# Research Papers — 2026-09 · bundle page 191 of 266 > Full-text markdown of papers #9501–9550 of 13298 published in 2026-09, newest first. AI-readable mirrors of the full 13298-paper 2026-09 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-09/page/191 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/190.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/192.md) ### Epistemic orientation predicts legislative effectiveness among members of the US Congress Authors: Segun Aroyehun, Stephan Lewandowsky, David Garcia Published: 2026-09-10 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.11865v1 Truth and evidence-based communication provide important foundations for democratic governance, accountability, and collective decision-making. Prior work shows that evidence-oriented language in US congressional floor speeches has declined since the mid-1970s, alongside broader changes in legislative productivity and polarization. This study shifts the analysis from congressional sessions to individual members of Congress to examine whether epistemic orientation varies systematically across legislators and whether it relates to political behavior and legislative effectiveness. Using the Evidence-Minus-Intuition (EMI) score, we measure the relative prevalence of evidence-oriented versus intuition-oriented language in congressional floor speeches and Twitter posts. We link these measures to legislator-level data on ideology, institutional position, communication context, and Legislative Effectiveness Score (LES). The results show that more ideologically extreme members use less evidence-oriented language on the congressional floor. EMI also exhibits cross-platform consistency with members who use more evidence-oriented language in floor speeches also being more evidence-oriented on Twitter, although EMI is lower on Twitter overall. Finally, EMI in congressional speeches is positively associated with individual legislative effectiveness, even after accounting for ideology and extensive political, institutional, demographic, topical, and communication volume controls. These findings suggest that evidence-oriented language is not only an aggregate feature of congressional discourse but also a meaningful attribute of individual-level legislative communication and effectiveness. --- ### RetroThinker: Enabling Retrospective Thinking in Speech LLMs Authors: Yi-Jen Shih, Puyuan Peng, Abdelrahman Mohamed, David Harwath Published: 2026-09-10 Categories: eess.AS, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.11864v1 Speech large language models (SpeechLLMs) offer reduced latency and retain paralinguistic nuances that are typically lost in cascaded automatic speech recognition (ASR) and text-based LM architectures. However, they continue to lag behind text-only LLMs on complex reasoning tasks, while real-time spoken interaction imposes strict latency constraints. Although prior works employ Chain-of-Thought (CoT) and concurrent reasoning to enhance reasoning capabilities without inducing prohibitive delays, an inherent accuracy-latency trade-off persists. In this paper, we investigate whether a streaming SpeechLLM can dynamically revise its reasoning traces on the fly. We introduce RetroThinker, a multi-stage post-training framework that equips the Moshi model to self-verify and forward-correct CoT steps during inference. RetroThinker combines supervised fine-tuning (SFT) on curated retrospective thinking data with length-based direct preference optimization (DPO) to optimize retrospective during early reasoning (i.e., reasoning concurrently while the user speaks). Evaluated on the GSM8K benchmark, RetroThinker significantly improves the accuracy-latency trade-off over non-retrospective baselines, achieving an 11% absolute accuracy gain at a comparable latency. --- ### Explainability Assistant: A Conversational XAI Interface for Interpreting Energy Consumption Models Authors: Rodion Krjutškov, Eduard Barbu, Nikos Sakkas, Sofia Yfanti Published: 2026-09-10 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.11860v1 Energy consumption forecasting relies on increasingly complex machine learning (ML) models, such as Genetic Programming-based symbolic regressors, whose predictions can be difficult for facility managers and building operators to interpret. Explainable Artificial Intelligence (XAI) techniques address this opacity, but traditional XAI dashboards require substantial technical expertise and provide limited flexibility for dynamic, context-aware inquiry. Conversational XAI systems offer a promising alternative; however, previous approaches, such as TalkToModel, were constrained by rigid custom grammars and achieved only 76.8% intent-parsing accuracy. This paper introduces the Explainability Assistant, an open-source conversational XAI system that leverages the function-calling capabilities of modern Large Language Models (LLMs) to overcome these limitations. The system achieves 94% intent-parsing accuracy, supports flexible natural language interaction, and adapts to different ML problem types without task-specific fine-tuning. We present the system's architecture and report results from a comparative evaluation conducted with energy domain specialists, contrasting the Explainability Assistant with a traditional XAI dashboard. The evaluation suggests improved usability and consistent task accuracy, with all experts unanimously preferring the conversational interface for practical use. --- ### From Parameters to Answers: How LLMs Retrieve and Use Their Internal Knowledge Authors: Wenkang Wei, Yuan Fang, Renhe Jiang, Hong Cheng, Xingtong Yu Published: 2026-09-10 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.11859v1 How does a language model's dependence on query-routing information and target knowledge change as it answers a question? We study this question through layerwise interventions on the hidden state at the end of the question. Across Qwen, Llama, and Gemma, we compare country-continent questions with noun, adjective, and code answers while keeping several fitted measurements distinct. A pair-conditioned request direction describes which country is queried in natural single-country questions; a global request direction describes first- versus second-country requests in paired questions; separate selection candidates test control among contents already available in the hidden state. A diagnostic reanalysis of frozen Qwen natural-question states shows that the pair-conditioned direction grows stronger before interventions on it begin to alter later fitted knowledge, with this causal window opening while answer-supporting content is still forming. The paired three-model trajectories are not uniform: Gemma shows a partially overlapping mid-layer routing-content profile, whereas Llama has no sustained routing-effect window under the same gates. In the paired protocol, dependence on the global request direction decreases from fixed earlier to later layer sets while dependence on fitted content persists. A matched Qwen comparison shows that the pair-conditioned direction retains a late effect, so this operational handoff concerns the global fitted direction rather than all request information. These results separate early readability, natural strength, causal steering, and later content dependence. --- ### IndicTriMix: Developing Language Identification Datasets and Models for Tri-Language Code-Mixing Authors: Pruthwik Mishra, Rudra Trivedi, Avi Patel, Ashok Urlana, Shrikant Malviya Published: 2026-09-10 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.11851v1 Language identification in code-mixed text, largely observed in social media, is highly essential when users frequently switch between multiple languages within a single utterance. Accurately identifying the languages of code-mixed tokens becomes an urgent necessity. Traditional language identification models, designed for monolingual text, are not well suited for token-level language identification in code-mixed settings. We formulate the task as a sequence labeling problem and fine-tune contextual transformer-based models MuRIL and XLM-RoBERTa best suited for Indian languages. We evaluate these systems on three different data configurations (Hindi, Gujarati, and Bengali) to predict language labels for individual tokens. We release a benchmark for language identification in code-mixed tokens with manually annotated test sets. We propose two approaches of code-mixed generation using parallel sentences of three languages. The trained models demonstrate the effectiveness of contextual embeddings for token-level language identification in multilingual social media text. For reproducibility and to facilitate future research, we publicly release our fine-tuned models. --- ### Model-Aware Schedules Improve Generation via Fiberwise Optimal Transport Authors: Luyi Jia, Boyan Zhang, Yilun Liu, Steffen Rulands Published: 2026-09-10 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.11842v1 Diffusion and flow-matching schedules control the signal and noise coefficients that mix data and noise along affine probability paths. Minimizing a kinetic action defined on coefficient paths, motivated by optimal transport, helps explain strong baselines but remains model-agnostic and ignores prediction error. Here we introduce a model-aware schedule construction based on fiberwise optimal transport. At a fixed time and state on the probability path, compatible signal/noise decompositions form an affine fiber. We define a fiberwise prediction risk by averaging optimal-transport costs between the true and predictor-induced decompositions within these fibers. On a fixed coefficient curve, combining this risk with coefficient-path kinetic action yields a closed-form optimal time allocation. This construction extends to general linear prediction targets, and the risk profile can be estimated from an early baseline checkpoint. We evaluate DDPMs and flow matching across prediction targets, training configurations, risk-estimation checkpoints, datasets, and architectures. Our model-aware schedules consistently outperform strong baselines, including a 38.6% relative FID reduction for flow matching on CIFAR-10 at 16 function evaluations. Each model-agnostic kinetic baseline determines its own kinetic reference coordinate. In these coordinates, fiberwise-risk profiles from independently trained models in different settings align closely after normalization to unit area. The resulting schedule deformations used in training also align, suggesting empirical universality across the evaluated models and settings. Pretrained-checkpoint diagnostics extend this normalized-risk agreement to larger conditional latent diffusion and 2-RF models. A frozen analytic allocation template retains most of the model-aware improvement without further risk estimation or model-specific fitting. --- ### Target leakage, not model class, explains reported accuracy in survey-based cardiovascular screening: a leakage-tiered audit of glass-box and tabular foundation models Authors: Raad Bin Tareaf, Murad Al-Rajab, Samia Loucif, Samer Ellaham, Cedric Schmitz Published: 2026-09-10 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.11838v1 Cardiovascular screening models trained on national health surveys routinely report areas under the receiver operating characteristic curve (AUROC) near 0.89. We asked whether that accuracy reflects learning or target leakage, whether tabular foundation models change the answer, and whether the properties deployment requires survive joint examination. We benchmarked ten classifiers spanning linear, tree-ensemble, neural, glass-box, and tabular foundation classes for prevalent myocardial infarction in 442,067 respondents of the 2022 Behavioral Risk Factor Surveillance System across five feature tiers of decreasing leakage risk. Each was audited for discrimination, calibration, fairness at an explicit screening threshold, conformal coverage, explanation faithfulness, and inference cost, then applied -- models and thresholds frozen -- to 430,755 respondents of 2023. Removing two post-diagnostic features cost every model 0.049-0.051 AUROC, collapsing the field into a 0.0045-wide band. The glass-box explainable boosting machine was non-inferior to every alternative within a pre-specified 0.005 margin while scoring the cohort roughly 104 times faster than the strongest foundation model. One threshold detected 75.4% of women's infarctions against 89.0% of men's; editing the model's shape functions reduced the gap to 0.010. Marginal conformal prediction gave 0.86 coverage to men and 0.82 to adults over 60; Mondrian calibration repaired every stratum. Frozen models transported within 0.002 AUROC. Reported headroom in this literature is a property of the feature set, not the learner. Transparency cost nothing measurable and made fairness repair and uncertainty conditioning directly auditable. Evaluation practice, not model capacity, is the binding constraint. --- ### Quantitative Diffusive Limits for Singular Nonlocal Transport Authors: Andrea Agazzi, Giuseppe Bruno, Federico Pasqualotto, Philippe Rigollet Published: 2026-09-10 Categories: math.AP, stat.ML arXiv: https://arxiv.org/abs/2609.11837v1 We study the nonlocal continuity equation \( \partial_tμ_b =\operatorname{div}\!\left( μ_b\nabla\log\bigl((I-b^2Δ)^{-1}μ_b\bigr) \right) \) on a closed connected Riemannian manifold. For smooth strictly positive initial data, we prove that as $b \to 0$, its global solution converges to heat flow $μ(t)$ at the sharp, uniform-in-time rate \( \sup_{t\ge0}\|μ_b(t)-μ(t)\|_{L^1}\le Cb^2. \) The key estimate is the uniform dissipation of a $b$-weighted higher-order resolvent energy, which yields exponential relaxation despite the absence of a Wasserstein gradient-flow structure. On the circle, we also analyze the corresponding deterministic $N$-particle dynamics. A weak--strong modulated energy argument gives \( \mathbb E\!\left( \sup_{t\ge0}W_1(μ_b^N(t),μ_b(t)) \right) \le C(Nb)^{-1/2} \) for iid initialization. Consequently, the choice $b\asymp N^{-1/5}$ approximates heat flow uniformly in time at rate $N^{-2/5}$. --- ### Near-Optimal Reinforcement Learning with Multi-Step Transition Lookahead Authors: Corentin Pla, Hugo Richard, Marc Abeille, Vianney Perchet Published: 2026-09-10 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2609.11807v2 We study reinforcement learning (RL) with transition look-ahead, where the agent may observe which states would be visited upon playing any sequence of actions before deciding its course of action. Although look-ahead can substantially improve achievable performance, (1) showed that optimal planning with multi-step transition look-ahead is NP-hard. However, this hardness was established using a discount factor close to one. It was therefore unknown whether the problem remains hard for every discount factor, and whether near-optimal planning can nevertheless be performed efficiently. We resolve both questions. First, we show that for every fixed discount factor, exact planning remains NP-hard. Second, we introduce a randomized polynomial-time approximation scheme for every fixed look-ahead depth. Third, we extend our approach to account for unknown transitions. We empirically validate the soundness of our results on the wind-farm storage-control benchmark of (2), showing that our approach, optimally accounting for -step look-ahead information, offers substantially better performance than existing algorithms. (1) Corentin Pla, Hugo Richard, Marc Abeille, Nadav Merlis, Vianney Perchet : On the Hardness of Reinforcement Learning with Transition Look-Ahead (2) Chenbei Lu, Zaiwei Chen, Tongxin Li, Chenye Wu, Adam Wierman : Reinforcement Learning with Imperfect Transition Predictions: A Bellman-Jensen Approach --- ### Near-Optimal Reinforcement Learning with Multi-Step Transition Lookahead Authors: Corentin Pla, Hugo Richard, Marc Abeille, Vianney Perchet Published: 2026-09-10 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2609.11807v1 We study reinforcement learning (RL) with transition look-ahead, where the agent may observe which states would be visited upon playing any sequence of $\ell$ actions before deciding its course of action. Although look-ahead can substantially improve achievable performance, it is known that optimal planning with multi-step transition look-ahead is NP-hard, but this hardness was established using discount factors arbitrarily close to one. It was therefore unknown whether the problem remains hard for any discount factor, and whether near-optimal planning can nevertheless be performed efficiently. We resolve both questions. First, we show that for every fixed rational discount factor ($γ\in(0,1)$), exact planning remains NP-hard. Second, we introduce a randomized polynomial-time approximation scheme for every fixed look-ahead depth. We then extend our approach to unknown transitions and stochastic rewards using optimism and variance-adaptive confidence bounds. The resulting algorithm achieves cumulative regret whose leading term matches classical tabular discounted RL up to logarithmic factors. Thus, although exact planning with transition look-ahead is NP-hard, efficient near-optimal planning and learning remain possible. --- ### Understanding Operator Attitudes Toward AI-Supported Decision Making in Maritime Operations Authors: Doreen Jirak, Armeen Saroukanoff, Dirk van Rooy Published: 2026-09-10 Categories: cs.HC, cs.AI arXiv: https://arxiv.org/abs/2609.11805v1 Maritime Autonomous Surface Ships (MASS) and AI- supported decision assistants are expected to transform maritime operations, but their safe integration depends on how maritime professionals perceive and trust such systems. This paper presents a survey study on maritime stakeholders' attitudes toward an AI-supported assistant in collision-avoidance scenarios. Participants evaluated technology anxiety, trust in automation, and explanation quality using established and adapted questionnaires, complemented by sentiment and thematic analysis of open-ended responses Results indicate a generally positive disposition toward maritime technology, no clear age-related differences in openness, stable trust across scenarios, and more scenario-sensitive, multidimensional explanation ratings. Open responses showed that participants valued support for decision-making, situation awareness, and confidence-building, while raising concerns about AI reliability, over- reliance and loss of expertise. The findings suggest that maritime AI systems should not focus solely on increasing automation or trust, but on supporting calibrated reliance through transparent, reliable, and operationally meaningful design with domain experts in the loop. --- ### Logit Refiner: Improving Visual Autoregressive Models via Intra-Scale Dependency Modeling Authors: Meimingwei Li, Stefan Andreas Baumann, Felix Krause, Björn Ommer Published: 2026-09-10 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.11804v1 Visual Autoregressive Models (VAR) generate images through next-scale prediction, producing all tokens within each scale in parallel. We show that this parallel decoding constitutes a mean-field-style approximation that discards spatial dependencies among same-scale tokens, causing locally incoherent samples regardless of backbone capacity -- a limitation of the decoding rule. Addressing this limitation, we introduce the Logit Refiner, a lightweight autoregressive module that restores intra-scale dependencies by sequentially sampling tokens conditioned on frozen backbone features. Adding only ~10% parameters and less than 5% of the base model's training compute, it plugs into any pretrained VAR checkpoint without retraining. Controlled ablations isolate joint intra-scale sampling -- rather than additional capacity or training -- as the critical ingredient. Across backbones from 310M to 2B parameters on class-conditional ImageNet 256x256, the refiner consistently improves generation quality, enabling a 1.1B-parameter model to surpass one twice its size. The approach further generalizes to text-to-image generation, confirming that the mean-field bottleneck persists across VAR variants and is effectively alleviated by our method. Project page: https://compvis.github.io/logit-refiner/ --- ### Thinking with Looped Flows Authors: Ayhan Suleymanzade, Chanhyuk Lee, Floor Eijkelboom, Nicholas M. Boffi, İsmail İlkan Ceylan, Jinwoo Kim Published: 2026-09-10 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.11801v1 Humans and machines often solve harder problems by spending more time on computation. In deep learning, looped models implement this idea during inference by recurrently updating a hidden state. In practice, however, their training backpropagates through only one or a few updates, making it hard to train early updates to support future ones. We propose looped flows, an approach that sidesteps this issue by training the recurrence with local denoising objectives. By imposing temporal association across denoising objectives through progressively decreasing noise levels and shared noise, the model is incentivized to learn recurrent states that transfer useful computation over time, even when gradients cover only a few updates. We then formulate inference as integrating the velocity of a probability flow parameterized by the learned denoiser, coupled with recurrent states. This allows solving harder problems by spending more computation through a finer temporal grid and enables multiple valid predictions from different initial noise samples. Across six reasoning benchmarks including two multi-solution benchmarks, looped flows outperform prior state-of-the-art looped models overall, achieving 58.8% test accuracy on ARC-AGI-1 and 12.2% on ARC-AGI-2. --- ### SpecGuard: Inference-Time Backdoor Detection For Free Authors: Rui Wen, Ahmed Salem, Andrew Paverd, Mark Russinovich, Zheng Li Published: 2026-09-10 Categories: cs.CR, cs.CL arXiv: https://arxiv.org/abs/2609.11799v1 Large language models are often fine-tuned, shared, or downloaded from third parties, so a deployed model may carry a hidden backdoor that behaves normally on benign inputs but switches to attacker-controlled behavior when a secret trigger appears. While backdoors can be audited before deployment, runtime monitoring remains important for models that are frequently updated. The challenge is that LLM serving is latency-sensitive: existing inference-time detectors either rely on assumptions about the trigger form, which can fail on stealthy attacks, or require extra model computation, such as input perturbations or an additional generation pass. We introduce SpecGuard, an inference-time backdoor detector that repurposes speculative decoding at zero added model-computation cost. Speculative decoding speeds up inference by using a small draft model to propose tokens and a target model to verify them. We observe that this verification process already exposes a useful signal: when a backdoor is triggered, the target model shifts toward the attacker's behavior, while a clean draft model does not predict this shift, causing the draft-token acceptance rate to change. We formalize when this signal appears and show that an attacker who suppresses it must also weaken the backdoor. Across diverse backdoor types and model families, SpecGuard reliably detects triggered behavior, including stealthy cases where input-level filters are blind, while avoiding the extra generation cost of existing runtime detectors. Speculative decoding therefore doubles as a free, always-on signal for detecting backdoored LLM behavior. --- ### Dynamic language model representations for multi-objective reaction optimisation Authors: Joshua W. Sin, David Ming Segura, Bojana Ranković, Siu Lun Chau, Marius D. R. Lutz, Andrea Anelli, Ryan P. Burwood, Kurt Püntener, Maximilian J. Notheis, Raphael Bigler, Philippe Schwaller Published: 2026-09-10 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.11790v1 Optimising chemical reactions across multiple objectives, such as yield, selectivity, and safety, is central to chemical synthesis, and model-driven approaches depend critically on how reaction components are represented. Established featurisations are either chemically uninformative, as with one-hot encodings, or, as with molecular descriptors, do not readily extend across chemically distinct components. For structurally and functionally diverse components, it is therefore unclear what a shared representation should contain. Constructing such a representation is itself a challenging research undertaking that must be revisited for each new reaction system. Here we bypass this step by learning the reaction representation dynamically from text. Textual descriptions of reaction conditions are encoded by a fine-tuned language model trained jointly with Gaussian process surrogates, yielding task-adaptive representations within a multi-objective Bayesian optimisation loop. Across nickel- and palladium-catalysed cross-couplings in both sequential and parallel experimentation regimes, this approach reaches optimisation convergence in fewer experiments than descriptor libraries or one-hot encoding. Applied prospectively to a palladium-catalysed cyanation spanning mixed ligand denticity and heterogeneous additives, and to a three-objective asymmetric hydrogenation across chiral iridium and ruthenium catalyst families, two rounds of high-throughput experimentation (192 reactions, under 3% of each design space) delivered conditions translating directly to gram scale in 94% and 84% isolated yield, the latter at 99.6% enantiomeric excess. --- ### Beyond Word Error Rate: A Switch Aware Evaluation of ASR and Audio Language Models on English Yoruba Code-Switched Speech Authors: Chibuzor Okocha, Christan Earl Grant Published: 2026-09-10 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.11786v1 Automatic speech recognition (ASR) systems and audio language models (audio LMs) now report low error rates on monolingual benchmarks, but their behavior on code switched speech in low resource, diacritic rich languages remains poorly characterized. We present a switch aware evaluation of eleven modern systems (six ASR models and five audio LMs) on English Yoruba code-switched speech, using a deterministic 2000 utterance evaluation set and a shared scoring pipeline. Beyond word error rate (WER), we report switch localized diagnostics: a switch entry token error rate (SETER), windowed switch point error rates, language specific error rates, and a diacritic insensitive WER. Our central finding is that aggregate WER hides code switching behavior. The best system by WER (an ASR model) is statistically indistinguishable from a leading audio LM on WER, yet the audio LM is significantly better on every switch localized metric. Across faithful systems, Yoruba token recognition collapses (error 0.97 for almost all systems) while English tokens are recognized far better, and errors concentrate sharply at switches into Yoruba. Several generative audio LMs fail as exact transcribers, producing translation, verbosity, and prompt leakage that are strongly prompt dependent. We release manifests, metric implementations, and evaluation scripts to support reproducible, switch aware benchmarking for African code switched speech. --- ### Predicting Privacy Leakage from Weight Spectral Density Authors: Richard J. Preen, Jim Smith Published: 2026-09-10 Categories: cs.LG, cs.CR, cs.NE arXiv: https://arxiv.org/abs/2609.11780v1 Membership inference attacks (MIAs) are widely used to audit the privacy disclosure risk of machine learning models, however current state-of-the-art attacks require training computationally expensive shadow models, making large-scale privacy evaluation impractical. In this work, we investigate whether inexpensive spectral metrics derived from the heavy-tailed self-regularisation framework can serve as proxies for MIA vulnerability. We evaluate several WeightWatcher spectral metrics on image and tabular classification tasks and compare their relationship with MIA privacy leakage against conventional measures of generalisation. Across datasets, stable rank exhibits a strong positive correlation with overall MIA success, while Log alpha-Norm shows a consistent negative correlation with MIA vulnerability at the low false-positive regime. These associations are observed to be stronger than those obtained using the generalisation gap. The results indicate that neural network spectra may contain information about privacy leakage that is not fully captured by conventional measures of overfitting, motivating spectral analysis as a promising direction for scalable privacy auditing. --- ### Differentially Private EEG Feature Anonymization: A Privacy-Utility Case Study in Clinical Neurophysiology Authors: Noman Sadiq, Mohsen Toorani Published: 2026-09-10 Categories: cs.CR, cs.LG, eess.SP arXiv: https://arxiv.org/abs/2609.11777v1 Clinical electroencephalography (EEG) data are valuable for healthcare research and for developing artificial intelligence (AI)-based clinical decision-support systems, but EEG recordings and derived features may contain sensitive patient-specific information. This creates privacy risks when data are reused, analyzed, or shared across clinical and research environments. Conventional anonymization methods are often insufficient for high-dimensional biomedical signals, since removing direct identifiers does not necessarily prevent re-identification, linkage, or inference risks. At the same time, strong privacy protection may distort clinically relevant signal characteristics and reduce data utility. This paper studies subject-level differential privacy for protecting clinical EEG-derived feature representations using Gaussian and Laplace perturbations. The proposed framework considers three deployment scenarios: client-side anonymization, centralized server-side anonymization, and decentralized local training. Following EEG preprocessing and feature extraction, Gaussian and Laplace perturbations are applied to the resulting patient-level EEG feature representations. The Laplace experiments evaluate the implemented noise scales, while the scales required for formal full-vector calibration are derived separately. The effects of both perturbations are assessed using statistical utility measures and a downstream machine-learning-based utility check. The results show that differentially private perturbation can be integrated into EEG processing workflows, but the selected mechanism, privacy parameters, and sensitivity calibration strongly influence data utility. The study highlights the practical privacy-utility trade-off in DP-based EEG feature anonymization and the challenges of preserving downstream utility in small and imbalanced clinical EEG datasets. --- ### Whisper-Based Speech Transcription from Videos Across Multiple Languages for Cross-Cultural Understanding Authors: Michael Picheny Published: 2026-09-10 Categories: eess.AS, cs.CL arXiv: https://arxiv.org/abs/2609.11772v1 Cross-cultural understanding has become increasingly important in today's highly connected, cross-national world. The success of LLM-based technologies is now driving the development of automated tools to aid understanding for nonnative people trying to succeed in cross-cultural environments. Building such automated tools is often done by leveraging in-thewild text, audio, and video data. This paper presents techniques for improving speech recognition-based transcript creation in multiple languages from videos to better train these automated tools. The focus is on processes and speech tools that can easily be used by cross-cultural tool builders without requiring deep speech processing expertise. Using publicly available videos from YouTube and Whisper-based tools, average transcription error rate across seven languages (Spanish, Japanese, Korean, Mandarin, Turkish, Russian, and Hebrew) of 30% are observed. With a modest amount of fine-tuning data, the average error rate can be reduced to 20% making such output much more usable for downstream processing. Speech and metadata associated with these videos that can be used by the community to further refine these experiments are released as well. --- ### The widening evaluation gap in medical large language model research 2023 to 2026 Authors: Raad Bin Tareaf, Murad Al-Rajab, Samia Loucif Published: 2026-09-10 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.11770v1 Large language models are superseded every few quarters; clinical evidence takes years. We asked whether medical research is keeping pace with the systems it evaluates. PubMed returned 11,628 records for January 2023 to June 2026 across fourteen clinical domains, growing 45-fold; 2.5% used a randomised, controlled or prospective design. Evaluation lag, from a study's newest named model release to its own publication, widened from 1.33 to 6.08 quarters. Because discontinued models age mechanically, we benchmarked this against a counterfactual holding model composition fixed: migration to newer systems offset only 56% of the drift (95% CI 50-65). Randomised trials evaluated models a median 4.6 quarters older than other designs (P = 3 x 10^-19), yet among studies naming a model still under development no design differed from any other; 62% of randomised trials evaluated a discontinued family. Rigour and currency are in tension, and that tension reflects model selection rather than research timelines. --- ### Recognizing Is Not Reversing: A Controlled Inversion Test of Fact-Preserving News Framing Authors: Yi Liu Published: 2026-09-10 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.11769v1 Large language models (LLMs) are increasingly used to analyze and rewrite news, yet current framing studies mainly evaluate generation, detection, or whether rewritten text appears more neutral. They do not directly show whether a model can undo a known framing transformation while keeping the facts fixed. We introduce a controlled inversion test over three established textual realizations of framing: evaluative lexis, agency realization, and information salience. Across 60 news articles and three intervention strengths, this yields 540 paired variants with preserved atomic facts and recorded edits. Across Qwen, DeepSeek, and Kimi, factual preservation remains near 0.84, whereas intervention reversal is 0.044--0.068. Even when both framing type and direction are recognized correctly, pooled reversal reaches 0.071. These results reveal a clear separation between factual fidelity, framing recognition, and framing inversion: recognizing how an article is framed does not imply that the framing can be undone. --- ### A Unified Per-Token Gating Family for On-Policy Distillation: FKL/RKL Mixing with Multi-Channel and Bias Coefficients Authors: Suwan Wu, Yumeng Lin, Pengcheng Yuan, Xiaolong Jiang Published: 2026-09-10 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.11768v1 Per-token gating of forward/reverse KL losses has become a standard technique for on-policy knowledge distillation (OPD), but existing methods such as EOPD (Jin et al., 2026) and ToDi (Jung et al., 2025) each fix a single gating signal and a single gating direction, and the two have never been compared directly. We introduce a four-coefficient parameterization lambda_t = sigma(a * h_t + b * u(x) + c + d * gap_t) in which direction-aligned proxies of EOPD and ToDi appear as one-dimensional (1D) restrictions, and which adds multi-channel composition and an explicit bias as further degrees of freedom. On TweetEval (Barbieri et al., 2020) emotion and hate, with a Qwen3-32B teacher and a Qwen3-4B student, configurations in the full family reach higher accuracy than the matched-magnitude single-channel (entropy-only / gap-only) 1D restrictions in 33 of 36 comparable cells, and a 26-cell mean-match isolation experiment places dynamic gating ahead of effective-KL-matched static baselines in 19 of 26 cells. Because cells share training data, models, and parameter substructure, we report both counts as exploratory aggregate directional evidence rather than as independent hypothesis tests. Targeted three-seed paired replications of the nine headline comparisons singled out by that sweep -- including a third task, offensive -- are directionally consistent, but individually smaller than the single-seed estimates and not significant at n=3. We therefore present the parameterization primarily as a shared coordinate system for comparing per-token gating designs in short-output classification OPD. --- ### Component-Aware Differential Privacy for Federated Multilingual Speech-LLMs Authors: Jordi Luque, Fernando López, Aleix Sant Published: 2026-09-10 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.11762v1 Per-layer differential privacy (DP) clipping improves gradient fidelity in federated learning by allocating per-matrix clipping budgets proportional to parameter count. We show that this recipe breaks for speech large language models (speech-LLMs), when the acoustic encoder and the language decoder differ by an order of magnitude in update norm. Single-pool per-layer methods suffer \emph{cross-component budget collapse}, dragging word error rate (WER) far from flat global clipping or collapsing training entirely. When the norm imbalance is milder, adaptive single-pool methods partially recover, confirming that collapse severity scales with the inter-component norm ratio. We empirically diagnose the root cause across six per-layer methods and three speech-LLM architectures. We then propose \emph{$α$-split}, a two-pool allocation that normalises encoder and LLM parameters into independent pools, and show that joint $\ell_2$ sensitivity and the original $(\varepsilon,δ)$-DP guarantee are unchanged. At architecture-calibrated $α$, our method recovers WER utility compared to flat DP, while granting the encoder $4.47{\times}$ tighter per-component noise protection against speaker voice-based gradient-inversion attacks at only $+2.6\%$ LLM noise overhead. --- ### RAG-Safety-Bench: Reliable Evaluation of Retrieval-Augmented LLM Safety Authors: Adithiyan Rajan Indira Saravanan, Kathleen C. Fraser Published: 2026-09-10 Categories: cs.CL, cs.IR arXiv: https://arxiv.org/abs/2609.11758v1 Allowing large language models (LLMs) to retrieve information from a set of trusted documents can increase reliability and reduce hallucination. However, recent work has demonstrated that retrieval-augmented generation (RAG) can have unintended side effects on the overall safety of the generated responses, when prompted for harmful or dangerous content. A clearer understanding of the mechanisms leading to this result is needed, as increasing numbers of end users turn to RAG to incorporate corporate documents and knowledge bases into LLM-based systems. We introduce RAG-Safety-Bench, a benchmark to measure the safety impact of RAG on LLM models. By removing the confounding effect of retriever quality, and cleanly separating the problem into four conditions -- non-RAG, RAG with an oracle document containing the answer to the harmful request, RAG with documents related to the harmful request but without the specific answer, and RAG with random, safe documents -- the benchmark isolates the impacts of different factors in the observed safety degradation. We report results across five open-source LLMs, showing an inverse relationship between benign and unsafe capability, strong evidence that baseline safety guardrails do not lead to downstream safety guarantees in the RAG case, and model-specific support for previous findings that even benign documents can lead to unsafe generation in retrieval-enabled systems. --- ### SIRF: A Spec-Internalized Risk Foundation Model for Industrial Content Risk Control Authors: Suwan Wu, Yumeng Lin, Pengcheng Yuan, Xiaolong Jiang Published: 2026-09-10 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.11752v1 For industrial content risk control, the real deployment constraint is not average accuracy but how much risk can be auto-handled under high precision and second-level latency. We present SIRF (Spec-Internalized Risk Foundation Model), which internalizes a platform's complex policies, synthesized without additional human annotation via EntiGraph, MAGA rewriting and account-level chain-of-thought (CoT), into the weights via continued pretraining (CPT), so rules are applied at high precision under an ultra-low-latency, verdict-only deployment. A controlled same-source comparison (Qwen3-8B-SFT vs. SIRF-8B-SFT, identical policy injection and verdict-only output form, differing only in policy-grounded CPT) attributes the gain to internalization: SIRF-8B-SFT reaches 71.3% Black Recall@P95, +15.1pp over the baseline, using only ~70M CPT tokens without harming general ability, and among included, logprob-available models under this interface it matches or exceeds far larger systems. SIRF is deployed as a tree-model adjudication layer (20% more mis-penalized samples recovered) and transfers to a freezing scenario at low cost (~70% relative mis-penalization reduction). --- ### Sparsity Regularized and Robust Mean Variance Portfolio Selection Under Ellipsoidal Uncertainty Authors: Deniz Akkaya, Emre Can Yayla, Buse Şen, Mustafa Ç. Pınar Published: 2026-09-10 Categories: math.OC, cs.LG, stat.ML arXiv: https://arxiv.org/abs/2609.11749v1 We investigate mean-variance portfolio selection with an $\ell_0$-penalty to promote sparsity in asset allocations. Uncertainty in the mean return vector is incorporated through an ellipsoidal uncertainty set, yielding a robust sparse optimization framework. We characterize the structure of both local and global minimizers and exploit these properties in the risk minimization and return maximization formulations. Building on this structural insight, we develop a branch-and-bound algorithm tailored to the resulting robust sparse portfolio problems, together with a new pruning rule that can discard exponentially many candidate portfolios in a single step. Extensive computational experiments on real market data, together with comparisons against a mixed-integer second-order cone programming solver, demonstrate the effectiveness and competitiveness of the proposed approach. --- ### Building py-kvcache: A Performance Characterization of External KV Caching for vLLM with NVMe SSDs Authors: Joseph Kanichai, Tiziano De Matteis, Animesh Trivedi Published: 2026-09-10 Categories: cs.DC, cs.LG arXiv: https://arxiv.org/abs/2609.11744v1 Prefix caching can reduce the time to first token (TTFT) of long-context LLM requests by reusing previously computed key-value (KV) states, but for short prefixes or fast GPUs, recomputation can be faster than loading from an external cache. We characterize this tradeoff in vLLM across GPU, CPU, and NVMe tiers using synthetic workloads, long-context benchmarks, production traces, and find that cache performance depends on transfer granularity, intermediate memory use, and when transfers enter the request schedule, not only on device bandwidth. These findings motivate py-kvcache, a vLLM KV Offload connector with asynchronous direct I/O, bounded shared staging, and scheduler-aware preloading, which starts disk reads while requests are still waiting, overlapping with compute. At 80k tokens, py-kvcache loading from disk is 2.0x faster than LMCache, with preloading contributing 1.34x. With GPU, CPU, and disk caching enabled, it is 1.23x faster than LMCache and within approximately 4% of the native vLLM KV Offload implementation. LongBench and SCBench show that these benefits extend to irregular prefix chains and multi-turn workloads. Bailian trace replays improve TTFT on a weaker GPU, but on an H100 the average request falls below the break-even point and GPU memory alone retains enough prefixes. External KV caching should therefore be treated as a setup specific admission decision. The py-kvcacheimplementation is available at: https://github.com/atlarge-research/py-kvcache. --- ### LOCUS: Task-Aware Low-Rank Post-Training for Token-Efficient Language Generation Authors: Dongfang Zhao Published: 2026-09-10 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.11739v1 Large language model serving costs scale directly with output sequence length, yet standard preference alignment often inflates response verbosity without improving utility. We study whether the parameterization of post-training updates affects generation length: low-rank subspaces alter sequence length without modifying the alignment loss. We present LOCUS, a method that selects a task-aware low-rank adaptation subspace to minimize output-token cost subject to a utility constraint. Within this subspace, post-training retains the native preference objective with a frozen backbone. Across Anthropic HH-RLHF dialogue preferences, we evaluate two $\sim$3B decoder backbones, Pythia-2.8B and Qwen2.5-3B, against protocol-matched full-parameter DPO and DrDPO branches and the released SamPO checkpoint. LOCUS reduces continuation length by up to 39.84\% on Pythia-2.8B and by 14.87--17.58\% on Qwen2.5-3B while updating only 0.24--0.28\% of model parameters, with no material change in the internal preference diagnostic. --- ### Organizational Principles Enable Collective Intelligence in Embodied AI Authors: Zhengran Ji, Jonathan Hyun, Boyuan Chen Published: 2026-09-10 Categories: cs.MA, cs.AI, cs.LG, cs.RO arXiv: https://arxiv.org/abs/2609.11737v2 Collective intelligence depends not only on the capabilities of individual members, but also on how those members are organized. Yet artificial multi-agent systems are typically assembled using fixed organizational structures, even when the physical tasks they perform impose fundamentally different coordination requirements. Here we show that principles from human organization theory can be operationalized to organize large, heterogeneous collectives of embodied artificial agents. We introduce ORCH (Organizing Roles and Coordination Hierarchies), which constructs task-specific hierarchical organizations by combining pooled interdependence for work that can proceed concurrently with sequential interdependence for work governed by prerequisite relationships. Across 25 wildfire-response missions spanning reconnaissance, rescue, transportation, resource management, containment and suppression, we evaluated teams of up to 50 heterogeneous agents using eight large language models. Organizations constructed using these principles consistently outperformed four representative embodied multi-agent approaches across mission outcome, execution efficiency, exploration and computational resource use. Human-designed ORCH organizations improved final score by 63.97% and execution efficiency by 74.29% on average relative to the four prior frameworks. Organizations generated automatically by language models improved these measures by 43.63% and 52.53%, respectively. These advantages persisted across missions and underlying language models. Notably, collective performance was not monotonically determined by model scale. Analysis of long-horizon missions showed that hierarchical organization enabled teams to preserve concurrent activity within specialized groups while coordinating ordered transitions between mission phases. --- ### ORCH: Organizational Principles Enable Collective Intelligence in Embodied AI Authors: Zhengran Ji, Jonathan Hyun, Boyuan Chen Published: 2026-09-10 Categories: cs.MA, cs.AI, cs.LG, cs.RO arXiv: https://arxiv.org/abs/2609.11737v1 Collective intelligence depends not only on the capabilities of individual members, but also on how those members are organized. Yet artificial multi-agent systems are typically assembled using fixed organizational structures, even when the physical tasks they perform impose fundamentally different coordination requirements. Here we show that principles from human organization theory can be operationalized to organize large, heterogeneous collectives of embodied artificial agents. We introduce ORCH (Organizing Roles and Coordination Hierarchies), which constructs task-specific hierarchical organizations by combining pooled interdependence for work that can proceed concurrently with sequential interdependence for work governed by prerequisite relationships. Across 25 wildfire-response missions spanning reconnaissance, rescue, transportation, resource management, containment and suppression, we evaluated teams of up to 50 heterogeneous agents using eight large language models. Organizations constructed using these principles consistently outperformed four representative embodied multi-agent approaches across mission outcome, execution efficiency, exploration and computational resource use. Human-designed ORCH organizations improved final score by 63.97% and execution efficiency by 74.29% on average relative to the four prior frameworks. Organizations generated automatically by language models improved these measures by 43.63% and 52.53%, respectively. These advantages persisted across missions and underlying language models. Notably, collective performance was not monotonically determined by model scale. Analysis of long-horizon missions showed that hierarchical organization enabled teams to preserve concurrent activity within specialized groups while coordinating ordered transitions between mission phases. --- ### Learning structural balance of graphs from quantum spectral features Authors: Stefano Scali, Oleksandr Kyriienko Published: 2026-09-10 Categories: quant-ph, cond-mat.dis-nn, cs.LG, cs.SI arXiv: https://arxiv.org/abs/2609.11736v1 We develop a quantum approach to spectral feature extraction from the density of states (DOS) of a problem-dependent Hamiltonian, and apply it to machine learning on signed graphs. We propose to embed a signed graph as an Ising model instance with positive and negative interactions, and use the standardized moments of the Ising DOS as features for learning. We show that these moments count signed closed walks, are switching-invariant, and are size-free by construction. As a benchmark, we target learning the frustration index, an NP-hard measure of structural balance that can be labeled exactly at moderate size. At zero field, the models can be sampled classically, allowing the quantum extraction procedure to be certified against exact ground truth. We propose DOS-QPE, a phase estimation on a purified maximally mixed probe, which samples the spectral density with orders of magnitude fewer shots than Hadamard test-based trace sampling and feeds the resulting features directly into classically trained models. On $1.4\times10^5$ labeled graphs the exact DOS determines the frustration index, and five moments recover it with a mean error of 0.4, well below one sign flip. Beyond zero field, the underlying trace-estimation problem is DQC1-complete, providing access to spectral features for which no efficient classical sampling method is known. Our work opens routes towards quantum applications in social network balance analysis, spin-glass studies, correlation clustering, and protein-interaction networks. --- ### Reflex-Informed Neuromuscular Reinforcement Learning for Muscle-Driven Locomotion Authors: Jian Zhou, Xingyu Zhang, Rui Ma, Yu Cao, Shane Xie, Zhi-qiang Zhang Published: 2026-09-10 Categories: cs.RO, cs.GR, cs.LG arXiv: https://arxiv.org/abs/2609.11733v1 Muscle-driven locomotion provides a physically grounded approach to generating realistic human movement. However, achieving both physiological plausibility and adaptability to changes in musculoskeletal capacity and external disturbances remains a fundamental challenge. To address this limitation, we propose a Reflex-Informed Neuromuscular Reinforcement Learning framework for muscle-driven locomotion. Within this framework, a fixed phase-dependent reflex controller serves as the underlying neuromuscular control mechanism, while the reinforcement learning policy produces four biomechanically meaningful residual parameters to modulate key reflex gains and thresholds associated with hip swing, knee support, and ankle propulsion according to the current state. Experimental results demonstrate that the proposed framework generates physiologically plausible locomotion with improved kinematic accuracy and dynamic consistency, as well as better bilateral symmetry and stride-to-stride consistency under nominal walking conditions. The learned policy remains robust under muscle weakness and external perturbations without retraining. --- ### Continuous-Time Acoustic Modelling with Neural Controlled Differential Equations Authors: Mattias Cross, Minghui Zhao, Anton Ragni Published: 2026-09-10 Categories: cs.SD, cs.AI arXiv: https://arxiv.org/abs/2609.11725v1 Text-to-speech (TTS) models commonly address text--speech alignment by expanding phone-level encoder states to frame-level decoder inputs using predicted durations. While this length-regulation step resolves alignment structurally, this use of duration typically changes only where and how often latent states appear, not the values of the states themselves. This paper proposes a continuous-time mechanism for duration-aware acoustic modelling in TTS using neural controlled differential equations (CDEs). We formulate the phone representation as a temporally parameterised control path and use a neural acoustic vector field to produce a continuous-time hidden state whose values evolve with phonetic content and duration-derived timing. The resulting trajectory can be sampled at discrete points and integrated into a standard acoustic decoder pipeline. Objective results contrast CDEs and typical recurrent models. Subjective results suggest that CDE-based models evaluating one phone per step can improve rank-order agreement between synthesised and reference emotion intensity while maintaining comparable emotion-expression quality to a strong baseline. Additional experiments with half-phone step-sizes suggest that temporal resolution changes the trade-off between style tracking and absolute calibration. These results position CDEs as a promising design space for continuous-time and duration-aware style-sensitive TTS. --- ### The Eloquence submission for Task 2 of the Interspeech 2026 MLC-SLM challenge Authors: Jordi Luque, Lorenzo Concina, Marco Matassoni, Alessio Brutti, Filippo Vella Published: 2026-09-10 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.11724v1 This paper details the Eloquence team's approach to Task 2 of the 2nd MLC-SLM challenge at Interspeech 2026, which involves multilingual Multiple-Choice Question Answering (MCQA) across 21 languages. Three approaches are explored. First, we fine-tune Voxtral-Mini-3B via LoRA with cross-lingual data augmentation, ASR transcript augmentation and timestamp-aware audio cropping, achieving 0.72 macro-accuracy on evaluation Phase 2. Second, we apply multimodal in-context learning (ICL) to the frozen Voxtral-24B model to correct a strong label bias, reaching 0.81, our best result. Third, a training-free retrieval system based on a three-layer voice-anchored memory combining acoustic identity, semantic content, and a knowledge graph achieves 0.68. All three systems substantially outperform the official baseline. --- ### Revisiting Avatar-As-Image: High-Fidelity Registration is All You Need Authors: Margaret Kostyrko, Yuxuan Xue, Garvita Tiwari, Gerard Pons-Moll Published: 2026-09-10 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.11722v1 The representation of 3D clothed humans as standardized 2D UV texture and displacement maps over an underlying body model has long been studied. This compact representation is enticing as it enables pretrained image networks to process, generate, and edit 3D avatars, but is only useful if scans are accurately aligned and brought into correspondence via high-fidelity registration. This prerequisite has never been met, which we argue explains the limited quality of prior UV-based methods for clothed humans. Despite its significance, no public method produces high-fidelity SMPL(-X)+D registrations with UV texture from arbitrary clothed scans. We present AvaImg, a multi-stage optimization pipeline, to close this gap: it enforces body-inside-clothing constraint via signed winding numbers, made viable by a three-level efficiency cascade (~10x runtime reduced, ~95% storage saved), and recovers fine surface detail using coarse-to-fine displacement optimization. AvaImg outperforms all baselines in body fitting, shape estimation, and surface registration across six datasets, yielding textured registrations near-indistinguishable from scans (PSNR=34.48dB). For validation of AvaImg's Avatar-as-Image representation as imminently compatible with image foundation models, we auto-encode our UV maps via the frozen FLUX VAE. This achieves only 0.76mm added Chamfer error relative to scan and shows that the resulting maps lie within natural-image distributions, supporting the use of 2D generative priors for 3D avatar generation. Code, data, and Singularity containers will be at https://yuxuan-xue.com/avaimg. --- ### MC-DeTra: Motion-Consistent Joint Object Detection and Socially-Aware Trajectory Forecasting in Bird's-Eye-View Images Authors: Vladislav Diuzhev, Dmitry Yudin Published: 2026-09-10 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2609.11717v1 Unified models for object detection and trajectory forecasting aim to merge perception and prediction for autonomous driving, refining actor trajectories directly over shared bird's-eye-view (BEV) images rasterized from LiDAR and high-definition maps. Their accuracy on dynamic, moving actors, however, remains the hardest part of the task, and the strongest such model, DeTra, has no public implementation. We contribute an openly released DeTra reimplementation with documented approximations, and on top of it MC-DeTra: a family of motion-consistency mechanisms that add supervision through two annotation-derived auxiliary signals -- each actor's observed past motion and the occupancy of the surrounding traffic that forms its social context -- and one inter-output consistency constraint that aligns an actor's predicted heading with its predicted direction of motion. Every proposed loss is train-only and inference-safe: it shapes the shared BEV representation during training and is removed at test time, adding no inference latency. On the Waymo Open Dataset, evaluated under a strict, detection-conditioned forecasting protocol, MC-DeTra improves dynamic, socially-situated trajectory forecasting while preserving or improving detection accuracy; a gradient-based loss-calibration analysis exposes how the auxiliary objectives compete at the shared backbone, and our ablation identifies which signals contribute most. We release code, configurations, and evaluation tooling at https://github.com/diuzhevVlad/MC-DeTra. --- ### Why Does Post-Training Quantization Work? Authors: Yuxiang Chen, Michael Beyer, Jun Zhu, Jianfei Chen Published: 2026-09-10 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.11716v1 Post-training quantization compresses large language models (LLMs) by storing their weights at reduced precision, and each quantized weight introduces an error into the hidden states. Naively, these errors should accumulate with depth and corrupt next-token prediction; randomly initialized models accumulate these discrepancies rapidly, whereas quantized pretrained models accumulate much less hidden-state error and largely maintain downstream task performance, even though they were never trained with quantization noise. This raises the question we address: why does post-training quantization work? Comparing full-precision and quantized forward passes, we identify two mechanisms that characterize pretrained quantization robustness. First, the error a layer newly introduces tends to oppose the error it inherits from the layer's input. The two cancel partially such that the discrepancy between full-precision and quantized passes grows slowly. This counteracting residual interaction develops during pretraining. Our quantitative analysis identifies it as a major factor slowing hidden-error growth. Second, LM-head geometry preferentially preserves the scores and probabilities of high-ranked tokens, which typically represent the model's most confident predictions. Together, these mechanisms explain why quantization error that passes through numerous layers can still produce only small output changes, and we verify the findings across models and quantization settings. --- ### A Time-Based Readout for Vector-Matrix Multiplication in Fully Analog Memristive SNNs Authors: Elia Mateu-Barriendos, Álvaro Gómez-Pau, Josep Rius, Daniel Arumí, Rosa Rodríguez-Montañés, Salvador Manich Published: 2026-09-10 Categories: cs.ET, cs.AI, cs.AR arXiv: https://arxiv.org/abs/2609.11713v1 Artificial neural networks rely on vector-matrix multiplications (VMMs), whose implementation in von Neumann architectures is dominated by costly data movement between memory and processing units. Spiking neural networks (SNNs) mitigate this bottleneck by performing in-memory, analog VMMs using memristive crossbar arrays. However, conventional current-mode readout circuits incur significant area and power overhead. This work proposes a fully analog readout architecture based on voltage-to-time conversion of the VMM output. By sensing the column voltage, the proposed approach avoids current-mode summing and scaling circuitry, improving area and energy efficiency. Post-layout simulations of a 10x1 SNN implemented in a 130 nm CMOS technology validate the proposed architecture, while application to a trained 64x10 SNN for digit classification further demonstrates its feasibility for SNN inference. --- ### Generalization Analysis of Distributed Kernel-based Robust Gradient Descent Algorithms Authors: Jun-Yi Meng, Zheng-Chu Guo, Yuan Mao Published: 2026-09-10 Categories: stat.ML, cs.LG, math.OA, math.PR arXiv: https://arxiv.org/abs/2609.11712v1 In this paper, we investigate the generalization performance of distributed gradient descent algorithms in a reproducing kernel Hilbert space under a robust loss function $l_σ$. By exploiting the spectral characterization of gradient descent together with the intrinsic properties of robust loss functions, we establish optimal learning rates for the distributed kernel-based robust gradient descent (DKRGD) algorithm with an appropriately chosen scale parameter $σ$. The proposed parameter choice of $σ$ simultaneously alleviates the saturation phenomenon and guarantees statistical robustness. A key technical contribution is a novel error analysis that provides substantially sharper bounds for products of operators, thereby significantly relaxing existing restrictions on the maximum number of local machines while retaining optimal learning rates. Finally, we develop a communication-efficient strategy that further improves the convergence performance of DKRGD. --- ### When Agents Disagree: Bayesian Backward Reasoning as a Label-Free Anchor for Multi-Agent Collective Decision-Making Authors: Ken Chen, Wei Wang, Sachith Seneviratne, Hansani Weeratunge, Saman Halgamuge Published: 2026-09-10 Categories: cs.AI, cs.MA arXiv: https://arxiv.org/abs/2609.11709v1 When multiple LLM agents yield conflicting answers, the decision-making process dictates whether agent diversity improves performance or merely compounds shared errors. Existing collective decision-making methods, including voting, electoral rules, and LLM judges, rely on forward reasoning: they map evidence to labels in one direction. Although these methods can combine diverse forward traces, they still aggregate estimates that share this evidence-to-label factorization and can inherit correlated errors within the forward pool. We therefore construct a reverse posterior for each instance through Bayesian backward reasoning from an explicit likelihood. The forward and reverse posteriors provide differently factorized approximations of the underlying posterior. Because estimates from different factorizations may tend to share the same error less often, we use Jensen-Shannon divergence to rank agents by cross-path consistency. This cross-path consistency signal underlies three strategies: hard selection (MinJS), soft reweighting (FwdJS), and log-linear fusion (LogLin). Evaluated on DDXPlus across five LLM backbones, our proposed strategies show consistent improvements: MinJS outperforms random selection across all backbones, FwdJS generally improves over the strongest baseline, and LogLin achieves the best performance among the evaluated methods, with its largest gains on the subset where the agents disagree. Despite its weaker standalone accuracy, the reverse posterior serves as a more useful anchor than forward-only alternatives, providing complementary information for collective decision-making. When labeled data are available, a lightweight two-stage calibration can further refine the reverse anchor and improve aggregation performance. --- ### Language-Augmented Semantic Priors for B-Spline Surface Fitting Authors: Yunzhong Lou, Yusheng Luo, Jiahao Li, Yu Song, Xiangdong Zhou Published: 2026-09-10 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.11708v1 The use of B-splines and Non-Uniform Rational B-Splines surfaces constitutes the mathematical foundation of contemporary computer-aided design (CAD) systems. Despite long-term progress, geometric kernels in traditional CAD still rely heavily on predetermined heuristic initialization for surface fitting and parameterization. Meanwhile, the procedural semantics and design intent encoded in modeling histories are largely ignored during geometry generation. This disconnect creates a gap between high-level design intent and solver-executable geometric configuration, often leading to suboptimal and semantically inconsistent fitting results. To bridge this gap, we introduce LASP, a Language-Augmented Semantic Priors framework that leverages large language models (LLMs) to infer structured, solver-usable B-spline priors from procedural modeling histories. Rather than modifying the geometric kernel itself, LASP operates as a semantic reasoning layer above existing solvers. It first translates modeling histories into rich textual descriptions that capture design intent, geometric context, and functional relationships, and then uses a fine-tuned LLM to predict structured B-spline prior parameters. LASP is trained through a two-stage scheme that combines local geometric regularities with long-range contextual dependencies, producing priors that are both interpretable and semantically coherent. This approach furnishes inductive signals that direct the conventional B-spline fitting process toward solutions that more accurately encapsulate the intended design objectives and demonstrate heightened semantic coherence. Compared to traditional machine learning schemes, the experiments demonstrate that language-driven reasoning can serve as a powerful inductive bias for geometric solving, establishing a new paradigm of language-guided geometric optimization in modern CAD systems. --- ### Spectral Adapters for Segment Anything Model-based Segmentation of Colorectal Liver Metastases in Computed Tomography Authors: Ramtin Mojtahedi, Mohammad Hamghalam, Jacob J. Peoples, Natalie Gangai, Mithat Gonen, Yun Shin Chun, HyunSeon Christine Kang, Richard K. G. Do, Amber L. Simpson Published: 2026-09-10 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.11703v1 Accurate segmentation of colorectal liver metastases (CRLM) in contrast-enhanced computed tomography (CT) is important for response assessment, surgical planning, and follow-up. We propose two parameter-efficient spectral adapters for the Segment Anything Model (SAM): the Directional Spectral Adapter (DiSECT) and Spectral Instance-Guided Adapter (SiGA). DiSECT uses singular value decomposition of frozen weights to constrain residual updates to leading spectral directions, while SiGA adds global and input-conditioned gating through a multilayer perceptron. We evaluate these methods on 446 contrast-enhanced CT volumes (355 training, 91 testing) and compare them with LoRA, QLoRA, convolutional adapters (CAD), and a 3D nnU-Net baseline. Experiments consider single-point, three-point, bounding-box, and no-prompt regimes. SiGA achieves the best single-point performance with a Dice score of 0.77, IoU of 0.69, and HD95 of 35.39 mm. Under no-prompt inference, SiGA reaches 0.76 Dice, 0.68 IoU, and 46.76 mm HD95, comparable to the nnU-Net baseline (0.758 Dice). DiSECT uses only 0.14 million trainable parameters. These results show that spectral adapters can efficiently adapt SAM for CRLM segmentation while retaining strong accuracy with limited trainable parameters. --- ### Negative Self-Distillation: Learning to Reason by Avoiding Flaws Authors: Rongcan Pei, Zhepei Wei, Shuyao Xu, Xinyu Zhu, Wei-Lin Chen, Yu Meng Published: 2026-09-10 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.11699v1 On-Policy Self-Distillation (OPSD) has emerged as a popular paradigm for large language model (LLM) self-improvement, allowing models to act as their own teachers by leveraging privileged information such as ground-truth solutions. However, recent findings indicate that OPSD can severely degrade the performance of LLMs on complex reasoning tasks: By forcing the student to imitate an artificially confident reasoning trace conditioned on privileged information, OPSD inadvertently suppresses expressions of uncertainty and penalizes the exploratory, self-corrective behaviors required to solve challenging problems. To address this, we introduce Negative Self-Distillation (NSD), a new framework that optimizes LLMs by diverging from flawed reasoning rather than imitating privileged solutions. Instead of relying on ground-truth answers or external supervision, NSD uses the model itself to generate a question-specific negative condition (eg, acting as a ``careless reasoner'') and pushes the student's distribution away from this self-generated negative teacher. Naively applying unlearning objectives to achieve this divergence is problematic, as flawed reasoning tokens are confounded with basic linguistic tokens; indiscriminately penalizing both risks catastrophically degrading the model's foundational language capabilities. We resolve this by designing a dynamic gating mechanism that automatically identifies and isolates reasoning-critical tokens, ensuring gradient updates target only behavioral flaws while preserving the model's linguistic priors. Empirically, NSD consistently outperforms OPSD and other label-free, self-bootstrapping reinforcement learning (RL) baselines. --- ### ActSafeGuard: Differentiable and Training-Aligned Constraint Enforcement for Flow-Matching Policies Authors: Jianming Ma, Rongjun Jin, Xiaxi Si, Yang Zhang, Yiheng Li, Yue Gao Published: 2026-09-10 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.11697v2 Vision-Language-Action (VLA) and World-Action Models (WAMs) have demonstrated strong capabilities in general-purpose robotic manipulation, yet their generated actions may violate hard physical constraints and therefore be unsafe or infeasible for deployment. Existing safety approaches either optimize statistical safety objectives without deterministic per-step guarantees or correct unsafe actions only during inference, creating a mismatch between policy training and execution. We introduce ActSafeGuard, a differentiable and training-aligned safeguard layer for flow-matching based policies. ActSafeGuard integrates hard action feasibility into policy learning, not merely treating safety as an inference-time external component. Through an analytical ray-scaling operator design, ActSafeGuard enables boundary-aware gradients to guide the model to naturally learn constrained manifolds. Extensive experiments on multiple standard foundation backbones ($π_{0.5}$ and Fast-WAM) across various tasks demonstrate that ActSafeGuard consistently achieves a $100\%$ step safety rate while fully preserving or even boosting task success rates, providing a scalable and minimally invasive solution for safe embodied AI deployment. --- ### ActSafeGuard: Differentiable and Training-Aligned Constraint Enforcement for Flow-Matching Policies Authors: Jianming Ma, Rongjun Jin, Xiaxi Si, Yang Zhang, Yiheng Li, Yue Gao Published: 2026-09-10 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.11697v1 Vision-Language-Action (VLA) and World-Action Models (WAMs) have demonstrated strong capabilities in general-purpose robotic manipulation, yet their generated actions may violate hard physical constraints and therefore be unsafe or infeasible for deployment. Existing safety approaches either optimize statistical safety objectives without deterministic per-step guarantees or correct unsafe actions only during inference, creating a mismatch between policy training and execution. We introduce ActSafeGuard, a differentiable and training-aligned safeguard layer for flow-matching based policies. ActSafeGuard integrates hard action feasibility into policy learning, not merely treating safety as an inference-time external component. Through an analytical ray-scaling operator design, ActSafeGuard enables boundary-aware gradients to guide the model to naturally learn constrained manifolds. Extensive experiments on multiple standard foundation backbones ($π_{0.5}$ and Fast-WAM) across various tasks demonstrate that ActSafeGuard consistently achieves a $100\%$ step safety rate while fully preserving or even boosting task success rates, providing a scalable and minimally invasive solution for safe embodied AI deployment. --- ### Stress-Testing Dynamical and Generative Downscaling Using Subseasonal Extreme Precipitation Forecasts Authors: Mauricio Lima, Marika Koukoula, Romain Pilon, Monika Feldmann, Erwan Koch, Daniela I. V. Domeisen, Tom Beucler Published: 2026-09-10 Categories: physics.comp-ph, physics.geo-ph, stat.ML arXiv: https://arxiv.org/abs/2609.11696v1 Coarse spatial resolution limits the ability of subseasonal prediction models to resolve extreme precipitation. Downscaling with either dynamical or deep generative models can overcome this issue, but the comparative performance of these models for extremes across different atmospheric regimes remains poorly understood. In this work, we evaluate the Weather Research and Forecasting (WRF) model against a diffusion-based generative model by downscaling two physically distinct, extreme precipitation events up to lead times of 3 weeks. For a fair comparison with WRF, which can downscale boundary conditions from different driving models without model-specific training, the diffusion model is trained in an unpaired fashion. Both approaches improve upon the raw European Centre for Medium-Range Weather Forecasts forecasts, in comparison to fused rain gauge-radar observations in Switzerland (CombiPrecip), but exhibit regime-dependent strengths. WRF achieves the highest probabilistic skill for a multicell, non-stationary event. Conversely, the diffusion model is more consistent across different performance metrics for the two events, outperforming WRF in a more stationary supercell event. These results demonstrate that explicit dynamical modeling can add value for specific precipitation events for subseasonal lead times, and that generative downscaling adds value more broadly in different situations. --- ### Geospatial Foundation Models Capture Health-Relevant Dimensions of Place Beyond Conventional Social Risk Indices Authors: Nathaniel Hendrix, Carl Y. Zhang, Chris Heitzig, Andrew Bazemore, David H. Rehkopf Published: 2026-09-10 Categories: stat.AP, cs.LG arXiv: https://arxiv.org/abs/2609.11689v2 Area-based social risk indices summarize residents' socioeconomic conditions but incompletely capture physical features of place that may affect health. We evaluated whether numerical representations of physical place produced by four geospatial foundation model families from 2022 satellite data explained residual variance in tract-level associations between the Area Deprivation Index, Social Deprivation Index, and Social Vulnerability Index with health outcomes. We used LightGBM to predict variables from the American Community Survey and 40 chronic disease and health-behavior outcomes from CDC PLACES across 82,646 census tracts in the contiguous United States, evaluating performance across 10 held-out states. Among survey variables, models were moderately predictive of some variables including housing type (R-squared up to 0.54) but weak for disability, unemployment, and income disparity. For health outcomes, models explained up to 54% of variance left unexplained by social risk indices, with the largest gains for annual checkups, arthritis, and high blood pressure. Mean total variance explained by geospatial foundation models across the 40 health-related outcomes increased from 0.31 in the smallest tract-size decile to 0.39 in the largest. Geospatial foundation models capture health-relevant features of place not represented by conventional social risk indices and may usefully augment them in epidemiological analyses. --- ### Geospatial Foundation Models Capture Health-Relevant Dimensions of Place Beyond Conventional Social Risk Indices Authors: Nathaniel Hendrix, Carl Y. Zhang, Chris Heitzig, Andrew Bazemore, David H. Rehkopf Published: 2026-09-10 Categories: stat.AP, cs.LG arXiv: https://arxiv.org/abs/2609.11689v1 Area-based social risk indices summarize residents' socioeconomic conditions but incompletely capture physical features of place that may affect health. We evaluated whether numerical representations of physical place produced by four geospatial foundation model families from 2022 satellite data explained residual variance in tract-level associations between the Area Deprivation Index, Social Deprivation Index, and Social Vulnerability Index with health outcomes. We used LightGBM to predict variables from the American Community Survey and 40 chronic disease and health-behavior outcomes from CDC PLACES across 82,646 census tracts in the contiguous United States, evaluating performance across 10 held-out states. Among survey variables, models were moderately predictive of some variables including housing type (R-squared up to 0.54) but weak for disability, unemployment, and income disparity. For health outcomes, models explained up to 54% of variance left unexplained by social risk indices, with the largest gains for annual checkups, arthritis, and high blood pressure. Mean total variance explained by geospatial foundation models across the 40 health-related outcomes increased from 0.31 in the smallest tract-size decile to 0.39 in the largest. Geospatial foundation models capture health-relevant features of place not represented by conventional social risk indices and may usefully augment them in epidemiological analyses. --- ### Structured Transforms for Low-Overhead Quantization of Language Models Authors: Daria Cherniuk, Alexander Rudikov, Boris Kashin, Ivan Oseledets Published: 2026-09-10 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.11687v1 We revisit Kashin-decomposition-based weight quantization for large language models and propose an improved algorithm with stronger convergence properties and structured, efficient orthogonal transforms. The method retains the core factorization of each weight into two components -- one with bounded infinity norm and the other with bounded infinity norm after an orthogonal transformation -- but replaces the dense random orthogonal matrix with a sign-randomized Discrete Cosine Transform (DCT), reducing the per-iteration cost from $\mathcal{O}(N^2)$ to $\mathcal{O}(N \log N)$. The proposed greedy algorithm with alternating updates guarantees the four-peak distribution required for stable 2-bit clustering of each factor and admits closed-form initialization of cluster centers, removing the multi-restart k-means bottleneck of prior work. Composed with OPTQ-style sequential error compensation and QuIP-style incoherence preprocessing, the resulting JAX pipeline is competitive with OPTQ, QuIP, QuIP-RG and a fine-tuning- and vector-quantization-free variant of QuIP# at 4-bit per channel on OPT, Llama-2 and Pythia, with favorable wall-clock scaling. The bounded-$\ell_\infty$ factorization is also notably robust: on stress configurations where QuIP variants diverge to four-digit perplexity (Pythia-6.9B) or abort with NaNs in LDL back-substitution (Mistral-7B), Kashin-DCT remains numerically stable and stays close to FP16 baseline. At inference time, each weight decomposes into two 2-bit factor codes per channel that are structurally suited to native-2-bit hardware. --- ### COBRA-Skills: Contextual Bandit-Guided Evolution for Agent Skill Optimization Authors: Pingchen Lu, Xiangyi Wang, Xiang Li, Jie Mao, Zikun Qu, Junfeng Luo, Yao Shu, Bryan Kian Hsiang Low, Zhongxiang Dai Published: 2026-09-10 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.11682v1 Large language model (LLM) agents can benefit from reusable skills distilled from prior task experience, yet existing skill optimization methods often rely on costly execution-based evaluation and substantial task data. We introduce \textbf{COBRA-Skills}, an efficient framework that formulates skill optimization as budgeted sequential optimization over a dynamically evolving candidate space. COBRA-Skills couples contextual-bandit-guided prioritization with evidence-grounded skill evolution, selectively allocating evaluations to promising or informative candidates while continually refining the skill population from execution feedback. Across six heterogeneous agent benchmarks and three target models, COBRA-Skills consistently achieves the strongest average performance among compared methods, while reducing optimization cost by 55--58\% relative to SkillOpt and using only 50 unique optimization examples per benchmark. Further analyses show that COBRA-Skills remains robust to changes in the agent harness and performs effectively when the target model itself is used for skill generation and refinement. ---