete metadata, and more interpretable estimates of short-term market impact following cybersecurity disclosures.

Many benchmarks track Large Language Model (LLM) performance on tasks with verifiable answers, but less is known about how LLM performance is evolving on open-ended tasks, where creativity, originality and diversity may matter as much as quality. As LLMs increasingly support human ideation and creative work, understanding trends in LLM performance on open-ended tasks is critical. This paper presents a preliminary analysis of LLM creative outputs spanning three years of model releases, examining model responses to Infinity-Chat100, a real-world collection of open-ended user queries, and the Alternate Uses Task, an established psychometric creativity assessment. Using sentence-embedding similarity, we examine trends in LLM responses to these prompts. Our findings show a statistically significant decrease in model output diversity over time, suggesting that LLM outputs may be converging in creative substance across models. If this trend persists, LLM-driven homogenization may progressively diminish human agency in human-AI co-creative work, demanding careful consideration of LLMs' role in the human creative process.

Alzheimer's disease (AD) progresses as a continuous biological process, whereas most existing neuroimaging-based artificial intelligence methods remain limited to discrete diagnosis or clinical score prediction from cross-sectional imaging. In this work, we propose Disease Continuum Positioning (DCP), a longitudinal Bayesian Learning framework that continuously estimates disease severity from longitudinal diffusion tensor imaging (DTI). Specifically, DCP models disease severity as a low-dimensional probabilistic latent variable by jointly integrating longitudinal observations with weak clinical supervision, from which the proposed Disease Continuum Score (DCS) is derived to quantify an individual's position along the Alzheimer's disease continuum together with its associated uncertainty. Extensive experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort demonstrate that DCP consistently outperforms representative disease progression methods. More importantly, comprehensive validation analyses show that DCS accurately characterizes disease severity, exhibits strong clinical relevance, preserves longitudinal disease evolution, and predicts future disease conversion. These results suggest that DCS provides a quantitative imaging-derived representation for continuous assessment of Alzheimer's disease progression beyond conventional diagnostic labels and clinical scores.

Vulnerability prioritization is inherently preference dependent, since the same CVE can receive different remediation priority under different operational preference scenarios. Existing scoring systems and ranking methods typically assume a fixed criterion. In practice, organizations already operate under a preference scenario, but this preference is often implicit and difficult to express as a written prompt instruction, while triage queries usually do not encode it. Past validated triage cases under the current scenario are more readily available. We study query-based CVE prioritization in this setting and propose HARP, a graph-grounded multi-view framework that ranks candidates from a natural-language query together with a support bank of historical labeled examples from the current preference scenario, without requiring an explicit textual summary of that scenario. HARP retrieves evidence from a vulnerability knowledge graph, scores candidates with policy-conditioned global, enterprise, and user views, and fits view-fusion weights from sampled supports. Experiments across three preference scenarios and multiple backbone LLMs show that HARP outperforms multiple baselines, expressing our method's effectiveness.

Reliable performance evaluation is a central bottleneck for deploying robot-learning policies in real-world conditions. Real-world testing is faithful but costly and difficult to scale, whereas simulation-based testing scales easily but is inevitably biased by the sim-to-real gap. Existing simulation-augmented methods combine limited real-world rollouts with abundant simulation proxies, but focus on performance averaged over initial conditions and deployment settings. Such population-level averages obscure scenario-specific variation and provide limited guidance about when and where a policy can be safely deployed. We propose SCAPE, a scenario-conditioned simulation-augmented policy evaluation framework that predicts scenario-conditioned real-world policy performance using limited paired sim-and-real samples and large-scale simulation rollouts. SCAPE corrects sim-to-real bias in simulation labels before training the prediction model and calibrates prediction uncertainty through conformal prediction. We validate SCAPE on autonomous driving and quadruped velocity tracking. In sim-to-sim studies, SCAPE reduces scenario-level prediction error by 4.9%/34.7% (driving) and 14.5%/27.7% (quadruped) relative to scene-conditioned neural and aggregate statistical baselines on average. We further evaluate a velocity-tracking policy deployed on a physical Unitree Go2. SCAPE also improves testing sample efficiency, produces narrower calibrated prediction intervals, generalizes better to out-of-distribution scenarios, and enables fine-grained deployment strategies.

Microlensing can reveal populations of faint compact objects that are otherwise difficult to detect. Depending on their design, all-sky surveys have the potential to search for these objects across the sky. The Transiting Exoplanet Survey Satellite (TESS), primarily designed to detect transiting exoplanets, also provides near all-sky coverage with high cadence. In this work, we use TESS data to search for microlensing candidates using both traditional and machine-learning methods and to identify associated false positives in high-cadence surveys. Microlensify is a physics-informed, transformer-based variational autoencoder trained on simulated single-lens microlensing light curves and real TESS Sector 12 data. The model classifies events, reconstructs light curves, and estimates microlensing event durations. Applied to $\sim 5.6$ million TESS light curves, it identified between $0.036\%$ and $1.89\%$ as microlensing candidates across different TESS pipelines. After applying microlensing detection metrics and cross-matching with SIMBAD, we obtained a final list of candidates and identified false positives including long-period variables, Mira variables, cataclysmic variables, red giants, and transients. We also found Gaussian-like peaks caused by asteroid crossings, a potential source of false positives in high-cadence microlensing surveys. The model also predicts event duration with an accuracy of $R^2 = 0.97$. The model was further tested on published events from different ground-based microlensing surveys, confirming 92.7% as microlensing, demonstrating its applicability across surveys with different cadences.

On-policy distillation (OPD) has emerged as an effective framework for post-training language models by pairing student-generated trajectories with dense token-level supervision from a teacher. However, OPD implicitly assumes that teacher-derived rewards are an appropriate proxy for reasoning progress, and therefore treats all teacher feedback equally during policy optimization. While in practice, this assumption does not always hold. We observe that teacher-derived rewards often conflict with genuine reasoning progress, as reasoning steps with clear reasoning advancement may still receive lower distillation rewards, simply due to deviation from teacher's outputs. To address this mismatch, we propose Reasoning-Progress-Aware Reward Filtering for On-Policy Distillation (R2-OPD), which constructs two within-trajectory rankings of reasoning spans, one from teacher-derived rewards and the other from independently estimated progress reward. Distillation rewards are selectively suppressed whenever the two rankings disagree, reducing supervision that conflicts with reasoning progress while preserving effective teacher guidance. Our approach shows consistent improvement over standard OPD especially regarding reasoning performances.

Deep Learning models can include billions of parameters or more, making it difficult to explain their internal transformations and outputs. However, explainability is increasing in importance due to the use of AI in crucial applications. This paper focuses on the interpretability of convolutional neural networks (CNNs). Building on the popular gradient based method LayerCAM for extracting internal features in CNNs, we propose an improved method named HiRA-CAM, and show that it outperforms both LayerCAM and Grad-CAM on creating useful saliency maps for object classification. The main feature of HiRA-CAM is its adaptive use of activation maps from all the layers of the CNN to arrive at a more focused saliency map.

Hybrid Transformer-Mamba large language models (LLMs) enhance long-context efficiency, but their heterogeneous computation and communication patterns complicate efficient hardware acceleration. Chiplet-based architectures offer a scalable solution by integrating specialized compute and memory units. However, the design space spanning static architectural configurations and dynamic runtime policies is prohibitively large to explore exhaustively. To address this challenge, we present HYDRA, a comprehensive design space exploration framework for hybrid LLM serving on heterogeneous chiplet systems. HYDRA jointly explores chiplet composition, placement, inter-chiplet bandwidth provisioning, dynamic batching, and runtime scheduling. It integrates communication-aware placement, dynamic batching, elastic task scheduling, and a fast Markov-based performance estimator that captures multi-tenant runtime dynamics for efficient and accurate exploration. Across all workloads, HYDRA delivers 1.55x the throughput and 43.7 percent lower time-to-first-token on average, with throughput gains reaching up to 2.3x compared to state-of-the-art baselines. These results highlight that co-designing architecture and runtime policies is critical for efficient large-scale LLM serving on heterogeneous chiplet systems.

We introduce Deep-MKV-TS, a path-dependent McKean-Vlasov framework for financial scenario generation. The stochastic dynamics are chosen by matching selected path and volatility features of generated scenarios to those observed in the data. Starting from an interpretable reference model, Deep-MKV-TS preserves the reference drift and adjusts its volatility, while a regularization penalty limits unnecessary departures from the calibrated dynamics. We solve the resulting control problem using a neural, sample-based implementation of the stochastic maximum principle. We validate the method against an exactly computable oracle. On Heston and Heston-mixture models, Deep-MKV-TS substantially reduces path-dependent and volatility-related deficiencies of the reference model. In delayed-volatility experiments, the correction remains effective as the forecasting horizon increases, while direct training becomes less reliable. On held-out intraday equity-index futures, the corrected model improves conditional forecasts relative to the reference and reaches a level of performance comparable to flexible generative and historical baselines. The resulting scenarios also support greater exposure than the reference under a fixed drawdown-risk target. These results show that path-dependent McKean-Vlasov control can enrich an interpretable reference model without replacing it.

Automated market makers (AMMs) are a cornerstone of decentralised finance (DeFi). Constant product markets with concentrated liquidity, such as UniswapV3, are now a well-established design. In these markets, liquidity providers (LPs) face a sequential decision problem: they must decide when to rebalance their positions and which price ranges to allocate capital to as market conditions evolve. We formulate dynamic liquidity provision as a stochastic impulse control problem and use reinforcement learning (RL) to solve it, focusing on providing interpretable solutions. We show that learned policies exhibit rich state-dependent behaviour, allocating liquidity according to mispricing, rebalancing costs, uncertainty, inventory exposure, and heterogeneous risk preferences. These behaviours help compress the left tail of the Profit and Loss (PnL) distribution and avoid catastrophic outcomes under high uncertainty. Finally, we benchmark the RL agents against baseline and sophisticated agents from the AMM microstructure literature and analyse their performance.

Network embedding methods learn low-dimensional representations of graph-structured data to support downstream tasks such as node classification, link prediction, and influence maximization. However, real-world networks often reflect structural inequalities arising from demographic imbalances, homophily, and other societal biases, which fairness-agnostic embedding methods can encode and amplify. To address this issue, numerous fairness-aware network embedding methods have been proposed to mitigate bias while preserving embedding utility. This survey presents a comprehensive overview of fairness-aware network embeddings for complex networks. We propose a taxonomy that categorizes existing methods along three main complementary dimensions: underlying embedding approach (spectral, random walk, graph neural network, Bayesian, and method-agnostic), fairness intervention strategy (pre-processing, in-processing, and post-processing), and fairness objective criterion (embedding- or task-level). We further compare methods with respect to group versus individual fairness and assumptions regarding sensitive attributes. Finally, we discuss current limitations and highlight promising future research directions. This survey provides a unified perspective on fairness-aware network embedding and serves as a reference for developing fair and trustworthy network representation learning methods.

Bayesian inference via Markov Chain Monte Carlo (MCMC) provides effective parameter estimation, but its real-time application in complex physical systems is hindered by heavy computational bottlenecks and extreme sensitivity to statistical noise. We address this by proposing a neural-network-based probabilistic surrogate framework for rapid and robust MCMC inference. Using fusion plasma Thomson scattering diagnostics as a challenging, noise-dominated testbed, our approach employs a dual-head architecture to simultaneously estimate the expected physical emission spectrum and the channel-wise intrinsic measurement noise variance. By optimizing a Gaussian Negative Log-Likelihood (GNLL) objective, the learned aleatoric uncertainty dynamically buffers the sampler against pathological shot noise. Evaluations demonstrate that this surrogate framework achieves > 1500x acceleration over exact physical forward models, while simultaneously reducing inference error (RMSE) by >20% compared to standard homoscedastic neural baselines, offering a highly promising paradigm for real-time physical analysis.

Split-conformal prediction provides marginal coverage under exchangeability and is increasingly used as an abstention layer for zero-shot vision-language models (VLMs). We audit this practice under deployment shift for CLIP, OpenCLIP, and SigLIP across ImageNet and non-ImageNet settings. Marginal coverage can remain relatively high while class-conditional tail coverage collapses: on ImageNet-Sketch, worst-class coverage falls to $\approx 0$ and 10-12% of classes lie below a finite-sample null floor, despite marginal coverage of about 0.86. The failure is aligned with target-domain class accuracy but is not predicted by the source-domain diagnostics we test. Source-side Mondrian calibration improves the in-distribution tail but does not transfer, while clustered conformal and Conf-OT improve marginal or average metrics without recovering the worst-class tail. Target-side class calibration substantially lifts the tail, but requires labels for every class and remains set-size-intensive. We further identify a 2-3$\times$ cross-family efficiency gap and show that native SigLIP sigmoid scores remove APS's probability-mass interpretation. The findings persist across the tested model scale, pretraining corpus, prompt, miscoverage level $α$, and shifted non-ImageNet settings. Marginal conformal coverage should therefore be treated as an average reliability statistic, not as a safety guarantee for the class tail.

Statistical watermarks for language models live in the freedom of the signifier: they choose among tokens that are nearly equivalent in meaning, and they are therefore eroded by exactly those transformations which move the form of a text while leaving its content in place. The literature measures such transformations by their endpoint, through the semantic similarity between the original and the rewritten text. We show that the endpoint is the wrong statistic. Adapting the formalism of linguistic loops, we prove that the invariant of a chain of meaning-preserving transformations factorises canonically into an endpoint part and a holonomy in the stabiliser of the initial state, the second of which the semantic deficit cannot see; the loop rotation is parallel transport on the unit sphere of the embedding space, so that the analogy with the Wilson loop becomes a theorem rather than a figure of speech. On the side of the detector we prove an exact identity: the residual statistic is proportional to the number of positions whose seeding window survived intact, from which the decay law $ρ^{h+1}$ follows as the independent-edit corollary. The identity has a disconcerting consequence, which we confirm to three decimal places: at one and the same retention rate the surviving signal may be one half of the original, one quarter of it, or exactly nothing, according only to where the edits fall.

Mission-critical intelligent systems often operate under time-varying limitations that reduce control authority and change the admissible safe operating envelope. In such settings, a safety certificate learned under nominal conditions may become invalid as system capability changes. To address this challenge, this paper proposes a degradation-aware, data-driven safety-filtering framework that learns a safe set from data, updates it online, and enforces the resulting learned barrier through a time-varying control barrier function (CBF). A nominal safe envelope is first learned from operational data using a radial basis function (RBF)-kernel support vector machine (SVM), whose decision function serves as the initial CBF candidate. To capture capability-induced safe-set contraction, a continuous-time decremental SVM update law is developed so that selected support-vector coefficients are reduced according to a degradation signal. A homotopy-smoothed SVM-CBF is then introduced to avoid discontinuous changes in the learned barrier during active-set transitions. The resulting time-varying learned barrier is enforced using a quadratic-program-based safety filter under degraded input constraints. Forward invariance of the learned time-varying safe set and recursive feasibility of the safety filter are established. Simulation results on a vertical takeoff and landing (VTOL) model show that the proposed method maintains safety under reduced control authority and avoids abrupt barrier-switching effects during safe-set contraction.

This study reports the development of an Automatic Speech Recognition (ASR) system in Mizo, a low-resource language. The development included collecting 17.62 hours of speech data, curating it, and fine-tuning the Mizo ASR system with three Whisper multilingual models and with the SraVaani 1.0 Indic multilingual model. Whisper-large-v3 achieved the lowest conventional WER (18.08%), while morphology-aware evaluation yielded a WER of 7.22%. Zero-shot evaluation of the SraVaani 1.0 Indic multilingual model yielded a WER of 58.27%, while Mizo-specific fine-tuning reduced the conventional WER to 29.45% and the morphology-aware WER to 17.93%. The results demonstrate that the Whisper model can achieve a substantially low WER, even when adapted to an unseen language. In contrast, SraVaani 1.0 supports the Mizo language in its multilingual model; however, fine-tuning with carefully curated Mizo speech data substantially improves its performance.

Robustness verification of neural networks is increasingly important, due to their use in many critical domains. In certain scenarios, proof sharing has been shown to accelerate incomplete verification techniques by reusing intermediate-layer abstract states, or templates, across queries. However, questions remain as to the robustness of template-based acceleration across varying network architectures, properties, datasets, and training methods. In this work, we perform a systematic study of the effectiveness of template-based acceleration and its limits. Our study shows that template subsumption rates can vary widely across scenarios. We present a novel metric of jointly stable neurons to explain this variation, showing that in some cases template-based techniques are very unlikely to provide any speedup. Then, we present FastCert, a novel technique for automatically distributing templates across neural network layers to increase performance impact, eschewing templates entirely if they are unlikely to produce a speedup. Across a large set of covering-design based $L_0$-verification tasks, FastCert achieved an average speedup of 1.13x over an extant template-based reuse technique.

Search-based Automated Program Repair (APR) techniques rely on carefully designed mutation operators to explore the space of candidate fixes. Recent advances in Large Language Models (LLMs) suggest that generative models could replace such operators by dynamically proposing repairs. In this paper, we investigate this hypothesis in the context of Cyber-Physical Systems (CPSs) modeled in Simulink/Stateflow. We extend the state-of-the-art FlowRepair approach by replacing a subset of its mutation operators with LLM-generated mutations, enabling more flexible and expressive patch generation. We evaluate the approach on a benchmark of 19 real-world faulty Stateflow models across four CPS domains, using the same experimental setup as FlowRepair for controlled comparison under the same wall-clock budget. Contrary to expectations, in this controlled evaluation, the LLM-based mutation substantially degrades repair performance under the FlowRepair experimental setup. Across the tested LLM variants, the LLM-based repair produced plausible patches for 4-6 models and valid patches for 4 models, compared to 18 and 16, respectively, with the original approach. Our analysis reveals that, in this integration, LLMs struggle with precise symbolic edits, lack behavioral feedback, and generate a noisy search space that hinders effective exploration. Rather than showing a general limitation of LLMs for APR, these findings highlight fundamental limitations of naively integrating LLMs into search-based APR and motivate hybrid approaches that combine structured mutation with generative guidance.

Mechanistic interpretability seeks quantities that models do not expose directly: represented states, component effects, interactions, and responses to interventions. Patching, gradients, Hessian-vector products, and subset interventions provide different measurements under different access assumptions and may target different quantities. We formulate their shared measurement structure as mechanistic tomography: designed measurement for recovering internal mechanisms and intervention effects. For a chosen basis and intervention family, measurements take the form y = Ax + w, where A describes the interventions, x is the target map, and w contains nonlinear response, sampling error, and basis misspecification. This language gives a practical procedure: start with the least costly measurements, test on held-out interventions at the intended scale, calibrate simple mismatch, and expand the measurement family when structured residuals remain. Control provides a demanding validation setting because an estimate that guides an intervention acts as an observer. In a two-HMM model, control error rises with observer error, while target improvement can hide nuisance-state movement. Under forward-only access, sparse aggregate measurements recover a finite-effect map with fewer interventions than coordinate patching. With gradient access, finite probes improve a local attribution map. Lifted measurements and Hessian-vector products recover interactions missed by first-order maps, while Tracr shows that the required family depends on the basis. On GPT-2-small IOI, the Name Mover-Negative Name Mover interaction is the largest held-out predictive term among three tested cross-group pairs. On Qwen-2.5-7B, finite calibration makes an additive refusal-response map adequate, so held-out error does not support pairwise lifting.

We modify the NN/QFT duality [1] to incorporate the layerwise permutation symmetry of the network, resulting in a $(0\!+\!1)$-dimensional lattice gauge theory, in which each layer of $N$ neurons acts as an $N$-component lattice site, and the weight matrices play the role of gauge fields living on the links. In this framework, we compute the tree-level neuron-neuron propagator which describes the evolution of layer variance in the network, and develop the Feynman diagram machinery to compute interactions in the perturbative expansion in $1/N$. In particular, we obtain a recursive expression for all corrections to the exact propagator at $O(1)$, representing statistical fluctuations in the ensemble of networks, including infinitely-many loop diagrams mediating the interactions from previous layers. We also present a preliminary analysis of neuron scattering amplitudes that contribute order-by-order in $1/N$, which provides a field-theoretic framework for studying higher-point correlations, and by extension information propagation, in deep networks. We remark on some interesting directions for future work at the intersection of neural networks and quantum field theory.

Uncertainty quantification (UQ) is essential for the safe deployment of large language models (LLMs). Existing methods, from verbalized confidence to ones requiring multiple generations, are often zero-shot and produce scores quantifying uncertainty without the need for labelled data. Nonetheless, in practice one must always evaluate their performance on a dataset of interest before deployment. In this work we show that, by leveraging this dataset, we consistently outperform these existing scores. Specifically, we build simple classifiers that predict LLM response correctness by using these scores and the correctness of similar queries as features. Our method produces minimal computational overhead, making it a cheap and straightforward enhancement for UQ in LLMs for real-world applications.

Given a set of input states, we consider the task of predicting the expectation value of a Pauli observable at the output of an unknown quantum evolution, using only a limited number of measurements. Recently, quantum Gaussian process (QGP) regression was introduced for this task across various classes of unitary evolution. Here, we extend the QGP framework beyond unitary dynamics. In particular, we prove convergence of the channel's outputs to a QGP and derive the associated closed-form kernel under a uniform (Lebesgue measure) prior over quantum channels. The kernel's dimensional factor, however, dictates the required observation precision. While manageable when the channel and observable are restricted to small subsystems, exponential suppression precludes learning when the subsystem grows extensively with the system size. Since the Lebesgue prior is overly broad for many applications, we propose an empirical Bayes heuristic that replaces the dimensional factor with a learnable scale parameter while retaining the kernel's state-overlap correlation structure. In numerical simulations of up to 64 qubits, channel QGP regression with the Lebesgue kernel exhibits a strong inductive bias for local channels, enabling faithful extrapolation. For global 64-qubit channels, the rescaled kernel restores learnability, with predictions improving systematically with the shot budget. Results from a noisy quantum computer further demonstrate the robustness of QGP regression under experimental conditions. Beyond regression, we validate QGPs as Bayesian-optimization surrogates for state preparation under noisy XXZ dynamics.

Lung cancer screening with low-dose chest computed tomography reduces mortality, but its impact is limited by uptake, adherence, and management challenges. Blood-based cell-free DNA (cfDNA) biomarkers offer a complementary approach, although early detection remains difficult because of lung cancer heterogeneity and high-dimensional, nonlinear molecular signals. We evaluated quantum-classical hybrid machine learning for lung cancer detection using DNA fragmentomics and DNA methylation. After feature selection, models were trained using 20- and 40-feature subsets. Features were encoded into quantum Hilbert space using angle and dense-angle feature maps with multiple entanglement strategies. Fidelity-based quantum kernels were computed with exact statevector simulation and integrated with precomputed-kernel SVM and kernel-PCA logistic regression and compared with an SVM model trained on the original features. This framework enabled systematic evaluation of how encoding and entanglement design affect classification. Across repeated held-out evaluations, quantum-kernel models achieved competitive performance on both datasets. For fragmentomics, several 20-feature configurations improved AUC relative to a classical SVM baseline, suggesting effective capture of nonlinear cfDNA fragmentation structure. For methylation, the classical SVM achieved the highest AUC, although selected quantum models remained competitive and improved specificity in some cases. Increasing features from 20 to 40 did not consistently improve performance and often increased variability. Overall, these results support quantum kernel methods as a promising approach for cfDNA-based lung cancer detection.

When a tool call times out, the agent sees the failure and can route around it. A cached error page or negative price can instead arrive in the expected format and be consumed as fact. We introduce Outcome Monitors, which detect violations of outcome contracts mined from task-disjoint traces or derived from public schemas. On a violation, the monitor preserves the result and issues a nonbinding receipt naming the violated property and public recovery tools. In frozen, prespecified evaluations with injected failures, Outcome Monitors raise ToolMaze completion from 10.9% to 28.1% across four models in two provider families and replicate in a third. In tau-bench retail, completion improves by 14.0 and 12.0 points on two tiers. In separate ToolMaze controls, removing the recovery-tool list eliminates the measured gain and restoring it recovers the effect; diagnostic detail and timing produce no detectable differences. Gains concentrate where the fault blocks completion. On a suite transcribed from a published incident taxonomy, detection outside the mined vocabulary falls to 46%, though delivery continues and completion is unchanged. Recovery tools are the active receipt content in these controls; extending detection beyond the contract vocabulary remains open.

This article explores the performance of analytical and neural-based hillshading methods in a dense urban environment using high-resolution digital elevation model (DEM) and digital surface model (DSM) data for downtown Calgary. The study compares single-direction and multi-direction analytical hillshading with relief shading generated in Eduard, a machine-learning system originally developed to emulate Swiss-style shaded relief trained primarily on mountainous landscapes. Because Eduard was not designed for buildings, bridges, streets, trees, and other urban infrastructures, the central question is not whether it perfectly reproduces urban morphology, but whether parameter tuning can nevertheless produce visually strong, cartographically useful, and in some cases superior results when compared with conventional analytical methods. The analysis focuses especially on terrain type, micro and macro generalization, and flat-area detail parameters, while keeping the large-scale shading style constant throughout the neural experiments. The article is structured as an exploratory comparison rather than a benchmark of universal best practice. It aims to identify where analytical hillshading remains more reliable, where Eduard offers unexpected strengths, and where neural shading fails because of its training bias toward alpine terrain. The study contributes to current work on terrain representation by testing whether a neural approach designed for natural landforms can be adapted to a highly built urban setting, and it concludes by arguing for future model training and evaluation specifically targeted at urban relief shading.

Air traffic control (ATC) communication is a safety-critical dialogue that remains largely human-driven even as other parts of air traffic management have been semi-automated. In this article, we experimentally evaluate whether large language models (LLMs) can generate operationally realistic ATC transmissions. An experimental general-aviation flight flying over the San Francisco "Bay Tour" route is hand-transcribed and used as ground truth (P0). Through a pilot-in-the-loop process we design five prompt structures (P1-P5) of increasing constraint and embed them in a stateful multi-turn pipeline, where the model plays ATC to a fixed pilot transcript while conditioning on the accumulating dialogue history. Across nine open- and closed-source LLMs we vary the prompt, the presence of a worked transcript from a different experimental flight as an in-context example, and whether the model conditions on its own prior replies or on injected ground-truth history. Turns are scored with lexical, structural, and semantic similarity metrics and by an LLM-as-judge (GPT-5.5) validated against human expert annotation. Supplying a worked example improves similarity, but tightening the prompt does not: the lightest prompts perform best and the most heavily scripted one collapses as its own errors accumulate through the dialogue, which injecting correct history repairs. These results outline a concrete path and its current limits toward LLM-assisted ATC.

Anticipating how scenes evolve under ego actions is fundamental to safe autonomous driving, yet the full potential of world models for decision-making remains unrealized. The critical challenge lies in ensuring that future modeling is not merely predictive, but decision-informative: the predicted future must directly shape which trajectory is selected. Existing approaches decouple future representation learning from planning optimization, or share predicted states across trajectory candidates, thereby diluting the action-specific consequences that ought to guide selection. To bridge this gap, we propose DA-WAM, a framework that unifies predictive representation learning, action-conditioned future modeling, and trajectory scoring under a single decision-making objective. DA-WAM maintains predictive supervision throughout planner optimization via an online encoder and a stable momentum target, allowing future representations to co-evolve with the driving task. An action-conditioned predictor generates a distinct future latent state per trajectory candidate, which is then evaluated by a future-latent-conditioned factorized scorer. For the expert-matched trajectory, the predicted future latent is supervised by the observed future representation, while safety-critical hard negatives provide additional supervision near planning boundaries. Extensive experiments on NAVSIM-v1 and NAVSIM-v2 demonstrate state-of-the-art performance, while ablations and diagnostic analyses validate the key components.

While multimodal large language models (MLLMs) excel in medical applications, most of them favor static images or short-term signals. In the critical field of dynamic electrocardiograms (ECG), models struggle with complex temporal reasoning and diagnostic report generation due to a lack of high-quality datasets and benchmarks. To address this, we introduce (i) Holtercare-23K, a large-scale multimodal dynamic ECG dataset comprising 22,980 QA pairs derived from 788 clinical Holter records and featuring a novel signal-video-text tri-modal alignment. Based on this dataset, we present (ii) Holtercare-Bench, a multimodal benchmark that evaluates models on temporal localization, clinical diagnosis, and global summarization. Zero-shot evaluations of leading MLLMs reveal a significant performance gap in processing ultra-long pathological sequences. However, fine-tuning representative models yields substantial improvements. This work illuminates the limitations of current MLLMs in electrophysiology and provides a foundational benchmark for long-term medical MLLMs. Our project is available at https://github.com/ZJU4HealthCare/Holtercare-Bench.

Large language models (LLMs) are increasingly paired with verifiers (step checkers, self-consistency filters, tool-based fact checkers, formal proof assistants) that claim to detect the model's errors. Yet the verification literature uses the word "level" to mean at least five different things: verification granularity, concept abstraction, risk tier, system-stack layer, and the epistemic source of the ground truth. We propose Verification Autonomy Levels (VAL), a meta-standard that classifies any verification scheme along a single axis: where does the verification spec come from, and what does the verdict guarantee? VAL ranges from L0 (LLM self-declaration; no deterministic anchor) through L2 (objective ground truth; correctness only) to L3/L4 (decidable systems with single-property or domain-level completeness), with L5 impossible in the unrestricted case. Central to VAL is the completeness blind spot: substitution- and sampling-based verifiers can confirm that proposed candidates hold, but cannot prove that no candidate was missed. We further identify a dichotomy the literature has not stated: completeness is reachable only for formally specifiable properties, whereas empirical open-world verification (fact-checking, diagnosis) caps at anchored correctness (L2). We document this gap empirically across four domains (symbolic mathematics, behavior monitoring, medical diagnosis, and code generation, the last a reverse validation with predictions stated before evidence) and in the strongest formal-verification baseline in our survey, whose authors note the verifier focuses on the correctness of each step. We show the levels of granularity, concept hierarchy, risk, and system stack are orthogonal to VAL, resolving a systematic conflation across 17 surveyed papers. Code and full assessment are released as supplementary material.

Designing, implementing, and comparing interpretable architectures requires a formal language to represent them. The most common representations fall short in one of two ways. Symbolic equations give no global view of an architecture at a glance. Probabilistic graphical models and flowcharts do not describe actual tensor manipulations, thus hiding key insights and limiting reproducibility. To close this gap, we introduce a graphical notation for designing interpretable AI architectures, adapted from Penrose tensor notation. This graphical notation gives a global view of an architecture and maps one to one onto PyTorch einsum code. We first use this notation to describe architectures that are interpretable by construction, including concept bottlenecks, sparse probes, prototype networks, neural additive models, and mixtures of linear models. We then diagram the key architectural components of Steerling-8B, a frontier interpretable language model. The diagram yields global insights into the architecture (e.g., showing that Steerling is a residual model), a geometric interpretation of each individual operation, and a direct translation into 33 lines of PyTorch code.

As AI systems are increasingly deployed in safety-critical application domains (e.g., autonomous driving), associated risks increase too. Deep learning models underlying modern AI systems, therefore, must undergo thorough testing to ensure their correct behaviour. A single robustness test involves thousands of inferences to empirically verify if a model's outputs remain stable under a bounded perturbation of its inputs. However, existing testing frameworks lack the means to systematically explore and summarise robustness across a combinatorial space of perturbations. We propose TestifAI, a deep learning testing framework for efficient and accurate estimation of robustness against combinations of perturbations. TestifAI enables users to specify operational conditions as structured spaces of semantic input perturbations (e.g., image blur, brightness and zoom) and discrete severity levels (e.g., low, medium and high). Users can query model robustness for any combination (e.g., "low blur, high brightness, and medium zoom"). To achieve efficiency and accuracy, TestifAI introduces partial model tomography, a novel approach to reconstructing model behaviour in a multi-perturbation space from tests that apply only a small number of perturbations (lower-order projections). To estimate robustness against at least three perturbations, TestifAI trains an auxiliary model on the results of tests involving up to two perturbations only, avoiding execution of an exponential number of tests. Our experiments on five image and language classification tasks show that TestifAI can predict higher-order (3 and 4 perturbations) test outcomes from low-order (1 and 2 perturbations) observations with an aggregate robustness estimation error of less than 7%, while reducing the number of inferences by 60-80%.

Recent Visual-Language Models (VLMs) have enhanced the capabilities of pre-trained LLMs by adding vision tokens alongside text, with approaches like LLaVA showing impressive results. However, the computational burden of processing up to 576 or 729 visual tokens makes edge deployment challenging. While various token pruning techniques require retraining, some are training-free and thus can easily adapt to architecture changes. We introduce ClustRS, a two-part, training-free algorithm for robust token pruning. Its first component is an attention-weighted, clustering algorithm that selects representative tokens from each semantic cluster. The second component, Residual Shrinkage, is a one-pass denoising step on the selected tokens. These training-free lightweight steps make LLaVA ready for real-world data, improving robustness to a wide range of image-noise types and intensities. Experimental results on the ScienceQA-IMG and MM-VET benchmarks show our method outperforms attention- and diversity-based methods by up to 20\% under extreme noise and token conditions (reducing tokens by 97\%, down to 16 tokens) on LLaVA 1.5 7b and achieves exceptional results on LLaVA-OneVision, where we match baseline performance with fewer than one-third of their tokens under mild noise conditions. Our study demonstrates a simple yet powerful alternative to both score-only and diversity-only pruning rules, paving the way for compute-efficient and noise-resilient VLM deployment.

Statutory retrieval is necessary for citation-grounded legal question answering, but remains underexplored for Greek. We introduce GreekBarRetrieval, a public retrieval benchmark derived from, and complementing GreekBarBench, which did not include retrieval. The new benchmark comprises 283 bar-exam questions, each accompanied by the facts of the case it refers to, and 6,308 candidate statutory articles to retrieve from. Questions and facts are stated in everyday language, but need to be mapped to the formal terminology of statutes and their abstract legal concepts. A further complication is that not all of the case facts are relevant to each question of a case. Experimenting with three BM25 variants and nine dense retrievers, we find that vanilla dense retrieval far outperforms vanilla sparse retrieval in Recall@100. However, LLM-based query reformulation helps BM25 close that gap, while also improving dense retrieval. With a ten-round ReAct-like LLM reformulation loop that we introduce, BM25 improves further in Recall@100 and obtains the best nDCG and MAP scores of all tested retrievers. Query reformulation also outperforms pseudo-relevance feedback, sparse-dense fusion, and English translation.

When factor scores replace true latent scores in nonlinear prediction, measurement error attenuates the recoverable variance of any $k$th-order component of the regression function by $ρ^k$ -- the $k$th power of the score's coefficient of determination -- for any linear score type. This study derives the bound via Hermite polynomial expansion and proposes PV-ANN -- plausible values (posterior draws preserving latent variance) combined with artificial neural networks (learning functional form without prespecification). The bound governs recovery of the latent-scale function, not prediction of the outcome from observed indicators, for which factor scores are already sufficient; the two metrics are therefore predicted to dissociate. An 18-condition simulation supports both predictions: in the nonlinear low-reliability conditions PV-ANN closes about four fifths of the function-shape recovery gap between a factor-score learner and one given the true latent values, and the margin widens as reliability falls, while predictive accuracy is not improved, as the theory requires. A Big Five application illustrates the intended exploratory workflow and delineates boundary conditions under weak signal and measurement model misspecification.

Proactive interference (PI) is a documented failure mode in large language models in which retrieval of a repeatedly overwritten value degrades as prior overwrites accumulate, mirroring a classical phenomenon in human working memory. Post-training quantization (PTQ) is now the default deployment path for open-weight models, yet its effect on this failure mode has not been tested. We evaluate three precision levels (FP16, INT8, INT4/NF4, via bitsandbytes) across three architecturally distinct instruction-tuned models (Qwen2.5-7B-Instruct, Mistral-7B-Instruct-v0.3, Phi-3.5-mini-instruct), holding the retrieval task fixed. INT4 quantization significantly reduces accuracy under high interference in every model (e.g., from 81.0% to 68.3% for Qwen), confirmed by paired McNemar's tests ($p \le 2.6 \times 10^{-6}$) and a mixed-effects regression spanning all interference levels; INT8, often assumed safe, also carries a smaller but real penalty in two of three models. The effect is specific to semantically similar (word-type) distractors and reverses sign under a numeric control condition, and is mechanistically linked to a rise in same-key intrusion errors under INT4 (from 21.5% to 24.6% of trials, $p = 4.8 \times 10^{-7}$). A follow-up ablation shows the effect originates in the quantized transformer backbone rather than the output projection layer. These results suggest that bitsandbytes 4-bit quantization can impose an additional cost on applications relying on long, updatable, semantically dense contexts, even when aggregate benchmark accuracy appears largely unaffected. We release our code and tokenizer-verified vocabulary construction method at https://github.com/ShayanShahrabi/compress-and-forget

Reinforcement learning with verifiable rewards (RLVR) commonly post-trains reasoning models on multiple tasks, while rerunning multitask RLVR (MTRL) as new tasks are added makes capability expansion costly. We therefore study continual RLVR, which updates the existing model as each task arrives. The central question is whether a model updated this way can perform as well as a jointly trained model. To answer this question, we introduce Continual Reasoning Gym, a continual-RLVR environment that organizes text and visual reasoning tasks into five task sequences. In this setting, we identify two key observations: Sequential RLVR exhibits modest forgetting, yet its final performance remains below that of MTRL. To understand the latter, we decompose final performance and show that forgetting accounts for only part of the gap. To explain the former, we identify shared reasoning: transferable reasoning structure allows training on one task to support others on average. We therefore introduce Continual Prompt Replay (CPR), which harnesses shared reasoning to improve learning on the arriving and future tasks by replaying previous-task prompts and regenerating their responses with the current policy. On average, only CPR reaches MTRL-level performance.

Tactile Internet services couple cyber events directly to physical actuation, so security decisions must improve risk discrimination without perturbing the control path. This paper presents VQC-ZTI, a split-plane Variational Quantum Classifier framework for zero-trust protection of Tactile Internet services, in which an off-path VQC analyzes encrypted-flow telemetry while an on-path policy engine applies cached deterministic grant, restrict, step-up, and deny actions. By decoupling anomaly scoring from enforcement, VQC-ZTI preserves predictable control behavior and allows detector sensitivity and policy aggressiveness to be tuned independently. We evaluate the framework on CESNET-derived aggregated traffic using random, entity-group, and temporal holdouts with a hybrid PyTorch-PennyLane implementation. The full-hybrid Quantum Neural Network achieves mean areas under the receiver operating characteristic curve of 0.9981, 0.9974, and 0.9941 and reduces the false-positive rate relative to ExtraTrees by 44.6%, 49.6%, and 67.9%, respectively. A representative component-timing decomposition further illustrates that batched VQC scoring remains in the asynchronous evidence path rather than the immediate enforcement path.

Dense-caption retrieval has recently been improved by introducing segmentation, edge maps, LLM-filtered captions, and cross-modal modules into contrastive fine-tuning. However, these methods largely inherit the same InfoNCE objective, whose optimization can prematurely saturate under a strong pre-trained initialization: on dense captions, the loss falls below 10^-3 on 80% of batches within the first epoch, while its gradient becomes numerically zero in 47% of measurements. We find that this behavior is closely related to the large number of near-duplicate captions in dense-caption benchmarks, where a few highly similar negatives remain unresolved after the easy majority has already been separated. As a remedy, we introduce HN-CLIP, which uses the text encoder's own text-text geometry to construct per-negative adaptive similarity margins. Specifically, a detached caption-similarity matrix is added to the negative logits, assigning larger margins to more similar captions without mining, synthesizing, or resampling negatives. The resulting objective requires only one caption-similarity matrix and a masked logit addition during training, with no auxiliary data, additional parameters, offline preprocessing, or inference-time overhead. Extensive experiments on four dense-caption retrieval benchmarks show that HN-CLIP improves over the strongest competitors by +2.5 to +4.0 R@1 while training 2.4x faster than GOAL and 5.4x faster than StructXLIP. Moreover, the proposed objective improves all six tested fine-tuning frameworks on the in-domain benchmarks and reaches the strongest full-data baseline with only 20% of the training data.

The most capable general-purpose AI (GPAI) models are mostly built in two jurisdictions, the United States and China, but the risks they carry land globally. Regionally advanced economies hosting no frontier developer, which we call AI middle-powers, are writing their own rules to govern GPAI. This paper investigates which GPAI-relevant provisions these AI middle-powers have enacted, mapping twenty jurisdictions including the European Union at the level of the individual provision, across four governance areas that trace the accountability chain for the model layer: systemic risk assessment, evaluation and verification, prohibitions with monitoring and detection, and serious incident reporting. Confirmed absence is recorded as data alongside positive provision. We find that jurisdictions converge on form, but diverge on force. Sixteen engage in at least three of the four governance areas, yet only about one in five provisions sit in binding law, and three-quarters of the instruments that do bind do so without defining GPAI. The institutional infrastructure shows the same shape: four in five of the mapped governance actors hold mandates that predate GPAI, and obligations attach wherever the inherited regime already reached, which is the application layer rather than the model. Where these states engage the model layer, they build capacity to observe it rather than impose duties on those who build it, and almost every evaluation body was constituted without the power to act on what it finds. Nominal coverage of the full accountability chain reaches eleven jurisdictions, but only five hold more than one provision in every area and, outside the EU, no jurisdiction imposes a binding evaluation duty on a model developer. The dataset gives researchers and policymakers a provision-level basis for identifying where regimes could align, and where coordination would have to start from scratch.