The paper proposes an opinion‑dynamics approach to form client coalitions in federated learning for heterogeneous IoT systems, specifically targeting smart‑meter water‑consumption forecasting. By modeling coalition formation as a Hegselmann‑Krause bounded‑confidence process on local model weights, the method creates stable coalition structures within ten iterations without extra client computation or communication. Experiments on real smart‑meter data show the HK‑based coalitions reduce mean absolute error by up to 54% compared to FedAvg and achieve the highest global accuracy (83‑85%).
The paper investigates how adversarial patches affect Vision‑Language‑Action (VLA) policies, revealing that such patches can cause immediate action corruption and persistent state effects that linger after the patch is removed. A state‑restoration protocol is introduced to isolate these effects by removing the patch at action‑chunk boundaries and measuring recoverability within the remaining step budget. Experiments on OpenVLA-OFT with EDPA attacks show that only 36.2% of episodes recover after five chunks, whereas controls recover at 89.9% and 87.0%. A recovery adapter trained on attack‑induced states improves recovery from 7.7% to 47.4% at one‑chunk latency, but its effectiveness drops sharply with delayed intervention, underscoring the importance of timely recovery.
The paper presents modular deep learning augmentations for next‑day wildfire spread prediction, including wind‑ and slope‑conditioned attention biases, physics‑feature retrieval‑augmented output correction, and fire‑conditioned dual‑stream gating. These modules are evaluated on five backbone models using the Next Day Wildfire Spread benchmark, with staged ablations, directional audits, retrieval perturbations, calibration measures, and computational comparisons. The best augmented SwinUNETR model achieves an F1 score of 0.4216 and an AUC‑PR of 0.3673, while a mixed ensemble reaches 0.4292 and 0.3790, demonstrating that predictive performance, operational trustworthiness, and computational practicality can be simultaneously improved.
By Miguel Esparza, Aydin Ayanzadeh Ahmad Mousavi, Ali Mostafavi
QuanText is a training‑free, large‑language‑model‑agnostic mechanism for releasing textual datasets that protects dataset‑level secrets such as the proportion of records with a particular diagnosis or gender. It perturbs both the secret distribution and correlated attribute distributions by selecting candidate release distributions close to the private empirical distribution and rewriting each text sample to match the chosen distribution using attribute‑related snippets. The method is inspired by the Statistic Maximal Leakage framework and, under idealized conditions, satisfies an SML guarantee, while empirical evaluations show a superior privacy‑utility trade‑off compared to existing data generation baselines.
By Shuaiqi Wang, Zinan Lin, Giulia Fanti
The paper audits vision‑language models (CLIP) for gender bias using 1,500 artworks from the Metropolitan Museum of Art, focusing on zero‑shot logit differences for prompts like "masterpiece," "quality," and "influence." Unadjusted results show no significant gender effect and statistical equivalence across models, but high residual variance suggests that global zero‑shot metrics are largely noise‑dominated and may miss fine‑grained biases. The study underscores the need for multivariate confound control, equivalence testing, and provenance auditing when evaluating AI fairness in cultural heritage data.
By Manpreet Singh, Rhythm Bhatia, Rahul Joshi
The paper introduces PanicCognitivePath (PCP), a model that predicts the timing of panic emotional arousal by integrating appraisal emotion theory into a Belief‑Desire‑Emotion‑Intention (BDEI) framework. PCP uses a Psychological Safety Distance (PSD) model to fuse physical, social, cognitive, and informational signals into a unified risk metric, and confines large language models to a single parameter‑estimation step to reduce hallucinations. Experiments on Hurricane Sandy data show PCP improves individual prediction accuracy by 10.68% and reduces peak count error to 7.07%.
By Mengzhu Liu, Long Qin, Chuan Ai, Zhengqiu Zhu, Hongru Liang, Fangfang Li, Chen Gao, Yong Li, Xin Lu, Quanjun Yin
The paper introduces ASLEval, a framework for measuring privacy exposure displacement in large language model (LLM) agent sessions. It highlights that traditional local proxies—such as inspecting a single action or final response—often miss unauthorized data leaks elsewhere in a multi-step session. ASLEval pre-registers hidden target sets, tracks all declared visible exits, and preserves internal traces for diagnosis, revealing that a single outlet view can overlook nearly 47% of exposure and that internal evidence typically precedes visible leaks. The study underscores the need for benchmarks that define complete visible boundaries, ground claims in pre-specified targets, and report privacy alongside task utility.
By Guosen Wu, Huizhen Huang, Guoxiong Long, Tao Huang, Chen Hou
METALICA is a new method that combines Metadynamics with Replica Exchange on pretrained diffusion models to improve sampling of rare conformational states in proteins. It builds a bias potential along a chosen Collective Variable, pushes new samples away from previously visited ones, and reweights them to recover the unbiased distribution. By running one replica per diffusion level and allowing inter‑replica communication, METALICA can generate long Markov chains that uncover rare events, as demonstrated on a bimodal test system and on protein unfolding where it reveals a second free‑energy minimum that other methods miss.
By Alireza Omidi, Jiajun He, J\"org Gsponer, Saifuddin Syed
Safety-Flag is a unified benchmark that consolidates seven popular safety datasets into a single balanced flag/do‑not‑flag protocol, providing item‑level decisions and confidence scores for multiple large language models and dedicated guards. The benchmark evaluates moderator reliability across three dimensions—error direction, probability calibration, and confidence‑based error ranking—revealing that aggregate accuracy masks significant differences, such as one model flagging 85% of benign content while another misses 54% of harmful content. The study shows that general‑purpose models are overconfident, but temperature tuning can substantially improve calibration, and confidence‑based abstention can reduce selective risk, though performance varies with how well confidence ranks errors.
By Yibo Hu
The paper addresses the challenge of creating machine learning learners that can guarantee provably correct predictions in difficult test-time scenarios, such as adversarial attacks and natural distribution shifts. It introduces a reliable learner with optimal theoretical guarantees for these settings and discusses practical implementations. The authors demonstrate strong performance on examples like linear separators under log-concave distributions and smooth boundary classifiers under smooth probability distributions.
By Maria-Florina Balcan, Steve Hanneke, Rattana Pukdee, Dravyansh Sharma
The paper examines how large language models (LLMs) alter the expression of Dark Triad traits—Machiavellianism, narcissism, and psychopathy—when prompted to fake good or fake bad. Across seven state‑of‑the‑art models and two real‑world contexts (employment selection and forensic evaluation), most models lowered trait scores under fake‑good conditions and raised them under fake‑bad conditions, with varying consistency across traits and models. The study also finds that explicit fake‑bad instructions produce stronger distortions than contextual framing alone, underscoring the influence of motivational and situational context on personality‑related outputs.
By Victoria Popa, Guglielmo Cola, Caterina Senette, Maurizio Tesconi
The paper introduces STaR-CS, a method that generates multi‑party dialogues by modeling facilitator style and extracting structured skeletons, thereby easing data scarcity for cognitive stimulation therapy (CST) in low‑resource languages. Building on this corpus, the Reflective Cognitive Alignment (RCA) framework treats stimulation interactions as a sequential decision process, combining Protocol‑Constrained Chain‑of‑Cognition (PC‑CoC) for structured reasoning with Inference‑Time Value Alignment (IVA) to select responses that balance safety and engagement. Experiments with six large language models and two judges demonstrate that RCA improves protocol adherence, safety, and group facilitation compared to standard prompting.
By Jiyue Jiang, Ziyi Li, He Hu, Sheng Wang, Yuhan Chen, Yanyu Chen, Jingqi Zhou, Pengan Chen, Fei Ma, Irwin King, Yu Li, Chuan Wu
SAGE is a governed multi‑stage LLM pipeline that transforms enterprise guideline documents—containing narrative text, tables, and images—into structured artifacts. It uses a shared versioned rule store, schema‑validated contracts, and provenance tracking to validate, score, and reconcile extracted rules, automatically approving high‑confidence outputs while flagging uncertain items for human review. In a test on 120 documents, SAGE reduced processing time from days to 20–100 minutes and achieved a 96% success rate with only 3.2% hallucination.
By Mohammadreza Sediqin, Shivali Dalmia, Sumukha Thoppanahalli, Srinivasa Karthikeya Reddy Kovvuri, Abhishek Mukherji
The paper investigates whether large language models decide to gather safety-relevant evidence before acting. Using the SAFE benchmark, the authors evaluate models such as GPT‑5.5, o3, Claude Opus, and Claude Sonnet, finding distinct evidence‑acquisition strategies that vary with retrieval cost, severity, and presentation. Across models, expected‑value reasoning dominates Stage 1 rationales, and evidence framing can alter decisions near the inspection threshold while probability is often cited despite limited influence.
By Omer Tafveez
GuardEn is an executable safeguard framework that transforms complex safety policies into atomic propositions via Safety‑Rule Compilation and models their composition as executable code. At inference, Scene‑Grounded Execution uses visual scene graphs to instantiate these propositions, allowing rule‑grounded and interpretable safety reasoning. Experiments on SafetyVisionBench show GuardEn improves F1 by an average of 9.8 points over the strongest baseline.
By Jisoo Kim (Sungkyunkwan University), TaeYoon Kwack (Sungkyunkwan University), Jinwoo Jang (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
The paper introduces Market Signal Injection (MSI), an attack that alters how market data is formatted or described—without changing its numerical values—to influence large language model (LLM) pricing agents. Experiments on nine open‑weight and three proprietary models in simulated duopoly and triopoly markets show that sentiment‑based formatting changes cause significant shifts in firm behavior, profits, and consumer surplus. The study also demonstrates that model susceptibility varies across families, that larger models are not always more robust, and that techniques such as input canonicalization and decision boundary anchoring can partially mitigate these attacks.
By Dohun Lee, Hyunwoo Park
The paper introduces DISCERN, a two-tier protocol for certifying that updates to production models do not increase risk. It first uses unlabeled data to detect benign updates based on disagreement rates, then selectively labels only disagreements through an anytime-valid confidence sequence. The method achieves finite-sample validity with label-complexity bounds of order ρ²/ε², demonstrating significant label savings and strong empirical performance across 14,000+ audit streams.
By Vishnu Bindu Balachandran
EDCT-Bench is a benchmark that evaluates the faithfulness of Vision‑Language Models (VLMs) by using Explanation‑Driven Counterfactual Testing (EDCT). EDCT extracts visual concepts from a model’s natural language explanation, applies minimal verified edits to those concepts, and checks whether the model’s answer and explanation remain consistent with the edited image. The benchmark covers three domains—knowledge‑intensive VQA, safety‑critical driving, and 3D spatial reasoning—and reveals significant faithfulness gaps in current VLMs, while also showing that EDCT‑generated counterfactuals can improve training.
By Sihao Ding, Santosh Vasa, Aditi Ramadwar, Thomas Monninger
The paper examines three recent studies that highlight distinct reliability issues in large language models (LLMs). Each study points to a missing capability—whether a tool‑reliability representation, safe generation behavior, or a consistent‑reasoning function—yet all converge on the need for calibrated abstention. The authors argue that current benchmarks fail to reward abstention, preventing the development of this function, and propose four evaluation changes to address the gap.
By Srijith Ravikumar
CALOS is a runtime safety layer for quadrotor control that enforces attitude constraints without altering the underlying deep reinforcement learning algorithm. It formulates tilt-angle inequalities and a Lyapunov descent condition into a single quadratic program, solved exactly via active-set enumeration over a three-dimensional torque space. In NVIDIA Isaac Lab trajectory-tracking tasks, CALOS reduces lateral tracking error by 55‑60% compared to an unconstrained Proximal Policy Optimization baseline and eliminates attitude-constraint violations during training.
By Fabrizio Cesareo, Sebastiano Mengozzi, Nicola Mimmo, Andrea Acquaviva