Authors: Jieyi Bi, Zhiguang Cao, Jianan Zhou, Wen Song, Yaoxin Wu, Jie Zhang, Yining Ma, Cathy Wu
Abstract: Neural solvers have achieved impressive progress in addressing simple routing problems, particularly excelling in computational efficiency. However, their advantages under complex constraints remain nascent, for which current constraint-handling schemes via feasibility masking or implicit feasibility awareness can be inefficient or inapplicable for hard constraints. In this paper, we present Construct-and-Refine (CaR), the first general and efficient constraint-handling framework for neural routing solvers based on explicit learning-based feasibility refinement. Unlike prior construction-search hybrids that target reducing optimality gaps through heavy improvements yet still struggle with hard constraints, CaR achieves efficient constraint handling by designing a joint training framework that guides the construction module to generate diverse and high-quality solutions well-suited for a lightweight improvement process, e.g., 10 steps versus 5k steps in prior work. Moreover, CaR presents the first use of construction-improvement-shared representation, enabling potential knowledge sharing across paradigms by unifying the encoder, especially in more complex constrained scenarios. We evaluate CaR on typical hard routing constraints to showcase its broader applicability. Results demonstrate that CaR achieves superior feasibility, solution quality, and efficiency compared to both classical and neural state-of-the-art solvers.
Authors: Cameron Cagan, Pedram Fard, Jiazi Tian, Jingya Cheng, Shawn N. Murphy, Hossein Estiri
Abstract: Autonomous agentic workflows that iteratively refine their own behavior hold considerable promise, yet their failure modes remain poorly characterized. We investigate optimization instability, a phenomenon in which continued autonomous improvement paradoxically degrades classifier performance, using Pythia, an open-source framework for automated prompt optimization. Evaluating three clinical symptoms with varying prevalence (shortness of breath at 23%, chest pain at 12%, and Long COVID brain fog at 3%), we observed that validation sensitivity oscillated between 1.0 and 0.0 across iterations, with severity inversely proportional to class prevalence. At 3% prevalence, the system achieved 95% accuracy while detecting zero positive cases, a failure mode obscured by standard evaluation metrics. We evaluated two interventions: a guiding agent that actively redirected optimization, amplifying overfitting rather than correcting it, and a selector agent that retrospectively identified the best-performing iteration successfully prevented catastrophic failure. With selector agent oversight, the system outperformed expert-curated lexicons on brain fog detection by 331% (F1) and chest pain by 7%, despite requiring only a single natural language term as input. These findings characterize a critical failure mode of autonomous AI systems and demonstrate that retrospective selection outperforms active intervention for stabilization in low-prevalence classification tasks.
Authors: Hang Li, Kaiqi Yang, Xianxuan Long, Fedor Filippov, Yucheng Chu, Yasemin Copur-Gencturk, Peng He, Cory Miller, Namsoo Shin, Joseph Krajcik, Hui Liu, Jiliang Tang
Abstract: The rapid rise of large language models (LLMs) is reshaping the landscape of automatic assessment in education. While these systems demonstrate substantial advantages in adaptability to diverse question types and flexibility in output formats, they also introduce new challenges related to output uncertainty, stemming from the inherently probabilistic nature of LLMs. Output uncertainty is an inescapable challenge in automatic assessment, as assessment results often play a critical role in informing subsequent pedagogical actions, such as providing feedback to students or guiding instructional decisions. Unreliable or poorly calibrated uncertainty estimates can lead to unstable downstream interventions, potentially disrupting students' learning processes and resulting in unintended negative consequences. To systematically understand this challenge and inform future research, we benchmark a broad range of uncertainty quantification methods in the context of LLM-based automatic assessment. Although the effectiveness of these methods has been demonstrated in many tasks across other domains, their applicability and reliability in educational settings, particularly for automatic grading, remain underexplored. Through comprehensive analyses of uncertainty behaviors across multiple assessment datasets, LLM families, and generation control settings, we characterize the uncertainty patterns exhibited by LLMs in grading scenarios. Based on these findings, we evaluate the strengths and limitations of different uncertainty metrics and analyze the influence of key factors, including model families, assessment tasks, and decoding strategies, on uncertainty estimates. Our study provides actionable insights into the characteristics of uncertainty in LLM-based automatic assessment and lays the groundwork for developing more reliable and effective uncertainty-aware grading systems in the future.
Authors: Amir Hosseinian, MohammadReza Zare Shahneh, Umer Mansoor, Gilbert Szeto, Kirill Karlin, Nima Aghaeepour
Abstract: Background: Large language models have demonstrated strong performance on general medical examinations, but subspecialty clinical reasoning remains challenging due to rapidly evolving guidelines and nuanced evidence hierarchies. Methods: We evaluated January Mirror, an evidence-grounded clinical reasoning system, against frontier LLMs (GPT-5, GPT-5.2, Gemini-3-Pro) on a 120-question endocrinology board-style examination. Mirror integrates a curated endocrinology and cardiometabolic evidence corpus with a structured reasoning architecture to generate evidence-linked outputs. Mirror operated under a closed-evidence constraint without external retrieval. Comparator LLMs had real-time web access to guidelines and primary literature. Results: Mirror achieved 87.5% accuracy (105/120; 95% CI: 80.4-92.3%), exceeding a human reference of 62.3% and frontier LLMs including GPT-5.2 (74.6%), GPT-5 (74.0%), and Gemini-3-Pro (69.8%). On the 30 most difficult questions (human accuracy less than 50%), Mirror achieved 76.7% accuracy. Top-2 accuracy was 92.5% for Mirror versus 85.25% for GPT-5.2. Conclusions: Mirror provided evidence traceability: 74.2% of outputs cited at least one guideline-tier source, with 100% citation accuracy on manual verification. Curated evidence with explicit provenance can outperform unconstrained web retrieval for subspecialty clinical reasoning and supports auditability for clinical deployment.
Authors: Martin Klissarov, Jonathan Cook, Diego Antognini, Hao Sun, Jingling Li, Natasha Jaques, Claudiu Musat, Edward Grefenstette
Abstract: Adapting one's thought process based on corrective feedback is an essential ability in human learning, particularly in collaborative settings. In contrast, the current large language model training paradigm relies heavily on modeling vast, static corpora. While effective for knowledge acquisition, it overlooks the interactive feedback loops essential for models to adapt dynamically to their context. In this work, we propose a framework that treats this interactive in-context learning ability not as an emergent property, but as a distinct, trainable skill. We introduce a scalable method that transforms single-turn verifiable tasks into multi-turn didactic interactions driven by information asymmetry. We first show that current flagship models struggle to integrate corrective feedback on hard reasoning tasks. We then demonstrate that models trained with our approach dramatically improve the ability to interactively learn from language feedback. More specifically, the multi-turn performance of a smaller model nearly reaches that of a model an order of magnitude larger. We also observe robust out-of-distribution generalization: interactive training on math problems transfers to diverse domains like coding, puzzles and maze navigation. Our qualitative analysis suggests that this improvement is due to an enhanced in-context plasticity. Finally, we show that this paradigm offers a unified path to self-improvement. By training the model to predict the teacher's critiques, effectively modeling the feedback environment, we convert this external signal into an internal capability, allowing the model to self-correct even without a teacher.
Authors: Thinh Hung Truong, Jey Han Lau, Jianzhong Qi
Abstract: Large Language Models (LLMs) are increasingly deployed in applications that interact with the physical world, such as navigation, robotics, or mapping, making robust geospatial reasoning a critical capability. Despite that, LLMs' ability to reason about GPS coordinates and real-world geography remains underexplored. We introduce GPSBench, a dataset of 57,800 samples across 17 tasks for evaluating geospatial reasoning in LLMs, spanning geometric coordinate operations (e.g., distance and bearing computation) and reasoning that integrates coordinates with world knowledge. Focusing on intrinsic model capabilities rather than tool use, we evaluate 14 state-of-the-art LLMs and find that GPS reasoning remains challenging, with substantial variation across tasks: models are generally more reliable at real-world geographic reasoning than at geometric computations. Geographic knowledge degrades hierarchically, with strong country-level performance but weak city-level localization, while robustness to coordinate noise suggests genuine coordinate understanding rather than memorization. We further show that GPS-coordinate augmentation can improve in downstream geospatial tasks, and that finetuning induces trade-offs between gains in geometric computation and degradation in world knowledge. Our dataset and reproducible code are available at https://github.com/joey234/gpsbench
Authors: Kaiqu Liang, Julia Kruk, Shengyi Qian, Xianjun Yang, Shengjie Bi, Yuanshun Yao, Shaoliang Nie, Mingyang Zhang, Lijuan Liu, Jaime Fern\'andez Fisac, Shuyan Zhou, Saghar Hosseini
Abstract: Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit preference models on interaction history or encoding user profiles in external memory. However, these approaches struggle with new users and with preferences that change over time. We introduce Personalized Agents from Human Feedback (PAHF), a framework for continual personalization in which agents learn online from live interaction using explicit per-user memory. PAHF operationalizes a three-step loop: (1) seeking pre-action clarification to resolve ambiguity, (2) grounding actions in preferences retrieved from memory, and (3) integrating post-action feedback to update memory when preferences drift. To evaluate this capability, we develop a four-phase protocol and two benchmarks in embodied manipulation and online shopping. These benchmarks quantify an agent's ability to learn initial preferences from scratch and subsequently adapt to persona shifts. Our theoretical analysis and empirical results show that integrating explicit memory with dual feedback channels is critical: PAHF learns substantially faster and consistently outperforms both no-memory and single-channel baselines, reducing initial personalization error and enabling rapid adaptation to preference shifts.
Authors: Sushant Mehta, Logan Ritchie, Suhaas Garre, Nick Heiner, Edwin Chen
Abstract: We show that training AI agents on high-fidelity reinforcement learning environments produces capabilities that generalize beyond the training distribution. We introduce \corecraft{}, the first environment in \textsc{EnterpriseGym}, Surge AI's suite of agentic RL environments. \corecraft{} is a fully operational enterprise simulation of a customer support organization, comprising over 2,500 entities across 14 entity types with 23 unique tools, designed to measure whether AI agents can perform the multi-step, domain-specific work that real jobs demand. Frontier models such as GPT-5.2 and Claude Opus 4.6 solve fewer than 30\% of tasks when all expert-authored rubric criteria must be satisfied. Using this environment, we train GLM~4.6 with Group Relative Policy Optimization (GRPO) and adaptive clipping. After a single epoch of training, the model improves from 25.37\% to 36.76\% task pass rate on held-out evaluation tasks. More importantly, these gains transfer to out-of-distribution benchmarks: +4.5\% on BFCL Parallel, +7.4\% on $\tau^2$-Bench Retail, and +6.8\% on Toolathlon (Pass@1). We believe three environment properties are consistent with the observed transfer: task-centric world building that optimizes for diverse, challenging tasks; expert-authored rubrics enabling reliable reward computation; and enterprise workflows that reflect realistic professional patterns. Our results suggest that environment quality, diversity, and realism are key factors enabling generalizable agent capabilities.
Authors: Hiroaki Yamanaka, Daisuke Miyashita, Takashi Toi, Asuka Maki, Taiga Ikeda, Jun Deguchi
Abstract: Driven by our mission of "uplifting the world with memory," this paper explores the design concept of "memory" that is essential for achieving artificial superintelligence (ASI). Rather than proposing novel methods, we focus on several alternative approaches whose potential benefits are widely imaginable, yet have remained largely unexplored. The currently dominant paradigm, which can be termed "extract then store," involves extracting information judged to be useful from experiences and saving only the extracted content. However, this approach inherently risks the loss of information, as some valuable knowledge particularly for different tasks may be discarded in the extraction process. In contrast, we emphasize the "store then on-demand extract" approach, which seeks to retain raw experiences and flexibly apply them to various tasks as needed, thus avoiding such information loss. In addition, we highlight two further approaches: discovering deeper insights from large collections of probabilistic experiences, and improving experience collection efficiency by sharing stored experiences. While these approaches seem intuitively effective, our simple experiments demonstrate that this is indeed the case. Finally, we discuss major challenges that have limited investigation into these promising directions and propose research topics to address them.
Authors: Yun-Shiuan Chuang, Chaitanya Kulkarni, Alec Chiu, Avinash Thangali, Zijie Pan, Shivani Shekhar, Yirou Ge, Yixi Li, Uma Kona, Linsey Pang, Prakhar Mehrotra
Abstract: Interactive large language model (LLM) agents operating via multi-turn dialogue and multi-step tool calling are increasingly used in production. Benchmarks for these agents must both reliably compare models and yield on-policy training data. Prior agentic benchmarks (e.g., tau-bench, tau2-bench, AppWorld) rely on fully deterministic backends, which are costly to build and iterate. We propose Proxy State-Based Evaluation, an LLM-driven simulation framework that preserves final state-based evaluation without a deterministic database. Specifically, a scenario specifies the user goal, user/system facts, expected final state, and expected agent behavior, and an LLM state tracker infers a structured proxy state from the full interaction trace. LLM judges then verify goal completion and detect tool/user hallucinations against scenario constraints. Empirically, our benchmark produces stable, model-differentiating rankings across families and inference-time reasoning efforts, and its on-/off-policy rollouts provide supervision that transfers to unseen scenarios. Careful scenario specification yields near-zero simulator hallucination rates as supported by ablation studies. The framework also supports sensitivity analyses over user personas. Human-LLM judge agreement exceeds 90%, indicating reliable automated evaluation. Overall, proxy state-based evaluation offers a practical, scalable alternative to deterministic agentic benchmarks for industrial LLM agents.
Authors: Marissa A. Weis, Maciej Wo{\l}czyk, Rajai Nasser, Rif A. Saurous, Blaise Ag\"uera y Arcas, Jo\~ao Sacramento, Alexander Meulemans
Abstract: Achieving cooperation among self-interested agents remains a fundamental challenge in multi-agent reinforcement learning. Recent work showed that mutual cooperation can be induced between "learning-aware" agents that account for and shape the learning dynamics of their co-players. However, existing approaches typically rely on hardcoded, often inconsistent, assumptions about co-player learning rules or enforce a strict separation between "naive learners" updating on fast timescales and "meta-learners" observing these updates. Here, we demonstrate that the in-context learning capabilities of sequence models allow for co-player learning awareness without requiring hardcoded assumptions or explicit timescale separation. We show that training sequence model agents against a diverse distribution of co-players naturally induces in-context best-response strategies, effectively functioning as learning algorithms on the fast intra-episode timescale. We find that the cooperative mechanism identified in prior work-where vulnerability to extortion drives mutual shaping-emerges naturally in this setting: in-context adaptation renders agents vulnerable to extortion, and the resulting mutual pressure to shape the opponent's in-context learning dynamics resolves into the learning of cooperative behavior. Our results suggest that standard decentralized reinforcement learning on sequence models combined with co-player diversity provides a scalable path to learning cooperative behaviors.
Authors: Philipp Schoenegger, Matt Carlson, Chris Schneider, Chris Daly
Abstract: Multiagent AI systems require consistent communication, but we lack methods to verify that agents share the same understanding of the terms used. Natural language is interpretable but vulnerable to semantic drift, while learned protocols are efficient but opaque. We propose a certification protocol based on the stimulus-meaning model, where agents are tested on shared observable events and terms are certified if empirical disagreement falls below a statistical threshold. In this protocol, agents restricting their reasoning to certified terms ("core-guarded reasoning") achieve provably bounded disagreement. We also outline mechanisms for detecting drift (recertification) and recovering shared vocabulary (renegotiation). In simulations with varying degrees of semantic divergence, core-guarding reduces disagreement by 72-96%. In a validation with fine-tuned language models, disagreement is reduced by 51%. Our framework provides a first step towards verifiable agent-to-agent communication.
Authors: Arun Vignesh Malarkkan, Wangyang Ying, Yanjie Fu
Abstract: Automated feature engineering (AFE) enables AI systems to autonomously construct high-utility representations from raw tabular data. However, existing AFE methods rely on statistical heuristics, yielding brittle features that fail under distribution shift. We introduce CAFE, a framework that reformulates AFE as a causally-guided sequential decision process, bridging causal discovery with reinforcement learning-driven feature construction. Phase I learns a sparse directed acyclic graph over features and the target to obtain soft causal priors, grouping features as direct, indirect, or other based on their causal influence with respect to the target. Phase II uses a cascading multi-agent deep Q-learning architecture to select causal groups and transformation operators, with hierarchical reward shaping and causal group-level exploration strategies that favor causally plausible transformations while controlling feature complexity. Across 15 public benchmarks (classification with macro-F1; regression with inverse relative absolute error), CAFE achieves up to 7% improvement over strong AFE baselines, reduces episodes-to-convergence, and delivers competitive time-to-target. Under controlled covariate shifts, CAFE reduces performance drop by ~4x relative to a non-causal multi-agent baseline, and produces more compact feature sets with more stable post-hoc attributions. These findings underscore that causal structure, used as a soft inductive prior rather than a rigid constraint, can substantially improve the robustness and efficiency of automated feature engineering.
Authors: Zihao Li, Fabrizio Russo
Abstract: Causal discovery seeks to uncover causal relations from data, typically represented as causal graphs, and is essential for predicting the effects of interventions. While expert knowledge is required to construct principled causal graphs, many statistical methods have been proposed to leverage observational data with varying formal guarantees. Causal Assumption-based Argumentation (ABA) is a framework that uses symbolic reasoning to ensure correspondence between input constraints and output graphs, while offering a principled way to combine data and expertise. We explore the use of large language models (LLMs) as imperfect experts for Causal ABA, eliciting semantic structural priors from variable names and descriptions and integrating them with conditional-independence evidence. Experiments on standard benchmarks and semantically grounded synthetic graphs demonstrate state-of-the-art performance, and we additionally introduce an evaluation protocol to mitigate memorisation bias when assessing LLMs for causal discovery.
Authors: Felix Fricke, Simon Malberg, Georg Groh
Abstract: Prompting schemes such as Chain of Thought, Tree of Thoughts, and Graph of Thoughts can significantly enhance the reasoning capabilities of large language models. However, most existing schemes require users to define static, problem-specific reasoning structures that lack adaptability to dynamic or unseen problem types. Additionally, these schemes are often under-optimized in terms of hyperparameters, prompts, runtime, and prompting cost. To address these limitations, we introduce Framework of Thoughts (FoT)--a general-purpose foundation framework for building and optimizing dynamic reasoning schemes. FoT comes with built-in features for hyperparameter tuning, prompt optimization, parallel execution, and intelligent caching, unlocking the latent performance potential of reasoning schemes. We demonstrate FoT's capabilities by implementing three popular schemes--Tree of Thoughts, Graph of Thoughts, and ProbTree--within FoT. We empirically show that FoT enables significantly faster execution, reduces costs, and achieves better task scores through optimization. We release our codebase to facilitate the development of future dynamic and efficient reasoning schemes.
Authors: Vered Tohar, Tsahi Hayat, Amir Leshem
Abstract: Can a machine write good poetry? Any positive answer raises fundamental questions about the nature and value of art. We report a seven-month poetry workshop in which a large language model was shaped into a digital poet through iterative in-context expert feedback, without retraining. Across sessions, the model developed a distinctive style and a coherent corpus, supported by quantitative and qualitative analyses, and it produced a pen name and author image. In a blinded authorship test with 50 humanities students and graduates (three AI poems and three poems by well-known poets each), judgments were at chance: human poems were labeled human 54% of the time and AI poems 52%, with 95% confidence intervals including 50%. After the workshop, a commercial publisher released a poetry collection authored by the model. These results show that workshop-style prompting can support long-horizon creative shaping and renew debates on creativity and authorship.
Authors: Yangjie Xu, Lujun Li, Lama Sleem, Niccolo Gentile, Yewei Song, Yiqun Wang, Siming Ji, Wenbo Wu, Radu State
Abstract: Agent Skill framework, now widely and officially supported by major players such as GitHub Copilot, LangChain, and OpenAI, performs especially well with proprietary models by improving context engineering, reducing hallucinations, and boosting task accuracy. Based on these observations, an investigation is conducted to determine whether the Agent Skill paradigm provides similar benefits to small language models (SLMs). This question matters in industrial scenarios where continuous reliance on public APIs is infeasible due to data-security and budget constraints requirements, and where SLMs often show limited generalization in highly customized scenarios. This work introduces a formal mathematical definition of the Agent Skill process, followed by a systematic evaluation of language models of varying sizes across multiple use cases. The evaluation encompasses two open-source tasks and a real-world insurance claims data set. The results show that tiny models struggle with reliable skill selection, while moderately sized SLMs (approximately 12B - 30B) parameters) benefit substantially from the Agent Skill approach. Moreover, code-specialized variants at around 80B parameters achieve performance comparable to closed-source baselines while improving GPU efficiency. Collectively, these findings provide a comprehensive and nuanced characterization of the capabilities and constraints of the framework, while providing actionable insights for the effective deployment of Agent Skills in SLM-centered environments.
Authors: Stephan Rabanser, Sayash Kapoor, Peter Kirgis, Kangheng Liu, Saiteja Utpala, Arvind Narayanan
Abstract: AI agents are increasingly deployed to execute important tasks. While rising accuracy scores on standard benchmarks suggest rapid progress, many agents still continue to fail in practice. This discrepancy highlights a fundamental limitation of current evaluations: compressing agent behavior into a single success metric obscures critical operational flaws. Notably, it ignores whether agents behave consistently across runs, withstand perturbations, fail predictably, or have bounded error severity. Grounded in safety-critical engineering, we provide a holistic performance profile by proposing twelve concrete metrics that decompose agent reliability along four key dimensions: consistency, robustness, predictability, and safety. Evaluating 14 agentic models across two complementary benchmarks, we find that recent capability gains have only yielded small improvements in reliability. By exposing these persistent limitations, our metrics complement traditional evaluations while offering tools for reasoning about how agents perform, degrade, and fail.
Authors: Weiwen Su, Yuhan Zhou, Zihan Wang, Naoki Yoshinaga, Masashi Toyoda
Abstract: Existing user simulations, where models generate user-like responses in dialogue, often lack verification that sufficient user personas are provided, questioning the validity of the simulations. To address this core concern, this work explores the task of identifying relevant but unknown personas of the simulation target for a given simulation context. We introduce PICQ, a novel dataset of context-aware choice questions, annotated with unknown personas (e.g., ''Is the user price-sensitive?'') that may influence user choices, and propose a multi-faceted evaluation scheme assessing fidelity, influence, and inaccessibility. Our benchmark of leading LLMs reveals a complex ''Fidelity vs. Insight'' dilemma governed by model scale: while influence generally scales with model size, fidelity to human patterns follows an inverted U-shaped curve. We trace this phenomenon to cognitive differences, particularly the human tendency for ''cognitive economy.'' Our work provides the first comprehensive benchmark for this crucial task, offering a new lens for understanding the divergent cognitive models of humans and advanced LLMs.
Authors: Mengyun Liu, Shanshan Huang, Jianan Jiang
Abstract: Large Action Models (LAMs) have shown immense potential in autonomous navigation by bridging high-level reasoning with low-level control. However, deploying these multi-billion parameter models on edge devices remains a significant challenge due to memory constraints and latency requirements. In this paper, we propose EdgeNav-QE, a novel framework that integrates Quantized Low-Rank Adaptation (QLoRA) with a dynamic early-exit (DEE) mechanism to optimize LAMs for real-time edge navigation. By quantizing the backbone to 4-bit precision and strategically placing early-exit branches, we enable the model to terminate inference early for simple navigation tasks while retaining full depth for complex decision-making. Experimental results on the Habitat-Sim environment with Matterport3D dataset using OpenVLA-7B backbone, demonstrate that EdgeNav-QE reduces inference latency by 82.7% and memory footprint by 66.7% compared to full-precision baselines, while maintaining 81.8% navigation success rate. Furthermore, it outperforms state-of-the-art static early-exit method by 17.9% in latency, demonstrating the superiority of content-aware adaptive computation for safety-critical applications.
Authors: Warren Johnson
Abstract: In "Compress or Route?" (Johnson, 2026), we found that code generation tolerates aggressive prompt compression (r >= 0.6) while chain-of-thought reasoning degrades gradually. That study was limited to HumanEval (164 problems), left the "perplexity paradox" mechanism unvalidated, and provided no adaptive algorithm. This paper addresses all three gaps. First, we validate across six code benchmarks (HumanEval, MBPP, HumanEval+, MultiPL-E) and four reasoning benchmarks (GSM8K, MATH, ARC-Challenge, MMLU-STEM), confirming the compression threshold generalizes across languages and difficulties. Second, we conduct the first per-token perplexity analysis (n=723 tokens), revealing a "perplexity paradox": code syntax tokens are preserved (high perplexity) while numerical values in math problems are pruned despite being task-critical (low perplexity). Signature injection recovers +34 percentage points in pass rate (5.3% to 39.3%; Cohen's h=0.890). Third, we propose TAAC (Task-Aware Adaptive Compression), achieving 22% cost reduction with 96% quality preservation, outperforming fixed-ratio compression by 7%. MBPP validation (n=1,800 trials) confirms systematic variation: 3.6% at r=0.3 to 54.6% at r=1.0.
Authors: Inwon Kang, Parikshit Ram, Yi Zhou, Horst Samulowitz, Oshani Seneviratne
Abstract: The Web is a rich source of structured data in the form of tables, from product catalogs and knowledge bases to scientific datasets. However, the heterogeneity of the structure and semantics of these tables makes it challenging to build a unified method that can effectively leverage the information they contain. Meanwhile, Large language models (LLMs) are becoming an increasingly integral component of web infrastructure for tasks like semantic search. This raises a crucial question: can we leverage these already-deployed LLMs to classify structured data in web-native tables (e.g., product catalogs, knowledge base exports, scientific data portals), avoiding the need for specialized models or extensive retraining? This work investigates a lightweight paradigm, $\textbf{Ta}$ble $\textbf{R}$epresentation with $\textbf{L}$anguage Model~($\textbf{TaRL}$), for few-shot tabular classification that directly utilizes semantic embeddings of individual table rows. We first show that naive application of these embeddings underperforms compared to specialized tabular models. We then demonstrate that their potentials can be unlocked with two key techniques: removing the common component from all embeddings and calibrating the softmax temperature. We show that a simple meta-learner, trained on handcrafted features, can learn to predict an appropriate temperature. This approach achieves performance comparable to state-of-the-art models in low-data regimes ($k \leq 32$) of semantically-rich tables. Our findings demonstrate the viability of reusing existing LLM infrastructure for efficient semantics-driven pathway to reuse existing LLM infrastructure for Web table understanding.
Authors: Pranav Bhandari, Usman Naseem, Mehwish Nasim
Abstract: Personality steering in large language models (LLMs) commonly relies on injecting trait-specific steering vectors, implicitly assuming that personality traits can be controlled independently. In this work, we examine whether this assumption holds by analysing the geometric relationships between Big Five personality steering directions. We study steering vectors extracted from two model families (LLaMA-3-8B and Mistral-8B) and apply a range of geometric conditioning schemes, from unconstrained directions to soft and hard orthonormalisation. Our results show that personality steering directions exhibit substantial geometric dependence: steering one trait consistently induces changes in others, even when linear overlap is explicitly removed. While hard orthonormalisation enforces geometric independence, it does not eliminate cross-trait behavioural effects and can reduce steering strength. These findings suggest that personality traits in LLMs occupy a slightly coupled subspace, limiting fully independent trait control.
Authors: Andrius Mat\v{s}enas, Anet Lello, T\~onis Lees, Hans Peep, Kim Lilii Tamm
Abstract: This study validates Large Language Models (LLMs) as a dynamic alternative to questionnaire-based personality assessment. Using a within-subjects experiment (N=33), we compared Big Five personality scores derived from guided LLM conversations against the gold-standard IPIP-50 questionnaire, while also measuring user-perceived accuracy. Results indicate moderate convergent validity (r=0.38-0.58), with Conscientiousness, Openness, and Neuroticism scores statistically equivalent between methods. Agreeableness and Extraversion showed significant differences, suggesting trait-specific calibration is needed. Notably, participants rated LLM-generated profiles as equally accurate as traditional questionnaire results. These findings suggest conversational AI offers a promising new approach to traditional psychometrics.
Authors: Karun Sharma, Vidushee Vats, Shengzhi Li, Yuxiang Wang, Zhongtian Sun, Prayag Tiwari
Abstract: Peer review relies on substantive, evidence-based questions, yet existing LLM-based approaches often generate surface-level queries, drawing over 50\% of their question tokens from a paper's first page. To bridge this gap, we develop IntelliReward, a novel reward model built from a frozen autoregressive LLM with trainable multi-head transformers over the final 50 token states, which outperforms API-based SFT baselines in predicting expert-level human preferences. By applying Decoupled Clip and Dynamic Sampling Policy Optimization (DAPO) with IntelliReward, we train IntelliAsk, a question-generation model aligned with human standards of effort, evidence, and grounding. We find consistent improvements on reasoning and writing benchmarks, suggesting reviewer-question quality correlates with broader capabilities. Compared to the Qwen3-32B base model, IntelliAsk shows measurable gains across diverse benchmarks, specifically improving performance on reasoning tasks like MuSR (68.3 vs 64.7 Acc) and complex writing evaluations such as WritingBench (8.31 vs 8.07). We release our implementation, expert preference annotations, and the IntelliReward model to provide an automatic evaluation benchmark for grounding, effort, and evidence in LLM-generated review questions.
Authors: David Y. Liu, Aditya Joshi, Paul Dawson
Abstract: Applications of narrative theories using large language models (LLMs) deliver promising use-cases in automatic story generation and understanding tasks. Our survey examines how natural language processing (NLP) research engages with fields of narrative studies, and proposes a taxonomy for ongoing efforts that reflect established distinctions in narratology. We discover patterns in the following: narrative datasets and tasks, narrative theories and NLP pipeline and methodological trends in prompting and fine-tuning. We highlight how LLMs enable easy connections of NLP pipelines with abstract narrative concepts and opportunities for interdisciplinary collaboration. Challenges remain in attempts to work towards any unified definition or benchmark of narrative related tasks, making model comparison difficult. For future directions, instead of the pursuit of a single, generalised benchmark for 'narrative quality', we believe that progress benefits more from efforts that focus on the following: defining and improving theory-based metrics for individual narrative attributes to incrementally improve model performance; conducting large-scale, theory-driven literary/social/cultural analysis; and creating experiments where outputs can be used to validate or refine narrative theories. This work provides a contextual foundation for more systematic and theoretically informed narrative research in NLP by providing an overview to ongoing research efforts and the broader narrative studies landscape.
Authors: Ha Na Cho, Sairam Sutari, Alexander Lopez, Hansen Bow, Kai Zheng
Abstract: Clinical natural language processing (NLP) models have shown promise for supporting hospital discharge planning by leveraging narrative clinical documentation. However, note-based models are particularly vulnerable to temporal and lexical leakage, where documentation artifacts encode future clinical decisions and inflate apparent predictive performance. Such behavior poses substantial risks for real-world deployment, where overconfident or temporally invalid predictions can disrupt clinical workflows and compromise patient safety. This study focuses on system-level design choices required to build safe and deployable clinical NLP under temporal leakage constraints. We present a lightweight auditing pipeline that integrates interpretability into the model development process to identify and suppress leakage-prone signals prior to final training. Using next-day discharge prediction after elective spine surgery as a case study, we evaluate how auditing affects predictive behavior, calibration, and safety-relevant trade-offs. Results show that audited models exhibit more conservative and better-calibrated probability estimates, with reduced reliance on discharge-related lexical cues. These findings emphasize that deployment-ready clinical NLP systems should prioritize temporal validity, calibration, and behavioral robustness over optimistic performance.
Authors: Md Asiful Islam, Mihai Surdeanu
Abstract: We propose a lightweight explainable guardrail (LEG) method for the classification of unsafe prompts. LEG uses a multi-task learning architecture to jointly learn a prompt classifier and an explanation classifier, where the latter labels prompt words that explain the safe/unsafe overall decision. LEG is trained using synthetic data for explainability, which is generated using a novel strategy that counteracts the confirmation biases of LLMs. Lastly, LEG's training process uses a novel loss that captures global explanation signals and combines cross-entropy and focal losses with uncertainty-based weighting. LEG obtains equivalent or better performance than the state-of-the-art for both prompt classification and explainability, both in-domain and out-of-domain on three datasets, despite the fact that its model size is considerably smaller than current approaches. If accepted, we will release all models and the annotated dataset publicly.
Authors: Jingyi Xu, Xingyu Ren, Zhiqiang You, Yumeng Zhang, Zhoupeng Shou
Abstract: Large language models show potential in task-oriented dialogue systems, yet existing training methods often rely on token-level likelihood or preference optimization, which poorly align with long-horizon task success. To address this, we propose Goal-Oriented Preference Optimization (GOPO), a hierarchical reinforcement learning framework that decouples strategy planning from response generation via an Expert Agent and a Customer Service Agent. The Expert Agent optimizes multi-turn goal preferences at the dialogue-trajectory level, while the Customer Service Agent generates responses strictly aligned with the selected strategy. We evaluate GOPO on public benchmarks and e-commerce customer service datasets, and introduce Task-focused Sequential Engagement (TSE), a sequence-level metric derived from real e-commerce interaction data. On the Mgshop dataset, GOPO improves TSE by 7.7% and 10.3% over PPO and Memento, with consistent gains in sequence-level reward and generation quality. Furthermore, a 14B model trained with GOPO achieves 2.7% and 1.5% higher TSE than Qwen-235B and GPT-5.2, respectively. Ablation studies confirm the Expert Agent's critical role in long-horizon optimization. GOPO demonstrates consistent improvements across other datasets as well. This work establishes a new paradigm for task-oriented dialogue systems in commercial scenarios, with code and datasets to be made public.
Authors: Barak Or
Abstract: Hybrid reasoning systems that combine learned components with model-based inference are increasingly deployed in tool-augmented decision loops, yet their runtime behavior under partial observability and sustained evidence mismatch remains poorly understood. In practice, failures often arise as gradual divergence of internal reasoning dynamics rather than as isolated prediction errors. This work studies runtime stability in hybrid reasoning systems from a Kalman-inspired perspective. We model reasoning as a stochastic inference process driven by an internal innovation signal and introduce cognitive drift as a measurable runtime phenomenon. Stability is defined in terms of detectability, bounded divergence, and recoverability rather than task-level correctness. We propose a runtime stability framework that monitors innovation statistics, detects emerging instability, and triggers recovery-aware control mechanisms. Experiments on multi-step, tool-augmented reasoning tasks demonstrate reliable instability detection prior to task failure and show that recovery, when feasible, re-establishes bounded internal behavior within finite time. These results emphasize runtime stability as a system-level requirement for reliable reasoning under uncertainty.
Authors: Yunhao Liu, Zian Jia, Xinyu Gao, Kanjun Xu, Yun Xiong
Abstract: Retrieval-Augmented Generation (RAG) effectively grounds Large Language Models (LLMs) with external knowledge and is widely applied to Web-related tasks. However, its scalability is hindered by excessive context length and redundant retrievals. Recent research on soft context compression aims to address this by encoding long documents into compact embeddings, yet they often underperform non-compressed RAG due to their reliance on auto-encoder-like full-compression that forces the encoder to compress all document information regardless of relevance to the input query. In this work, we conduct an analysis on this paradigm and reveal two fundamental limitations: (I) Infeasibility, full-compression conflicts with the LLM's downstream generation behavior; and (II) Non-necessity: full-compression is unnecessary and dilutes task-relevant information density. Motivated by these insights, we introduce SeleCom, a selector-based soft compression framework for RAG that redefines the encoder's role as query-conditioned information selector. The selector is decoder-only and is trained with a massive, diverse and difficulty-graded synthetic QA dataset with curriculum learning. Extensive experiments show that SeleCom significantly outperforms existing soft compression approaches and achieves competitive or superior performance to non-compression baselines, while reducing computation and latency by 33.8%~84.6%.
Authors: Annie Wong, Aske Plaat, Thomas B\"ack, Niki van Stein, Anna V. Kononova
Abstract: As large language models (LLMs) move from static reasoning tasks toward dynamic environments, their success depends on the ability to navigate and respond to an environment that changes as they interact at inference time. An underexplored factor in these settings is the representation of the state. Holding model parameters fixed, we systematically vary three key aspects: (1) state granularity (long form versus summary), (2) structure (natural language versus symbolic), and (3) spatial grounding (text-only versus images or textual map encodings) across sequential decision-making benchmarks. We find that trajectory summarisation improves performance by reducing noise and stabilising long-horizon reasoning. Second, natural language representations are the most robust across models, whereas structured encodings help mainly for models with strong code or structured output priors, such as JSON schemas. Third, while image-inputs show some benefit, text-based spatial encodings prove most effective. This advantage stems not from the spatial information itself, but from the act of construction, which compels the model to perform the spatial reasoning that static input does not elicit. Overall, we demonstrate that design choices for representing state are a decisive factor in performance, distinct from the availability of information itself. We note, however, that even with improved representations, current LLMs and VLMs remain brittle over long horizons, particularly when they must synthesise information to manage multiple subtasks to reach a goal.
Authors: Jinxiang Xie, Zihao Li, Wei He, Rui Ding, Shi Han, Dongmei Zhang
Abstract: Text analysis of tabular data relies on two core operations: \emph{summarization} for corpus-level theme extraction and \emph{tagging} for row-level labeling. A critical limitation of employing large language models (LLMs) for these tasks is their inability to meet the high standards of output stability demanded by data analytics. To address this challenge, we introduce \textbf{CAST} (\textbf{C}onsistency via \textbf{A}lgorithmic Prompting and \textbf{S}table \textbf{T}hinking), a framework that enhances output stability by constraining the model's latent reasoning path. CAST combines (i) Algorithmic Prompting to impose a procedural scaffold over valid reasoning transitions and (ii) Thinking-before-Speaking to enforce explicit intermediate commitments before final generation. To measure progress, we introduce \textbf{CAST-S} and \textbf{CAST-T}, stability metrics for bulleted summarization and tagging, and validate their alignment with human judgments. Experiments across publicly available benchmarks on multiple LLM backbones show that CAST consistently achieves the best stability among all baselines, improving Stability Score by up to 16.2\%, while maintaining or improving output quality.
Authors: Guoshan Liu, Bin Zhu, Yian Li, Jingjing Chen, Chong-Wah Ngo, Yu-Gang Jiang
Abstract: Recent advances in Multimodal Large Language Models (MLMMs) have enabled recipe generation from food images, yet outputs often contain semantically incorrect actions or ingredients despite high lexical scores (e.g., BLEU, ROUGE). To address this gap, we propose a semantically grounded framework that predicts and validates actions and ingredients as internal context for instruction generation. Our two-stage pipeline combines supervised fine-tuning (SFT) with reinforcement fine-tuning (RFT): SFT builds foundational accuracy using an Action-Reasoning dataset and ingredient corpus, while RFT employs frequency-aware rewards to improve long-tail action prediction and ingredient generalization. A Semantic Confidence Scoring and Rectification (SCSR) module further filters and corrects predictions. Experiments on Recipe1M show state-of-the-art performance and markedly improved semantic fidelity.
Authors: Daehoon Gwak, Minseo Jung, Junwoo Park, Minho Park, ChaeHun Park, Junha Hyung, Jaegul Choo
Abstract: Recent studies have shown that Large Language Models (LLMs) can improve their reasoning performance through self-generated few-shot examples, achieving results comparable to manually curated in-context examples. However, the underlying mechanism behind these gains remains unclear, making it hard to decide when and how to apply the technique effectively. In this work, we argue that the key benefit arises not from the generated examples themselves but from the act of creating them. To validate this, on reasoning-intensive tasks across diverse LLM architectures, we systematically evaluate three prompting strategies for in-context learning: (1) Zero-shot prompting; (2) Integrated prompting, where LLMs create and solve problems within a single, unified prompt; and (3) Decoupled prompting, where self-generated examples are reused as in-context examples, but the context of their creation itself is excluded. We conduct experiments across five widely used model architectures, demonstrating that Integrated prompting consistently outperforms both Zero-shot and Decoupled prompting. In contrast, Decoupled prompting offers only marginal gains over Zero-shot. Further, for a more in-depth analysis, we conduct an attention analysis and observe significant differences in attention patterns between Integrated and Decoupled prompting. These findings suggest that the advantage of self-generation prompting comes from the process of problem creation, not the examples themselves, providing valuable insights for designing more effective prompting strategies.
Authors: Most. Sharmin Sultana Samu, Nafisa Khan, Kazi Toufique Elahi, Tasnuva Binte Rahman, Md. Rakibul Islam, Farig Sadeque
Abstract: The integration of Artificial Intelligence (AI) necessitates determining whether systems function as tools or collaborative teammates. In this study, by synthesizing Human-AI Interaction (HAI) literature, we analyze this distinction across four dimensions: interaction design, trust calibration, collaborative frameworks and healthcare applications. Our analysis reveals that static interfaces and miscalibrated trust limit AI efficacy. Performance hinges on aligning transparency with cognitive workflows, yet a fluency trap often inflates trust without improving decision-making. Consequently, an overemphasis on explainability leaves systems largely passive. Our findings show that current AI systems remain largely passive due to an overreliance on explainability-centric designs and that transitioning AI to an active teammate requires adaptive, context-aware interactions that support shared mental models and the dynamic negotiation of authority between humans and AI.
Authors: Dhiman Goswami, Jai Kruthunz Naveen Kumar, Sanchari Das
Abstract: Natural Language Processing (NLP) is integral to social media analytics but often processes content containing Personally Identifiable Information (PII), behavioral cues, and metadata raising privacy risks such as surveillance, profiling, and targeted advertising. To systematically assess these risks, we review 203 peer-reviewed papers and propose the NLP Privacy Risk Identification in Social Media (NLP-PRISM) framework, which evaluates vulnerabilities across six dimensions: data collection, preprocessing, visibility, fairness, computational risk, and regulatory compliance. Our analysis shows that transformer models achieve F1-scores ranging from 0.58-0.84, but incur a 1% - 23% drop under privacy-preserving fine-tuning. Using NLP-PRISM, we examine privacy coverage in six NLP tasks: sentiment analysis (16), emotion detection (14), offensive language identification (19), code-mixed processing (39), native language identification (29), and dialect detection (24) revealing substantial gaps in privacy research. We further found a (reduced by 2% - 9%) trade-off in model utility, MIA AUC (membership inference attacks) 0.81, AIA accuracy 0.75 (attribute inference attacks). Finally, we advocate for stronger anonymization, privacy-aware learning, and fairness-driven training to enable ethical NLP in social media contexts.
Authors: Berry Gerrits
Abstract: In this positioning paper, we evaluate the problem-solving and reasoning capabilities of contemporary Large Language Models (LLMs) through their performance in Zork, the seminal text-based adventure game first released in 1977. The game's dialogue-based structure provides a controlled environment for assessing how LLM-based chatbots interpret natural language descriptions and generate appropriate action sequences to succeed in the game. We test the performance of leading proprietary models - ChatGPT, Claude, and Gemini - under both minimal and detailed instructions, measuring game progress through achieved scores as the primary metric. Our results reveal that all tested models achieve less than 10% completion on average, with even the best-performing model (Claude Opus 4.5) reaching only approximately 75 out of 350 possible points. Notably, providing detailed game instructions offers no improvement, nor does enabling ''extended thinking''. Qualitative analysis of the models' reasoning processes reveals fundamental limitations: repeated unsuccessful actions suggesting an inability to reflect on one's own thinking, inconsistent persistence of strategies, and failure to learn from previous attempts despite access to conversation history. These findings suggest substantial limitations in current LLMs' metacognitive abilities and problem-solving capabilities within the domain of text-based games, raising questions about the nature and extent of their reasoning capabilities.
Authors: Chuyang Ye, Haoxian Jing, Qinting Jiang, Yixi Lin, Qiang Li, Xing Tang, Jingyan Jiang
Abstract: Tactile-vision-language (TVL) models are increasingly deployed in real-world robotic and multimodal perception tasks, where test-time distribution shifts are unavoidable. Existing test-time adaptation (TTA) methods provide filtering in unimodal settings but lack explicit treatment of modality-wise reliability under asynchronous cross-modal shifts, leaving them brittle when some modalities become unreliable. We study TTA for TVL models under such shifts and propose a reliability-aware framework that estimates per-modality reliability from prediction uncertainty and perturbation-based responses. This shared reliability signal is used to (i) filter unreliable test samples, (ii) adaptively fuse tactile, visual, and language features, and (iii) regularize test-time optimization with a reliability-guided objective. On the TAG-C benchmark and additional TVL scenarios, our approach consistently outperforms strong TTA baselines, achieving accuracy gains of up to 49.9\% under severe modality corruptions, underscoring the importance of explicit modality-wise reliability modeling for robust test-time adaptation.
Authors: Zhenxing Xu, Brikit Lu, Weidong Bao, Zhengqiu Zhu, Junsong Zhang, Hui Yan, Wenhao Lu, Ji Wang
Abstract: Current Visual-Language Navigation (VLN) methodologies face a trade-off between semantic understanding and control precision. While Multimodal Large Language Models (MLLMs) offer superior reasoning, deploying them as low-level controllers leads to high latency, trajectory oscillations, and poor generalization due to weak geometric grounding. To address these limitations, we propose Fly0, a framework that decouples semantic reasoning from geometric planning. The proposed method operates through a three-stage pipeline: (1) an MLLM-driven module for grounding natural language instructions into 2D pixel coordinates; (2) a geometric projection module that utilizes depth data to localize targets in 3D space; and (3) a geometric planner that generates collision-free trajectories. This mechanism enables robust navigation even when visual contact is lost. By eliminating the need for continuous inference, Fly0 reduces computational overhead and improves system stability. Extensive experiments in simulation and real-world environments demonstrate that Fly0 outperforms state-of-the-art baselines, improving the Success Rate by over 20\% and reducing Navigation Error (NE) by approximately 50\% in unstructured environments. Our code is available at https://github.com/xuzhenxing1/Fly0.
Authors: Kaaustaaub Shankar, Kelly Cohen
Abstract: Generalized Additive Models (GAMs) balance predictive accuracy and interpretability, but manually configuring their structure is challenging. We propose using the multi-objective genetic algorithm NSGA-II to automatically optimize GAMs, jointly minimizing prediction error (RMSE) and a Complexity Penalty that captures sparsity, smoothness, and uncertainty. Experiments on the California Housing dataset show that NSGA-II discovers GAMs that outperform baseline LinearGAMs in accuracy or match performance with substantially lower complexity. The resulting models are simpler, smoother, and exhibit narrower confidence intervals, enhancing interpretability. This framework provides a general approach for automated optimization of transparent, high-performing models. The code can be found at https://github.com/KaaustaaubShankar/GeneticAdditiveModels.
URLs: https://github.com/KaaustaaubShankar/GeneticAdditiveModels.
Authors: Mingchun Sun, Rongqiang Zhao, Zhennan Huang, Songyu Ding, Jie Liu
Abstract: In industrial scenarios, data augmentation is an effective approach to improve model performance. However, its benefits are not unidirectionally beneficial. There is no theoretical research or established estimation for the optimal sample size (OSS) in augmentation, nor is there an established metric to evaluate the accuracy of OSS or its deviation from the ground truth. To address these issues, we propose an information-theoretic optimal sample size estimation (IT-OSE) to provide reliable OSS estimation for industrial data augmentation. An interval coverage and deviation (ICD) score is proposed to evaluate the estimated OSS intuitively. The relationship between OSS and dominant factors is theoretically analyzed and formulated, thereby enhancing the interpretability. Experiments show that, compared to empirical estimation, the IT-OSE increases accuracy in classification tasks across baseline models by an average of 4.38%, and reduces MAPE in regression tasks across baseline models by an average of 18.80%. The improvements in downstream model performance are more stable. ICDdev in the ICD score is also reduced by an average of 49.30%. The determinism of OSS is enhanced. Compared to exhaustive search, the IT-OSE achieves the same OSS while reducing computational and data costs by an average of 83.97% and 93.46%. Furthermore, practicality experiments demonstrate that the IT-OSE exhibits generality across representative sensor-based industrial scenarios.
Authors: Jingjing Fan, Yushan Liu, Shoujie Li, Botao Ren, Siyuan Li, Xiao-Ping Zhang, Wenbo Ding, Zhidong Deng
Abstract: General vision-language models increasingly support unified spatiotemporal reasoning over long video streams, yet deploying such capabilities on robots remains constrained by the prohibitive latency of processing long-horizon histories and generating high-dimensional future predictions. To bridge this gap, we present FUTURE-VLA, a unified architecture that reformulates long-horizon control and future forecasting as a monolithic sequence-generation task. Adopting a dual-sided efficiency paradigm, FUTURE-VLA leverages a temporally adaptive compression strategy to maximize spatiotemporal information density, enabling the ingestion of extensive multi-view histories while maintaining constant inference latency. Simultaneously, it performs latent-space autoregression to align actionable dynamics with reviewable visual look-aheads in a single forward pass. These real-time predictive capabilities further enable a prediction-guided Human-In-the-Loop mechanism via interactive execution gating, allowing operators to dynamically validate behaviors based on interpretable future previews. Extensive evaluations demonstrate that FUTURE-VLA establishes new state-of-the-art performance, attaining success rates of 99.2% on LIBERO, 75.4% on RoboTwin, and 78.0% on a real-world Piper platform, all with a $16\times$ extended spatiotemporal window while maintaining the inference latency of a single-frame baseline.
Authors: Boyu Li, Xingchun Zhu, Yonghui Wu
Abstract: Reliable, continuous neural sensing on wearable edge platforms is fundamental to long-term health monitoring; however, for electroencephalography (EEG)-based sleep monitoring, dense high-frequency processing is often computationally prohibitive under tight energy budgets. To address this bottleneck, this paper proposes NeuroSleep, an integrated event-driven sensing and inference system for energy-efficient sleep staging. NeuroSleep first converts raw EEG into complementary multi-scale bipolar event streams using Residual Adaptive Multi-Scale Delta Modulation (R-AMSDM), enabling an explicit fidelity-sparsity trade-off at the sensing front end. Furthermore, NeuroSleep adopts a hierarchical inference architecture that comprises an Event-based Adaptive Multi-scale Response (EAMR) module for local feature extraction, a Local Temporal-Attention Module (LTAM) for context aggregation, and an Epoch-Leaky Integrate-and-Fire (ELIF) module to capture long-term state persistence. Experimental results using subject-independent 5-fold cross-validation on the Sleep-EDF Expanded dataset demonstrate that NeuroSleep achieves a mean accuracy of 74.2% with only 0.932 M parameters while reducing sparsity-adjusted effective operations by approximately 53.6% relative to dense processing. Compared with the representative dense Transformer baseline, NeuroSleep improves accuracy by 7.5% with a 45.8% reduction in computational load. By bridging neuromorphic encoding with state-aware modeling, NeuroSleep provides a scalable solution for always-on sleep analysis in resource-constrained wearable scenarios.
Authors: Paul Tschisgale, Peter Wulff
Abstract: Large language models (LLMs) are increasingly used in research both as tools and as objects of investigation. Much of this work implicitly assumes that LLM performance under fixed conditions (identical model snapshot, hyperparameters, and prompt) is time-invariant. If average output quality changes systematically over time, this assumption is violated, threatening the reliability, validity, and reproducibility of findings. To empirically examine this assumption, we conducted a longitudinal study on the temporal variability of GPT-4o's average performance. Using a fixed model snapshot, fixed hyperparameters, and identical prompting, GPT-4o was queried via the API to solve the same multiple-choice physics task every three hours for approximately three months. Ten independent responses were generated at each time point and their scores were averaged. Spectral (Fourier) analysis of the resulting time series revealed notable periodic variability in average model performance, accounting for approximately 20% of the total variance. In particular, the observed periodic patterns are well explained by the interaction of a daily and a weekly rhythm. These findings indicate that, even under controlled conditions, LLM performance may vary periodically over time, calling into question the assumption of time invariance. Implications for ensuring validity and replicability of research that uses or investigates LLMs are discussed.
Authors: Md Hossain Sahadath, Qiyun Cheng, Shaowu Pan, Wei Ji
Abstract: This work introduces a neural operator based surrogate modeling framework for neutron transport computation. Two architectures, the Deep Operator Network (DeepONet) and the Fourier Neural Operator (FNO), were trained for fixed source problems to learn the mapping from anisotropic neutron sources, Q(x,{\mu}), to the corresponding angular fluxes, {\psi}(x,{\mu}), in a one-dimensional slab geometry. Three distinct models were trained for each neural operator, corresponding to different scattering ratios (c = 0.1, 0.5, & 1.0), providing insight into their performance across distinct transport regimes (absorption-dominated, moderate, and scattering-dominated). The models were subsequently evaluated on a wide range of previously unseen source configurations, demonstrating that FNO generally achieves higher predictive accuracy, while DeepONet offers greater computational efficiency. Both models offered significant speedups that become increasingly pronounced as the scattering ratio increases, requiring <0.3% of the runtime of a conventional S_N solver. The surrogate models were further incorporated into the S_N k-eigenvalue solver, replacing the computationally intensive transport sweep loop with a single forward pass. Across varying fission cross sections and spatial-angular grids, both neural operator solvers reproduced reference eigenvalues with deviations up to 135 pcm for DeepONet and 112 pcm for FNO, while reducing runtime to <0.1% of that of the S_N solver on relatively fine grids. These results demonstrate the strong potential of neural operator frameworks as accurate, efficient, and generalizable surrogates for neutron transport, paving the way for real-time digital twin applications and repeated evaluations, such as in design optimization.
Authors: Maijunxian Wang, Yijiang Li, Bingyang Wang, Tianwei Zhao, Ran Ji, Qingying Gao, Emmy Liu, Hokin Deng, Dezhi Luo
Abstract: Visual perspective taking--inferring how the world appears from another's viewpoint--is foundational to social cognition. We introduce FlipSet, a diagnostic benchmark for Level-2 visual perspective taking (L2 VPT) in vision-language models. The task requires simulating 180-degree rotations of 2D character strings from another agent's perspective, isolating spatial transformation from 3D scene complexity. Evaluating 103 VLMs reveals systematic egocentric bias: the vast majority perform below chance, with roughly three-quarters of errors reproducing the camera viewpoint. Control experiments expose a compositional deficit--models achieve high theory-of-mind accuracy and above-chance mental rotation in isolation, yet fail catastrophically when integration is required. This dissociation indicates that current VLMs lack the mechanisms needed to bind social awareness to spatial operations, suggesting fundamental limitations in model-based spatial reasoning. FlipSet provides a cognitively grounded testbed for diagnosing perspective-taking capabilities in multimodal systems.
Authors: Pengcheng Zhou, Haochen Li, Zhiqiang Nie, JiaLe Chen, Qing Gong, Weizhen Zhang, Chun Yu
Abstract: Retrieval-Augmented Generation (RAG) effectively mitigates hallucinations in LLMs by incorporating external knowledge. However, the inherent discrete representation of text in existing frameworks often results in a loss of semantic integrity, leading to retrieval deviations. Inspired by the human episodic memory mechanism, we propose CogitoRAG, a RAG framework that simulates human cognitive memory processes. The core of this framework lies in the extraction and evolution of the Semantic Gist. During the offline indexing stage, CogitoRAG first deduces unstructured corpora into gist memory corpora, which are then transformed into a multi-dimensional knowledge graph integrating entities, relational facts, and memory nodes. In the online retrieval stage, the framework handles complex queries via Query Decomposition Module that breaks them into comprehensive sub-queries, mimicking the cognitive decomposition humans employ for complex information. Subsequently, Entity Diffusion Module performs associative retrieval across the graph, guided by structural relevance and an entity-frequency reward mechanism. Furthermore, we propose the CogniRank algorithm, which precisely reranks candidate passages by fusing diffusion-derived scores with semantic similarity. The final evidence is delivered to the generator in a passage-memory pairing format, providing high-density information support. Experimental results across five mainstream QA benchmarks and multi-task generation on GraphBench demonstrate that CogitoRAG significantly outperforms state-of-the-art RAG methods, showcasing superior capabilities in complex knowledge integration and reasoning.
Authors: Rujikorn Charakorn, Edoardo Cetin, Shinnosuke Uesaka, Robert Tjarko Lange
Abstract: Long input sequences are central to in-context learning, document understanding, and multi-step reasoning of Large Language Models (LLMs). However, the quadratic attention cost of Transformers makes inference memory-intensive and slow. While context distillation (CD) can transfer information into model parameters, per-prompt distillation is impractical due to training costs and latency. To address these limitations, we propose Doc-to-LoRA (D2L), a lightweight hypernetwork that meta-learns to perform approximate CD within a single forward pass. Given an unseen prompt, D2L generates a LoRA adapter for a target LLM, enabling subsequent queries to be answered without re-consuming the original context, reducing latency and KV-cache memory consumption during inference of the target LLM. On a long-context needle-in-a-haystack task, D2L successfully learns to map contexts into adapters that store the needle information, achieving near-perfect zero-shot accuracy at sequence lengths exceeding the target LLM's native context window by more than 4x. On real-world QA datasets with limited compute, D2L outperforms standard CD while significantly reducing peak memory consumption and update latency. We envision that D2L can facilitate rapid adaptation of LLMs, opening up the possibility of frequent knowledge updates and personalized chat behavior.
Authors: Pengfei Zhang, Tianxin Xie, Minghao Yang, Li Liu
Abstract: Deep learning-based respiratory auscultation is currently hindered by two fundamental challenges: (i) inherent information loss, as converting signals into spectrograms discards transient acoustic events and clinical context; (ii) limited data availability, exacerbated by severe class imbalance. To bridge these gaps, we present Resp-Agent, an autonomous multimodal system orchestrated by a novel Active Adversarial Curriculum Agent (Thinker-A$^2$CA). Unlike static pipelines, Thinker-A$^2$CA serves as a central controller that actively identifies diagnostic weaknesses and schedules targeted synthesis in a closed loop. To address the representation gap, we introduce a Modality-Weaving Diagnoser that weaves EHR data with audio tokens via Strategic Global Attention and sparse audio anchors, capturing both long-range clinical context and millisecond-level transients. To address the data gap, we design a Flow Matching Generator that adapts a text-only Large Language Model (LLM) via modality injection, decoupling pathological content from acoustic style to synthesize hard-to-diagnose samples. As a foundation for these efforts, we introduce Resp-229k, a benchmark corpus of 229k recordings paired with LLM-distilled clinical narratives. Extensive experiments demonstrate that Resp-Agent consistently outperforms prior approaches across diverse evaluation settings, improving diagnostic robustness under data scarcity and long-tailed class imbalance. Our code and data are available at https://github.com/zpforlove/Resp-Agent.
Authors: Chuang Niu, Pengwei Wu, Bruno De Man, Ge Wang
Abstract: Foundation models (FMs) are rapidly reshaping medical imaging, shifting the field from narrowly trained, task-specific networks toward large, general-purpose models that can be adapted across modalities, anatomies, and clinical tasks. In this review, we synthesize the emerging landscape of medical imaging FMs along three major axes: principles of FM design, applications of FMs, and forward-looking challenges and opportunities. Taken together, this review provides a technically grounded, clinically aware, and future-facing roadmap for developing FMs that are not only powerful and versatile but also trustworthy and ready for responsible translation into clinical practice.
Authors: Xianwei Mao, Kai Ye, Sheng Zhou, Nan Zhang, Haikuan Huang, Bin Li, Jiajun Bu
Abstract: Knowledge-based Visual Question Answering (KB-VQA) requires models to answer questions by integrating visual information with external knowledge. However, retrieved knowledge is often noisy, partially irrelevant, or misaligned with the visual content, while internal model knowledge is difficult to control and interpret. Naive aggregation of these sources limits reasoning effectiveness and reduces answer accuracy. To address this, we propose MaS-VQA, a selection-driven framework that tightly couples explicit knowledge filtering with implicit knowledge reasoning. MaS-VQA first retrieves candidate passages and applies a Mask-and-Select mechanism to jointly prune irrelevant image regions and weakly relevant knowledge fragments, producing compact, high-signal multimodal knowledge . This filtered knowledge then guides the activation of internal knowledge in a constrained semantic space, enabling complementary co-modeling of explicit and implicit knowledge for robust answer prediction. Experiments on Encyclopedic-VQA and InfoSeek demonstrate consistent performance gains across multiple MLLM backbones, and ablations verify that the selection mechanism effectively reduces noise and enhances knowledge utilization.
Authors: Zelin Xu, Yupu Zhang, Saugat Adhikari, Saiful Islam, Tingsong Xiao, Zibo Liu, Shigang Chen, Da Yan, Zhe Jiang
Abstract: Benchmarking spatial reasoning in multimodal large language models (MLLMs) has attracted growing interest in computer vision due to its importance for embodied AI and other agentic systems that require precise interaction with the physical world. However, spatial reasoning on Earth imagery has lagged behind, as it uniquely involves grounding objects in georeferenced images and quantitatively reasoning about distances, directions, and topological relations using both visual cues and vector geometry coordinates (e.g., 2D bounding boxes, polylines, and polygons). Existing benchmarks for Earth imagery primarily focus on 2D spatial grounding, image captioning, and coarse spatial relations (e.g., simple directional or proximity cues). They lack support for quantitative direction and distance reasoning, systematic topological relations, and complex object geometries beyond bounding boxes. To fill this gap, we propose \textbf{EarthSpatialBench}, a comprehensive benchmark for evaluating spatial reasoning in MLLMs on Earth imagery. The benchmark contains over 325K question-answer pairs spanning: (1) qualitative and quantitative reasoning about spatial distance and direction; (2) systematic topological relations; (3) single-object queries, object-pair queries, and compositional aggregate group queries; and (4) object references expressed via textual descriptions, visual overlays, and explicit geometry coordinates, including 2D bounding boxes, polylines, and polygons. We conducted extensive experiments on both open-source and proprietary models to identify limitations in the spatial reasoning of MLLMs.
Authors: Valentin Dorseuil (DI-ENS), Jamal Atif (CMAP), Olivier Capp\'e (DI-ENS)
Abstract: Can the privacy vulnerability of individual data points be assessed without retraining models or explicitly simulating attacks? We answer affirmatively by showing that exposure to membership inference attack (MIA) is fundamentally governed by a data point's influence on the learned model. We formalize this in the linear setting by establishing a theoretical correspondence between individual MIA risk and the leverage score, identifying it as a principled metric for vulnerability. This characterization explains how data-dependent sensitivity translates into exposure, without the computational burden of training shadow models. Building on this, we propose a computationally efficient generalization of the leverage score for deep learning. Empirical evaluations confirm a strong correlation between the proposed score and MIA success, validating this metric as a practical surrogate for individual privacy risk assessment.
Authors: Rakshit Kumar Singh, Aryan Amit Barsainyan, Bharath Ramsundar
Abstract: Density Functional Theory (DFT) is widely used for first-principles simulations in chemistry and materials science, but its computational cost remains a key limitation for large systems. Motivated by recent advances in ML-based exchange-correlation (XC) functionals, this paper introduces a differentiable framework that integrates machine learning models into density functional theory (DFT) for solids and other periodic systems. The framework defines a clean API for neural network models that can act as drop in replacements for conventional exchange-correlation (XC) functionals and enables gradients to flow through the full self-consistent DFT workflow. The framework is implemented in Python using a PyTorch backend, making it fully differentiable and easy to use with standard deep learning tools. We integrate the implementation with the DeepChem library to promote the reuse of established models and to lower the barrier for experimentation. In initial benchmarks against established electronic structure packages (GPAW and PySCF), our models achieve relative errors on the order of 5-10%.
Authors: Yuval Felendler, Parth A. Gandhi, Idan Habler, Yuval Elovici, Asaf Shabtai
Abstract: Model Context Protocols (MCPs) provide a unified platform for agent systems to discover, select, and orchestrate tools across heterogeneous execution environments. As MCP-based systems scale to incorporate larger tool catalogs and multiple concurrently connected MCP servers, traditional tool-by-tool invocation increases coordination overhead, fragments state management, and limits support for wide-context operations. To address these scalability challenges, recent MCP designs have incorporated code execution as a first-class capability, an approach called Code Execution MCP (CE-MCP). This enables agents to consolidate complex workflows, such as SQL querying, file analysis, and multi-step data transformations, into a single program that executes within an isolated runtime environment. In this work, we formalize the architectural distinction between context-coupled (traditional) and context-decoupled (CE-MCP) models, analyzing their fundamental scalability trade-offs. Using the MCP-Bench framework across 10 representative servers, we empirically evaluate task behavior, tool utilization patterns, execution latency, and protocol efficiency as the scale of connected MCP servers and available tools increases, demonstrating that while CE-MCP significantly reduces token usage and execution latency, it introduces a vastly expanded attack surface. We address this security gap by applying the MAESTRO framework, identifying sixteen attack classes across five execution phases-including specific code execution threats such as exception-mediated code injection and unsafe capability synthesis. We validate these vulnerabilities through adversarial scenarios across multiple LLMs and propose a layered defense architecture comprising containerized sandboxing and semantic gating. Our findings provide a rigorous roadmap for balancing scalability and security in production-ready executable agent workflows.
Authors: Carlo Cena, Mauro Martini, Marcello Chiaberge
Abstract: Reliable spacecraft attitude control depends on accurate prediction of attitude dynamics, particularly when model-based strategies such as Model Predictive Control (MPC) are employed, where performance is limited by the quality of the internal system model. For spacecraft with complex dynamics, obtaining accurate physics-based models can be difficult, time-consuming, or computationally heavy. Learning-based system identification presents a compelling alternative; however, models trained exclusively on data frequently exhibit fragile stability properties and limited extrapolation capability. This work explores Physics-Informed Neural Networks (PINNs) for modeling spacecraft attitude dynamics and contrasts it with a conventional data-driven approach. A comprehensive dataset is generated using high-fidelity numerical simulations, and two learning methodologies are investigated: a purely data-driven pipeline and a physics-regularized approach that incorporates prior knowledge into the optimization process. The results indicate that embedding physical constraints during training leads to substantial improvements in predictive reliability, achieving a 68.17% decrease in mean relative error relative. When deployed within an MPC architecture, the physics-informed models yield superior closed-loop tracking performance and improved robustness to uncertainty. Furthermore, a hybrid control formulation that merges the learned nonlinear dynamics with a nominal linear model enables consistent steady-state convergence and significantly faster response, reducing settling times by 61.52%-76.42% under measurement noise and reaction wheel friction.
Authors: Md Mofijul Islam, Md Sirajus Salekin, Nivedha Balakrishnan, Vincil C. Bishop III, Niharika Jain, Spencer Romo, Bob Strahan, Boyi Xie, Diego A. Socolinsky
Abstract: Document understanding in real-world applications often requires processing heterogeneous, multi-page document packets containing multiple documents stitched together. Despite recent advances in visual document understanding, the fundamental task of document packet splitting, which involves separating a document packet into individual units, remains largely unaddressed. We present the first comprehensive benchmark dataset, DocSplit, along with novel evaluation metrics for assessing the document packet splitting capabilities of large language models. DocSplit comprises five datasets of varying complexity, covering diverse document types, layouts, and multimodal settings. We formalize the DocSplit task, which requires models to identify document boundaries, classify document types, and maintain correct page ordering within a document packet. The benchmark addresses real-world challenges, including out-of-order pages, interleaved documents, and documents lacking clear demarcations. We conduct extensive experiments evaluating multimodal LLMs on our datasets, revealing significant performance gaps in current models' ability to handle complex document splitting tasks. The DocSplit benchmark datasets and proposed novel evaluation metrics provide a systematic framework for advancing document understanding capabilities essential for legal, financial, healthcare, and other document-intensive domains. We release the datasets to facilitate future research in document packet processing.
Authors: Yiwen Wang, Jiahao Qin
Abstract: High-speed optical-resolution photoacoustic microscopy (OR-PAM) with bidirectional raster scanning doubles imaging speed but introduces coupled domain shift and geometric misalignment between forward and backward scan lines. Existing registration methods, constrained by brightness constancy assumptions, achieve limited alignment quality, while recent generative approaches address domain shift through complex architectures that lack temporal awareness across frames. We propose GPEReg-Net, a scene-appearance disentanglement framework that separates domain-invariant scene features from domain-specific appearance codes via Adaptive Instance Normalization (AdaIN), enabling direct image-to-image registration without explicit deformation field estimation. To exploit temporal structure in sequential acquisitions, we introduce a Global Position Encoding (GPE) module that combines learnable position embeddings with sinusoidal encoding and cross-frame attention, allowing the network to leverage context from neighboring frames for improved temporal coherence. On the OR-PAM-Reg-4K benchmark (432 test samples), GPEReg-Net achieves NCC of 0.953, SSIM of 0.932, and PSNR of 34.49dB, surpassing the state-of-the-art by 3.8% in SSIM and 1.99dB in PSNR while maintaining competitive NCC. Code is available at https://github.com/JiahaoQin/GPEReg-Net.
Authors: Pedro Reynolds-Cu\'ellar (Robotics,AI Institute), Marisol Wong-Villacres (Escuela Superior Polit\'ecnica del Litoral), Adriana Alvarado Garcia (IBM Research), Heila Precel (Robotics,AI Institute)
Abstract: Dataset documentation is widely recognized as essential for the responsible development of automated systems. Despite growing efforts to support documentation through different kinds of artifacts, little is known about the motivations shaping documentation tool design or the factors hindering their adoption. We present a systematic review supported by mixed-methods analysis of 59 dataset documentation publications to examine the motivations behind building documentation tools, how authors conceptualize documentation practices, and how these tools connect to existing systems, regulations, and cultural norms. Our analysis shows four persistent patterns in dataset documentation conceptualization that potentially impede adoption and standardization: unclear operationalizations of documentation's value, decontextualized designs, unaddressed labor demands, and a tendency to treat integration as future work. Building on these findings, we propose a shift in Responsible AI tool design toward institutional rather than individual solutions, and outline actions the HCI community can take to enable sustainable documentation practices.
Authors: Cherish Puniani, Tushar Kumar, Arnav Bendre, Gaurav Kumar, Shree Singhi
Abstract: Inspired by non-equilibrium thermodynamics, diffusion models have achieved state-of-the-art performance in generative modeling. However, their iterative sampling nature results in high inference latency. While recent distillation techniques accelerate sampling, they discard intermediate trajectory steps. This sparse supervision leads to a loss of structural information and introduces significant discretization errors. To mitigate this, we propose B-DENSE, a novel framework that leverages multi-branch trajectory alignment. We modify the student architecture to output $K$-fold expanded channels, where each subset corresponds to a specific branch representing a discrete intermediate step in the teacher's trajectory. By training these branches to simultaneously map to the entire sequence of the teacher's target timesteps, we enforce dense intermediate trajectory alignment. Consequently, the student model learns to navigate the solution space from the earliest stages of training, demonstrating superior image generation quality compared to baseline distillation frameworks.
Authors: Junbo Jacob Lian, Yujun Sun, Huiling Chen, Chaoyu Zhang, Chung-Piaw Teo
Abstract: Large language models (LLMs) can translate natural language into optimization code, but silent failures pose a critical risk: code that executes and returns solver-feasible solutions may encode semantically incorrect formulations, creating a feasibility-correctness gap of up to 90 percentage points on compositional problems. We introduce ReLoop, addressing silent failures from two complementary directions. Structured generation decomposes code production into a four-stage reasoning chain (understand, formalize, synthesize, verify) that mirrors expert modeling practice, with explicit variable-type reasoning and self-verification to prevent formulation errors at their source. Behavioral verification detects errors that survive generation by testing whether the formulation responds correctly to solver-based parameter perturbation, without requiring ground truth -- an external semantic signal that bypasses the self-consistency problem inherent in LLM-based code review. The two mechanisms are complementary: structured generation dominates on complex compositional problems, while behavioral verification becomes the largest single contributor on problems with localized formulation defects. Together with execution recovery via IIS-enhanced diagnostics, ReLoop raises correctness from 22.6% to 31.1% and execution from 72.1% to 100.0% on the strongest model, with consistent gains across five models spanning three paradigms (foundation, SFT, RL) and three benchmarks. We additionally release RetailOpt-190, 190 compositional retail optimization scenarios targeting the multi-constraint interactions where LLMs most frequently fail.
Authors: Jayadev Billa
Abstract: Capability emergence during neural network training remains mechanistically opaque. We track five geometric measures across five model scales (405K-85M parameters), 120+ emergence events in eight algorithmic tasks, and three Pythia language models (160M-2.8B). We find: (1) training begins with a universal representation collapse to task-specific floors that are scale-invariant across a 210X parameter range (e.g., modular arithmetic collapses to RANKME ~ 2.0 regardless of model size); (2) collapse propagates top-down through layers (32/32 task X model consistency), contradicting bottom-up feature-building intuition; (3) a geometric hierarchy in which representation geometry leads emergence (75-100% precursor rate for hard tasks), while the local learning coefficient is synchronous (0/24 precursor) and Hessian measures lag. We also delineate prediction limits: geometric measures encode coarse task difficulty but not fine-grained timing (within-class concordance 27%; when task ordering reverses across scales, prediction fails at 26%). On Pythia, global geometric patterns replicate but per-task precursor signals do not -- the precursor relationship requires task-training alignment that naturalistic pre-training does not provide. Our contribution is the geometric anatomy of emergence and its boundary conditions, not a prediction tool.
Authors: Jose Rojas, Aristotelis Papatheodorou, Sergi Martinez, Ioannis Havoutis, Carlos Mastalli
Abstract: We introduce ODYN, a novel all-shifted primal-dual non-interior-point quadratic programming (QP) solver designed to efficiently handle challenging dense and sparse QPs. ODYN combines all-shifted nonlinear complementarity problem (NCP) functions with proximal method of multipliers to robustly address ill-conditioned and degenerate problems, without requiring linear independence of the constraints. It exhibits strong warm-start performance and is well suited to both general-purpose optimization, and robotics and AI applications, including model-based control, estimation, and kernel-based learning methods. We provide an open-source implementation and benchmark ODYN on the Maros-M\'esz\'aros test set, demonstrating state-of-the-art convergence performance in small-to-high-scale problems. The results highlight ODYN's superior warm-starting capabilities, which are critical in sequential and real-time settings common in robotics and AI. These advantages are further demonstrated by deploying ODYN as the backend of an SQP-based predictive control framework (OdynSQP), as the implicitly differentiable optimization layer for deep learning (ODYNLayer), and the optimizer of a contact-dynamics simulation (ODYNSim).
Authors: Adnan El Assadi, Isaac Chung, Chenghao Xiao, Roman Solomatin, Animesh Jha, Rahul Chand, Silky Singh, Kaitlyn Wang, Ali Sartaz Khan, Marc Moussa Nasser, Sufen Fong, Pengfei He, Alan Xiao, Ayush Sunil Munot, Aditya Shrivastava, Artem Gazizov, Niklas Muennighoff, Kenneth Enevoldsen
Abstract: We introduce the Massive Audio Embedding Benchmark (MAEB), a large-scale benchmark covering 30 tasks across speech, music, environmental sounds, and cross-modal audio-text reasoning in 100+ languages. We evaluate 50+ models and find that no single model dominates across all tasks: contrastive audio-text models excel at environmental sound classification (e.g., ESC50) but score near random on multilingual speech tasks (e.g., SIB-FLEURS), while speech-pretrained models show the opposite pattern. Clustering remains challenging for all models, with even the best-performing model achieving only modest results. We observe that models excelling on acoustic understanding often perform poorly on linguistic tasks, and vice versa. We also show that the performance of audio encoders on MAEB correlates highly with their performance when used in audio large language models. MAEB is derived from MAEB+, a collection of 98 tasks. MAEB is designed to maintain task diversity while reducing evaluation cost, and it integrates into the MTEB ecosystem for unified evaluation across text, image, and audio modalities. We release MAEB and all 98 tasks along with code and a leaderboard at https://github.com/embeddings-benchmark/mteb.
Authors: Ahmad Elallaf, Yu Zhang, Yuktha Priya Masupalli, Jeong Yang, Young Lee, Zechun Cao, Gongbo Liang
Abstract: Vision-language foundation models have emerged as powerful general-purpose representation learners with strong potential for multimodal understanding, but their deterministic embeddings often fail to provide the reliability required for high-stakes biomedical applications. This work introduces MedProbCLIP, a probabilistic vision-language learning framework for chest X-ray and radiology report representation learning and bidirectional retrieval. MedProbCLIP models image and text representations as Gaussian embeddings through a probabilistic contrastive objective that explicitly captures uncertainty and many-to-many correspondences between radiographs and clinical narratives. A variational information bottleneck mitigates overconfident predictions, while MedProbCLIP employs multi-view radiograph encoding and multi-section report encoding during training to provide fine-grained supervision for clinically aligned correspondence, yet requires only a single radiograph and a single report at inference. Evaluated on the MIMIC-CXR dataset, MedProbCLIP outperforms deterministic and probabilistic baselines, including CLIP, CXR-CLIP, and PCME++, in both retrieval and zero-shot classification. Beyond accuracy, MedProbCLIP demonstrates superior calibration, risk-coverage behavior, selective retrieval reliability, and robustness to clinically relevant corruptions, underscoring the value of probabilistic vision-language modeling for improving the trustworthiness and safety of radiology image-text retrieval systems.
Authors: Ruiwei Xiao, Runlong Ye, Xinying Hou, Jessica Wen, Harsh Kumar, Michael Liut, John Stamper
Abstract: Despite universal GenAI adoption, students cannot distinguish task performance from actual learning and lack skills to leverage AI for learning, leading to worse exam performance when AI use remains unreflective. Yet few interventions teaching students to prompt AI as a tutor rather than solution provider have been validated at scale through randomized controlled trials (RCTs). To bridge this gap, we conducted a semester-long RCT (N=979) with four ICAP framework-based instructional conditions varying in engagement intensity with a pre-test, immediate and delayed post-test and surveys. Mixed methods analysis results showed: (1) All conditions significantly improved prompting skills, with gains increasing progressively from Condition 1 to Condition 4, validating ICAP's cognitive engagement hierarchy; (2) for students with similar pre-test scores, higher learning gain in immediate post-test predict higher final exam score, though no direct between-group differences emerged; (3) Our interventions are suitable and scalable solutions for diverse educational contexts, resources and learners. Together, this study makes empirical and theoretical contributions: (1) theoretically, we provided one of the first large-scale RCTs examining how cognitive engagement shapes learning in prompting literacy and clarifying the relationship between learning-oriented prompting skills and broader academic performance; (2) empirically, we offered timely design guidance for transforming GenAI classroom policies into scalable, actionable prompting literacy instruction to advance learning in the era of Generative AI.
Authors: KC Santosh, Srikanth Baride, Rodrigue Rizk
Abstract: As machine learning (ML) continues its rapid expansion, the environmental cost of model training and inference has become a critical societal concern. Existing benchmarks overwhelmingly focus on standard performance metrics such as accuracy, BLEU, or mAP, while largely ignoring energy consumption and carbon emissions. This single-objective evaluation paradigm is increasingly misaligned with the practical requirements of large-scale deployment, particularly in energy-constrained environments such as mobile devices, developing regions, and climate-aware enterprises. In this paper, we propose AI-CARE, an evaluation tool for reporting energy consumption, and carbon emissions of ML models. In addition, we introduce the carbon-performance tradeoff curve, an interpretable tool that visualizes the Pareto frontier between performance and carbon cost. We demonstrate, through theoretical analysis and empirical validation on representative ML workloads, that carbon-aware benchmarking changes the relative ranking of models and encourages architectures that are simultaneously accurate and environmentally responsible. Our proposal aims to shift the research community toward transparent, multi-objective evaluation and align ML progress with global sustainability goals. The tool and documentation are available at https://github.com/USD-AI-ResearchLab/ai-care.
Authors: Kevin Wang, Hongqian Niu, Didong Li
Abstract: Generative Artificial Intelligence (AI), such as large language models (LLMs), has become a transformative force across science, industry, and society. As these systems grow in popularity, web data becomes increasingly interwoven with this AI-generated material and it is increasingly difficult to separate them from naturally generated content. As generative models are updated regularly, later models will inevitably be trained on mixtures of human-generated data and AI-generated data from earlier versions, creating a recursive training process with data contamination. Existing theoretical work has examined only highly simplified settings, where both the real data and the generative model are discrete or Gaussian, where it has been shown that such recursive training leads to model collapse. However, real data distributions are far more complex, and modern generative models are far more flexible than Gaussian and linear mechanisms. To fill this gap, we study recursive training in a general framework with minimal assumptions on the real data distribution and allow the underlying generative model to be a general universal approximator. In this framework, we show that contaminated recursive training still converges, with a convergence rate equal to the minimum of the baseline model's convergence rate and the fraction of real data used in each iteration. To the best of our knowledge, this is the first (positive) theoretical result on recursive training without distributional assumptions on the data. We further extend the analysis to settings where sampling bias is present in data collection and support all theoretical results with empirical studies.
Authors: Chenda Duan, Yipeng Zhang, Sotaro Kanai, Yuanyi Ding, Atsuro Daida, Pengyue Yu, Tiancheng Zheng, Naoto Kuroda, Shaun A. Hussain, Eishi Asano, Hiroki Nariai, Vwani Roychowdhury
Abstract: Epilepsy affects over 50 million people worldwide, and one-third of patients suffer drug-resistant seizures where surgery offers the best chance of seizure freedom. Accurate localization of the epileptogenic zone (EZ) relies on intracranial EEG (iEEG). Clinical workflows, however, remain constrained by labor-intensive manual review. At the same time, existing data-driven approaches are typically developed on single-center datasets that are inconsistent in format and metadata, lack standardized benchmarks, and rarely release pathological event annotations, creating barriers to reproducibility, cross-center validation, and clinical relevance. With extensive efforts to reconcile heterogeneous iEEG formats, metadata, and recordings across publicly available sources, we present $\textbf{Omni-iEEG}$, a large-scale, pre-surgical iEEG resource comprising $\textbf{302 patients}$ and $\textbf{178 hours}$ of high-resolution recordings. The dataset includes harmonized clinical metadata such as seizure onset zones, resections, and surgical outcomes, all validated by board-certified epileptologists. In addition, Omni-iEEG provides over 36K expert-validated annotations of pathological events, enabling robust biomarker studies. Omni-iEEG serves as a bridge between machine learning and epilepsy research. It defines clinically meaningful tasks with unified evaluation metrics grounded in clinical priors, enabling systematic evaluation of models in clinically relevant settings. Beyond benchmarking, we demonstrate the potential of end-to-end modeling on long iEEG segments and highlight the transferability of representations pretrained on non-neurophysiological domains. Together, these contributions establish Omni-iEEG as a foundation for reproducible, generalizable, and clinically translatable epilepsy research. The project page with dataset and code links is available at omni-ieeg.github.io/omni-ieeg.
Authors: Kevin Kai-Chun Chang, Ekin Beyazit, Alberto Sangiovanni-Vincentelli, Tichakorn Wongpiromsarn, Sanjit A. Seshia
Abstract: Developing autonomous driving systems for complex traffic environments requires balancing multiple objectives, such as avoiding collisions, obeying traffic rules, and making efficient progress. In many situations, these objectives cannot be satisfied simultaneously, and explicit priority relations naturally arise. Also, driving rules require context, so it is important to formally model the environment scenarios within which such rules apply. Existing benchmarks for evaluating autonomous vehicles lack such combinations of multi-objective prioritized rules and formal environment models. In this work, we introduce ScenicRules, a benchmark for evaluating autonomous driving systems in stochastic environments under prioritized multi-objective specifications. We first formalize a diverse set of objectives to serve as quantitative evaluation metrics. Next, we design a Hierarchical Rulebook framework that encodes multiple objectives and their priority relations in an interpretable and adaptable manner. We then construct a compact yet representative collection of scenarios spanning diverse driving contexts and near-accident situations, formally modeled in the Scenic language. Experimental results show that our formalized objectives and Hierarchical Rulebooks align well with human driving judgments and that our benchmark effectively exposes agent failures with respect to the prioritized objectives. Our benchmark can be accessed at https://github.com/BerkeleyLearnVerify/ScenicRules/.
Authors: Sean Trott, Samuel Taylor, Cameron Jones, James A. Michaelov, Pamela D. Rivi\`ere
Abstract: Research on mental state reasoning in language models (LMs) has the potential to inform theories of human social cognition--such as the theory that mental state reasoning emerges in part from language exposure--and our understanding of LMs themselves. Yet much published work on LMs relies on a relatively small sample of closed-source LMs, limiting our ability to rigorously test psychological theories and evaluate LM capacities. Here, we replicate and extend published work on the false belief task by assessing LM mental state reasoning behavior across 41 open-weight models (from distinct model families). We find sensitivity to implied knowledge states in 34% of the LMs tested; however, consistent with prior work, none fully ``explain away'' the effect in humans. Larger LMs show increased sensitivity and also exhibit higher psychometric predictive power. Finally, we use LM behavior to generate and test a novel hypothesis about human cognition: both humans and LMs show a bias towards attributing false beliefs when knowledge states are cued using a non-factive verb (``John thinks...'') than when cued indirectly (``John looks in the...''). Unlike the primary effect of knowledge states, where human sensitivity exceeds that of LMs, the magnitude of the human knowledge cue effect falls squarely within the distribution of LM effect sizes-suggesting that distributional statistics of language can in principle account for the latter but not the former in humans. These results demonstrate the value of using larger samples of open-weight LMs to test theories of human cognition and evaluate LM capacities.
Authors: Shankar Padmanabhan, Mustafa Omer Gul, Tanya Goyal
Abstract: Post-training endows pretrained LLMs with a variety of desirable skills, including instruction-following, reasoning, and others. However, these post-trained LLMs only encode knowledge up to a cut-off date, necessitating continual adaptation. Unfortunately, existing solutions cannot simultaneously learn new knowledge from an adaptation document corpora and mitigate the forgetting of earlier learned capabilities. To address this, we introduce Distillation via Split Contexts (DiSC), a simple context-distillation based approach for continual knowledge adaptation. \methodname~derives student and teacher distributions by conditioning on distinct segments of the training example and minimizes the KL divergence between the shared tokens. This allows us to efficiently apply context-distillation without requiring explicit generation steps during training. We run experiments on four post-trained models and two adaptation domains. Compared to prior finetuning and distillation methods for continual adaptation, DiSC consistently reports the best trade-off between learning new knowledge and mitigating forgetting of previously learned skills like instruction-following, reasoning, and factual knowledge.
Authors: Srikumar Nayak, James Walmesley
Abstract: Cross-border insider threats pose a critical challenge to government financial schemes, particularly when dealing with distributed, privacy-sensitive data across multiple jurisdictions. Existing approaches face fundamental limitations: they cannot effectively share intelligence across borders due to privacy constraints, lack reasoning capabilities to understand complex multi-step attack patterns, and fail to capture intricate graph-structured relationships in financial networks. We introduce FedGraph-AGI, a novel federated learning framework integrating Artificial General Intelligence (AGI) reasoning with graph neural networks for privacy-preserving cross-border insider threat detection. Our approach combines: (1) federated graph neural networks preserving data sovereignty; (2) Mixture-of-Experts (MoE) aggregation for heterogeneous jurisdictions; and (3) AGI-powered reasoning via Large Action Models (LAM) performing causal inference over graph data. Through experiments on a 50,000-transaction dataset across 10 jurisdictions, FedGraph-AGI achieves 92.3% accuracy, significantly outperforming federated baselines (86.1%) and centralized approaches (84.7%). Our ablation studies reveal AGI reasoning contributes 6.8% improvement, while MoE adds 4.4%. The system maintains epsilon = 1.0 differential privacy while achieving near-optimal performance and scales efficiently to 50+ clients. This represents the first integration of AGI reasoning with federated graph learning for insider threat detection, opening new directions for privacy-preserving cross-border intelligence sharing.
Authors: Tianwei Lin, Zhongwei Qiu, Wenqiao Zhang, Jiang Liu, Yihan Xie, Mingjian Gao, Zhenxuan Fan, Zhaocheng Li, Sijing Li, Zhongle Xie, Peng LU, Yueting Zhuang, Yingda Xia, Ling Zhang, Beng Chin Ooi
Abstract: Computed Tomography (CT) is one of the most widely used and diagnostically information-dense imaging modalities, covering critical organs such as the heart, lungs, liver, and colon. Clinical interpretation relies on both slice-driven local features (e.g., sub-centimeter nodules, lesion boundaries) and volume-driven spatial representations (e.g., tumor infiltration, inter-organ anatomical relations). However, existing Large Vision-Language Models (LVLMs) remain fragmented in CT slice versus volumetric understanding: slice-driven LVLMs show strong generalization but lack cross-slice spatial consistency, while volume-driven LVLMs explicitly capture volumetric semantics but suffer from coarse granularity and poor compatibility with slice inputs. The absence of a unified modeling paradigm constitutes a major bottleneck for the clinical translation of medical LVLMs. We present OmniCT, a powerful unified slice-volume LVLM for CT scenarios, which makes three contributions: (i) Spatial Consistency Enhancement (SCE): volumetric slice composition combined with tri-axial positional embedding that introduces volumetric consistency, and an MoE hybrid projection enables efficient slice-volume adaptation; (ii) Organ-level Semantic Enhancement (OSE): segmentation and ROI localization explicitly align anatomical regions, emphasizing lesion- and organ-level semantics; (iii) MedEval-CT: the largest slice-volume CT dataset and hybrid benchmark integrates comprehensive metrics for unified evaluation. OmniCT consistently outperforms existing methods with a substantial margin across diverse clinical tasks and satisfies both micro-level detail sensitivity and macro-level spatial reasoning. More importantly, it establishes a new paradigm for cross-modal medical imaging understanding.
Authors: Zehao Xu, Tony Paek, Kevin O'Sullivan, Attila Dobi
Abstract: Online media platforms often need to measure how frequently users are exposed to specific content attributes in order to evaluate trade-offs in A/B experiments. A direct approach is to sample content, label it using a high-quality rubric (e.g., an expert-reviewed LLM prompt), and estimate impression-weighted prevalence. However, repeatedly running such labeling for every experiment arm and segment is too costly and slow to serve as a default measurement at scale. We present a scalable \emph{surrogate-based prevalence measurement} framework that decouples expensive labeling from per-experiment evaluation. The framework calibrates a surrogate signal to reference labels offline and then uses only impression logs to estimate prevalence for arbitrary experiment arms and segments. We instantiate this framework using \emph{score bucketing} as the surrogate: we discretize a model score into buckets, estimate bucket-level prevalences from an offline labeled sample, and combine these calibrated bucket level prevalences with the bucket distribution of impressions in each arm to obtain fast, log-based estimates. Across multiple large-scale A/B tests, we validate that the surrogate estimates closely match the reference estimates for both arm-level prevalence and treatment--control deltas. This enables scalable, low-latency prevalence measurement in experimentation without requiring per-experiment labeling jobs.
Authors: Jiang Zhang, Yubo Wang, Wei Chang, Lu Han, Xingying Cheng, Feng Zhang, Min Li, Songhao Jiang, Wei Zheng, Harry Tran, Zhen Wang, Lei Chen, Yueming Wang, Benyu Zhang, Xiangjun Fan, Bi Xue, Qifan Wang
Abstract: Approximate nearest neighbor (ANN) search is widely used in the retrieval stage of large-scale recommendation systems. In this stage, candidate items are indexed using their learned embedding vectors, and ANN search is executed for each user (or item) query to retrieve a set of relevant items. However, ANN-based retrieval has two key limitations. First, item embeddings and their indices are typically learned in separate stages: indexing is often performed offline after embeddings are trained, which can yield suboptimal retrieval quality-especially for newly created items. Second, although ANN offers sublinear query time, it must still be run for every request, incurring substantial computation cost at industry scale. In this paper, we propose MultiFaceted Learnable Index (MFLI), a scalable, real-time retrieval paradigm that learns multifaceted item embeddings and indices within a unified framework and eliminates ANN search at serving time. Specifically, we construct a multifaceted hierarchical codebook via residual quantization of item embeddings and co-train the codebook with the embeddings. We further introduce an efficient multifaceted indexing structure and mechanisms that support real-time updates. At serving time, the learned hierarchical indices are used directly to identify relevant items, avoiding ANN search altogether. Extensive experiments on real-world data with billions of users show that MFLI improves recall on engagement tasks by up to 11.8\%, cold-content delivery by up to 57.29\%, and semantic relevance by 13.5\% compared with prior state-of-the-art methods. We also deploy MFLI in the system and report online experimental results demonstrating improved engagement, less popularity bias, and higher serving efficiency.
Authors: Hongyeon Yu, Dongchan Kim, Young-Bum Kim
Abstract: The rapid proliferation of AI-generated content on the Web presents a structural risk to information retrieval, as search engines and Retrieval-Augmented Generation (RAG) systems increasingly consume evidence produced by the Large Language Models (LLMs). We characterize this ecosystem-level failure mode as Retrieval Collapse, a two-stage process where (1) AI-generated content dominates search results, eroding source diversity, and (2) low-quality or adversarial content infiltrates the retrieval pipeline. We analyzed this dynamic through controlled experiments involving both high-quality SEO-style content and adversarially crafted content. In the SEO scenario, a 67\% pool contamination led to over 80\% exposure contamination, creating a homogenized yet deceptively healthy state where answer accuracy remains stable despite the reliance on synthetic sources. Conversely, under adversarial contamination, baselines like BM25 exposed $\sim$19\% of harmful content, whereas LLM-based rankers demonstrated stronger suppression capabilities. These findings highlight the risk of retrieval pipelines quietly shifting toward synthetic evidence and the need for retrieval-aware strategies to prevent a self-reinforcing cycle of quality decline in Web-grounded systems.
Authors: Wooyoung Jung, Kahyun Jeon, Prosper Babon-Ayeng
Abstract: This study aimed to comprehend how user domain knowledge and artificial intelligence (AI) literacy impact the effective use of human-AI interactive building energy management system (BEMS). While prior studies have investigated the potential of integrating large language models (LLMs) into BEMS or building energy modeling, very few studies have examined how user interact with such systems. We conducted a systematic role-playing experiment, where 85 human subjects interacted with an advanced generative pre-trained transformer (OpenAI GPT-4o). Participants were tasked with identifying the top five behavioral changes that could reduce home energy use with the GPT model that functioned as an LLM-integrated BEMS. Then, the collected prompt-response data and participant conclusions were analyzed using an analytical framework that hierarchically assessed and scored human-AI interactions and their home energy analysis approaches. Also, participants were classified into four groups based on their self-evaluated domain knowledge of building energy use and AI literacy, and Kruskal-Wallis H tests with post-hoc pairwise comparisons were conducted across 20 quantifiable metrics. Key takeaways include: most participants employed concise prompts (median: 16.2 words) and relied heavily on GPT's analytical capabilities; and notably, only 1 of 20 metrics, appliance identification rate, showed statistically significant group differences (p=0.037), driven by AI literacy rather than domain knowledge, suggesting an equalizing effect of LLMs across expertise levels. This study provides foundational insights into human-AI collaboration dynamics and promising development directions in the context of LLM-integrated BEMS and contributes to realizing human-centric LLM-integrated energy systems.
Authors: Megan Lee, Seung Ha Hwang, Inhyeok Choi, Shreyas Darade, Mengchun Zhang, Kateryna Shapovalenko
Abstract: Cross-subject generalization in EEG-based brain-computer interfaces (BCIs) remains challenging due to individual variability in neural signals. We investigate whether spectral representations offer more stable features for cross-subject transfer than temporal waveforms. Through correlation analyses across three EEG paradigms (SSVEP, P300, and Motor Imagery), we find that spectral features exhibit consistently higher cross-subject similarity than temporal signals. Motivated by this observation, we introduce ASPEN, a hybrid architecture that combines spectral and temporal feature streams via multiplicative fusion, requiring cross-modal agreement for features to propagate. Experiments across six benchmark datasets reveal that ASPEN is able to dynamically achieve the optimal spectral-temporal balance depending on the paradigm. ASPEN achieves the best unseen-subject accuracy on three of six datasets and competitive performance on others, demonstrating that multiplicative multimodal fusion enables effective cross-subject generalization.
Authors: Nithin Sivakumaran, Shoubin Yu, Hyunji Lee, Yue Zhang, Ali Payani, Mohit Bansal, Elias Stengel-Eskin
Abstract: Chain-of-thought (CoT) reasoning sometimes fails to faithfully reflect the true computation of a large language model (LLM), hampering its utility in explaining how LLMs arrive at their answers. Moreover, optimizing for faithfulness and interpretability in reasoning often degrades task performance. To address this tradeoff and improve CoT faithfulness, we propose Reasoning Execution by Multiple Listeners (REMUL), a multi-party reinforcement learning approach. REMUL builds on the hypothesis that reasoning traces which other parties can follow will be more faithful. A speaker model generates a reasoning trace, which is truncated and passed to a pool of listener models who "execute" the trace, continuing the trace to an answer. Speakers are rewarded for producing reasoning that is clear to listeners, with additional correctness regularization via masked supervised finetuning to counter the tradeoff between faithfulness and performance. On multiple reasoning benchmarks (BIG-Bench Extra Hard, MuSR, ZebraLogicBench, and FOLIO), REMUL consistently and substantially improves three measures of faithfulness -- hint attribution, early answering area over the curve (AOC), and mistake injection AOC -- while also improving accuracy. Our analysis finds that these gains are robust across training domains, translate to legibility gains, and are associated with shorter and more direct CoTs.
Authors: Jiangweizhi Peng, Yuanxin Liu, Ruida Zhou, Charles Fleming, Zhaoran Wang, Alfredo Garcia, Mingyi Hong
Abstract: Training LLMs as interactive agents for multi-turn decision-making remains challenging, particularly in long-horizon tasks with sparse and delayed rewards, where agents must execute extended sequences of actions before receiving meaningful feedback. Most existing reinforcement learning (RL) approaches model LLM agents as flat policies operating at a single time scale, selecting one action at each turn. In sparse-reward settings, such flat policies must propagate credit across the entire trajectory without explicit temporal abstraction, which often leads to unstable optimization and inefficient credit assignment. We propose HiPER, a novel Hierarchical Plan-Execute RL framework that explicitly separates high-level planning from low-level execution. HiPER factorizes the policy into a high-level planner that proposes subgoals and a low-level executor that carries them out over multiple action steps. To align optimization with this structure, we introduce a key technique called hierarchical advantage estimation (HAE), which carefully assigns credit at both the planning and execution levels. By aggregating returns over the execution of each subgoal and coordinating updates across the two levels, HAE provides an unbiased gradient estimator and provably reduces variance compared to flat generalized advantage estimation. Empirically, HiPER achieves state-of-the-art performance on challenging interactive benchmarks, reaching 97.4\% success on ALFWorld and 83.3\% on WebShop with Qwen2.5-7B-Instruct (+6.6\% and +8.3\% over the best prior method), with especially large gains on long-horizon tasks requiring multiple dependent subtasks. These results highlight the importance of explicit hierarchical decomposition for scalable RL training of multi-turn LLM agents.
Authors: Fatih Temiz, Shavbo Salehi, Melike Erol-Kantarci
Abstract: Mobile edge computing (MEC) based wireless metaverse services offer an untethered, immersive experience to users, where the superior quality of experience (QoE) needs to be achieved under stringent latency constraints and visual quality demands. To achieve this, MEC-based intelligent resource allocation for virtual reality users needs to be supported by coordination across MEC servers to harness distributed data. Federated learning (FL) is a promising solution, and can be combined with reinforcement learning (RL) to develop generalized policies across MEC-servers. However, conventional FL incurs transmitting the full model parameters across the MEC-servers and the cloud, and suffer performance degradation due to naive global aggregation, especially in heterogeneous multi-radio access technology environments. To address these challenges, this paper proposes Federated Split Decision Transformer (FSDT), an offline RL framework where the transformer model is partitioned between MEC servers and the cloud. Agent-specific components (e.g., MEC-based embedding and prediction layers) enable local adaptability, while shared global layers in the cloud facilitate cooperative training across MEC servers. Experimental results demonstrate that FSDT enhances QoE for up to 10% in heterogeneous environments compared to baselines, while offloadingnearly 98% of the transformer model parameters to the cloud, thereby reducing the computational burden on MEC servers.
Authors: Binchuan Qi
Abstract: In this work, we propose a notion of practical learnability grounded in finite sample settings, and develop a conjugate learning theoretical framework based on convex conjugate duality to characterize this learnability property. Building on this foundation, we demonstrate that training deep neural networks (DNNs) with mini-batch stochastic gradient descent (SGD) achieves global optima of empirical risk by jointly controlling the extreme eigenvalues of a structure matrix and the gradient energy, and we establish a corresponding convergence theorem. We further elucidate the impact of batch size and model architecture (including depth, parameter count, sparsity, skip connections, and other characteristics) on non-convex optimization. Additionally, we derive a model-agnostic lower bound for the achievable empirical risk, theoretically demonstrating that data determines the fundamental limit of trainability. On the generalization front, we derive deterministic and probabilistic bounds on generalization error based on generalized conditional entropy measures. The former explicitly delineates the range of generalization error, while the latter characterizes the distribution of generalization error relative to the deterministic bounds under independent and identically distributed (i.i.d.) sampling conditions. Furthermore, these bounds explicitly quantify the influence of three key factors: (i) information loss induced by irreversibility in the model, (ii) the maximum attainable loss value, and (iii) the generalized conditional entropy of features with respect to labels. Moreover, they offer a unified theoretical lens for understanding the roles of regularization, irreversible transformations, and network depth in shaping the generalization behavior of deep neural networks. Extensive experiments validate all theoretical predictions, confirming the framework's correctness and consistency.
Authors: Zirui Zang, Ahmad Amine, Nick-Marios T. Kokolakis, Truong X. Nghiem, Ugo Rosolia, Rahul Mangharam
Abstract: Robots executing iterative tasks in complex, uncertain environments require control strategies that balance robustness, safety, and high performance. This paper introduces a safe information-theoretic learning model predictive control (SIT-LMPC) algorithm for iterative tasks. Specifically, we design an iterative control framework based on an information-theoretic model predictive control algorithm to address a constrained infinite-horizon optimal control problem for discrete-time nonlinear stochastic systems. An adaptive penalty method is developed to ensure safety while balancing optimality. Trajectories from previous iterations are utilized to learn a value function using normalizing flows, which enables richer uncertainty modeling compared to Gaussian priors. SIT-LMPC is designed for highly parallel execution on graphics processing units, allowing efficient real-time optimization. Benchmark simulations and hardware experiments demonstrate that SIT-LMPC iteratively improves system performance while robustly satisfying system constraints.
Authors: Namkyung Yoon, Kyeonghyun Yoo, Wooyong Jung, Sanghong Kim, Hwangnam Kim
Abstract: Despite the continuous research and evolution of language models, they sometimes underperform previous versions. Existing approaches to overcome these challenges are resource-intensive, highlighting the need for alternatives that enable immediate action. We assume that each language model has a local module inside that is suitable for a specific function. First, this work identifies a set of modules showing consistent and local activation changes under an inference workload through activation-based analysis. Subsequently, we transplant an internal module that is properly activated for a specific task into the target model, leading to immediate and measurable functional changes without additional training or fine-tuning. To experimentally demonstrate the effectiveness of the transplant technique, we quantify the relationship between transplant strength and performance improvement under different conditions for two language models. In the cross-generation setting, we find that transplanting activation-selected modules can substantially improve the underperforming model, reaching up to twice the target baseline and achieving gap-based recovery above 100%. Moreover, in transplant experiments between a base model and its instruction-tuned counterpart, transplantation improves the underperforming model toward the stronger baseline, yielding up to about 2.33 times the target baseline with gap-based recovery reaching up to 100% in the best case. These results show that meaningful capacity transfer can be realized through the implantation of highly localized modules implied by language models. Overall, this work provides empirical evidence for task-localized modularity in language models and presents a new research area: model transplantation.
Authors: Zhenzhen Huang, Haoyu Bian, Jiaquan Zhang, Yibei Liu, Kuien Liu, Caiyan Qin, Guoqing Wang, Yang Yang, Chaoning Zhang
Abstract: Several complex physical systems are governed by multi-scale partial differential equations (PDEs) that exhibit both smooth low-frequency components and localized high-frequency structures. Existing physics-informed neural network (PINN) methods typically train with fixed coordinate system inputs, where geometric misalignment with these structures induces gradient stiffness and ill-conditioning that hinder convergence. To address this issue, we introduce a mapping paradigm that reshapes the input coordinates through differentiable geometric compactification mappings and couples the geometric structure of PDEs with the spectral properties of residual operators. Based on this paradigm, we propose Geometric Compactification (GC)-PINN, a framework that introduces three mapping strategies for periodic boundaries, far-field scale expansion, and localized singular structures in the input domain without modifying the underlying PINN architecture. Extensive empirical evaluation demonstrates that this approach yields more uniform residual distributions and higher solution accuracy on representative 1D and 2D PDEs, while improving training stability and convergence speed.
Authors: Yusuf Hakan Kalayci, Evi Micha
Abstract: Permanent citizens' assemblies are ongoing deliberative bodies composed of randomly selected citizens, organized into panels that rotate over time. Unlike one-off panels, which represent the population in a single snapshot, permanent assemblies enable shifting participation across multiple rounds. This structure offers a powerful framework for ensuring that different groups of individuals are represented over time across successive panels. In particular, it allows smaller groups of individuals that may not warrant representation in every individual panel to be represented across a sequence of them. We formalize this temporal sortition framework by requiring proportional representation both within each individual panel and across the sequence of panels. Building on the work of Ebadian and Micha (2025), we consider a setting in which the population lies in a metric space, and the goal is to achieve both proportional representation, ensuring that every group of citizens receives adequate representation, and individual fairness, ensuring that each individual has an equal probability of being selected. We extend the notion of representation to a temporal setting by requiring that every initial segment of the panel sequence, viewed as a cumulative whole, proportionally reflects the structure of the population. We present algorithms that provide varying guarantees of proportional representation, both within individual panels and across any sequence of panels, while also maintaining individual fairness over time.
Authors: Emile Anand, Richard Hoffmann, Sarah Liaw, Adam Wierman
Abstract: Coordinating large populations of interacting agents is a central challenge in multi-agent reinforcement learning (MARL), where the size of the joint state-action space scales exponentially with the number of agents. Mean-field methods alleviate this burden by aggregating agent interactions, but these approaches assume homogeneous interactions. Recent graphon-based frameworks capture heterogeneity, but are computationally expensive as the number of agents grows. Therefore, we introduce $\texttt{GMFS}$, a $\textbf{G}$raphon $\textbf{M}$ean-$\textbf{F}$ield $\textbf{S}$ubsampling framework for scalable cooperative MARL with heterogeneous agent interactions. By subsampling $\kappa$ agents according to interaction strength, we approximate the graphon-weighted mean-field and learn a policy with sample complexity $\mathrm{poly}(\kappa)$ and optimality gap $O(1/\sqrt{\kappa})$. We verify our theory with numerical simulations in robotic coordination, showing that $\texttt{GMFS}$ achieves near-optimal performance.
Authors: Sanket Badhe, Deep Shah, Nehal Kathrotia
Abstract: Large language models (LLMs) are trained on web-scale corpora that exhibit steep power-law distributions, in which the distribution of knowledge is highly long-tailed, with most appearing infrequently. While scaling has improved average-case performance, persistent failures on low-frequency, domain-specific, cultural, and temporal knowledge remain poorly characterized. This paper develops a structured taxonomy and analysis of long-Tail Knowledge in large language models, synthesizing prior work across technical and sociotechnical perspectives. We introduce a structured analytical framework that synthesizes prior work across four complementary axes: how long-Tail Knowledge is defined, the mechanisms by which it is lost or distorted during training and inference, the technical interventions proposed to mitigate these failures, and the implications of these failures for fairness, accountability, transparency, and user trust. We further examine how existing evaluation practices obscure tail behavior and complicate accountability for rare but consequential failures. The paper concludes by identifying open challenges related to privacy, sustainability, and governance that constrain long-Tail Knowledge representation. Taken together, this paper provides a unifying conceptual framework for understanding how long-Tail Knowledge is defined, lost, evaluated, and manifested in deployed language model systems.
Authors: Jiaquan Zhang, Fachrina Dewi Puspitasari, Songbo Zhang, Yibei Liu, Kuien Liu, Caiyan Qin, Fan Mo, Peng Wang, Yang Yang, Chaoning Zhang
Abstract: Neural operators offer an effective framework for learning solutions of partial differential equations for many physical systems in a resolution-invariant and data-driven manner. Existing neural operators, however, often suffer from instability in multi-layer iteration and long-horizon rollout, which stems from the unconstrained Euclidean latent space updates that violate the geometric and conservation laws. To address this challenge, we propose to constrain manifolds with low-rank Lie algebra parameterization that performs group action updates on the latent representation. Our method, termed Manifold Constraining based on Lie group (MCL), acts as an efficient \emph{plug-and-play} module that enforces geometric inductive bias to existing neural operators. Extensive experiments on various partial differential equations, such as 1-D Burgers and 2-D Navier-Stokes, over a wide range of parameters and steps demonstrate that our method effectively lowers the relative prediction error by 30-50\% at the cost of 2.26\% of parameter increase. The results show that our approach provides a scalable solution for improving long-term prediction fidelity by addressing the principled geometric constraints absent in the neural operator updates.
Authors: Ruibiao Zhu
Abstract: This paper introduces a novel approach to sea ice modeling using Graph Neural Networks (GNNs), utilizing the natural graph structure of sea ice, where nodes represent individual ice pieces, and edges model the physical interactions, including collisions. This concept is developed within a one-dimensional framework as a foundational step. Traditional numerical methods, while effective, are computationally intensive and less scalable. By utilizing GNNs, the proposed model, termed the Collision-captured Network (CN), integrates data assimilation (DA) techniques to effectively learn and predict sea ice dynamics under various conditions. The approach was validated using synthetic data, both with and without observed data points, and it was found that the model accelerates the simulation of trajectories without compromising accuracy. This advancement offers a more efficient tool for forecasting in marginal ice zones (MIZ) and highlights the potential of combining machine learning with data assimilation for more effective and efficient modeling.
Authors: Hamzeh Asgharnezhad, Pegah Tabarisaadi, Abbas Khosravi, Roohallah Alizadehsani, U. Rajendra Acharya
Abstract: Deep learning has improved automated electrocardiogram (ECG) classification, but limited insight into prediction reliability hinders its use in safety-critical settings. This paper proposes UCTECG-Net, an uncertainty-aware hybrid architecture that combines one-dimensional convolutions and Transformer encoders to process raw ECG signals and their spectrograms jointly. Evaluated on the MIT-BIH Arrhythmia and PTB Diagnostic datasets, UCTECG-Net outperforms LSTM, CNN1D, and Transformer baselines in terms of accuracy, precision, recall and F1 score, achieving up to 98.58% accuracy on MIT-BIH and 99.14% on PTB. To assess predictive reliability, we integrate three uncertainty quantification methods (Monte Carlo Dropout, Deep Ensembles, and Ensemble Monte Carlo Dropout) into all models and analyze their behavior using an uncertainty-aware confusion matrix and derived metrics. The results show that UCTECG-Net, particularly with Ensemble or EMCD, provides more reliable and better-aligned uncertainty estimates than competing architectures, offering a stronger basis for risk-aware ECG decision support.
Authors: Md. Najib Hasan, Touseef Hasan, Souvika Sarkar
Abstract: Large Language Models (LLMs) are increasingly used as automated annotators to scale dataset creation, yet their reliability as unbiased annotators--especially for low-resource and identity-sensitive settings--remains poorly understood. In this work, we study the behavior of LLMs as zero-shot annotators for Bangla hate speech, a task where even human agreement is challenging, and annotator bias can have serious downstream consequences. We conduct a systematic benchmark of 17 LLMs using a unified evaluation framework. Our analysis uncovers annotator bias and substantial instability in model judgments. Surprisingly, increased model scale does not guarantee improved annotation quality--smaller, more task-aligned models frequently exhibit more consistent behavior than their larger counterparts. These results highlight important limitations of current LLMs for sensitive annotation tasks in low-resource languages and underscore the need for careful evaluation before deployment.
Authors: Ryotaro Nagase, Ryoichi Takashima, Yoichi Yamashita
Abstract: Speech emotion recognition (SER) has traditionally relied on categorical or dimensional labels. However, this technique is limited in representing both the diversity and interpretability of emotions. To overcome this limitation, we focus on color attributes, such as hue, saturation, and value, to represent emotions as continuous and interpretable scores. We annotated an emotional speech corpus with color attributes via crowdsourcing and analyzed them. Moreover, we built regression models for color attributes in SER using machine learning and deep learning, and explored the multitask learning of color attribute regression and emotion classification. As a result, we demonstrated the relationship between color attributes and emotions in speech, and successfully developed color attribute regression models for SER. We also showed that multitask learning improved the performance of each task.
Authors: Fabian Walke, Veronika F\"oller
Abstract: This study investigates generative artificial intelligence (GenAI) usage of university students who study alongside their professional career. Previous literature has paid little attention to part-time students and the intersectional use of GenAI between education and business. This study examines with a grounded theory approach the characteristics of GenAI usage of part-time students. Eleven students from a distance learning university were interviewed. Three causal and four intervening conditions, as well as strategies were identified, to influence the use of GenAI. The study highlights both the potential and challenges of GenAI usage in education and business. While GenAI can significantly enhance productivity and learning outcomes, concerns about ethical implications, reliability, and the risk of academic misconduct persist. The developed grounded model offers a comprehensive understanding of GenAI usage among students, providing valuable insights for educators, policymakers, and developers of GenAI tools seeking to bridge the gap between education and business.
Authors: Jakub Breier, \v{S}tefan Ku\v{c}er\'ak, Xiaolu Hou
Abstract: Fault injection attacks on embedded neural network models have been shown as a potent threat. Numerous works studied resilience of models from various points of view. As of now, there is no comprehensive study that would evaluate the influence of number representations used for model parameters against electromagnetic fault injection (EMFI) attacks. In this paper, we investigate how four different number representations influence the success of an EMFI attack on embedded neural network models. We chose two common floating-point representations (32-bit, and 16-bit), and two integer representations (8-bit, and 4-bit). We deployed four common image classifiers, ResNet-18, ResNet-34, ResNet-50, and VGG-11, on an embedded memory chip, and utilized a low-cost EMFI platform to trigger faults. Our results show that while floating-point representations exhibit almost a complete degradation in accuracy (Top-1 and Top-5) after a single fault injection, integer representations offer better resistance overall. Especially, when considering the the 8-bit representation on a relatively large network (VGG-11), the Top-1 accuracies stay at around 70% and the Top-5 at around 90%.
Authors: Gabriele Barlacchi, Margherita Lalli, Emanuele Ferragina, Fosca Giannotti, Dino Pedreschi, Luca Pappalardo
Abstract: Recommender systems shape individual choices through feedback loops in which user behavior and algorithmic recommendations coevolve over time. The systemic effects of these loops remain poorly understood, in part due to unrealistic assumptions in existing simulation studies. We propose a feedback-loop model that captures implicit feedback, periodic retraining, probabilistic adoption of recommendations, and heterogeneous recommender systems. We apply the framework on online retail and music streaming data and analyze systemic effects of the feedback loop. We find that increasing recommender adoption may lead to a progressive diversification of individual consumption, while collective demand is redistributed in model- and domain-dependent ways, often amplifying popularity concentration. Temporal analyses further reveal that apparent increases in individual diversity observed in static evaluations are illusory: when adoption is fixed and time unfolds, individual diversity consistently decreases across all models. Our results highlight the need to move beyond static evaluations and explicitly account for feedback-loop dynamics when designing recommender systems.
Authors: Guy Bar-Shalom, Ami Tavory, Itay Evron, Maya Bechler-Speicher, Ido Guy, Haggai Maron
Abstract: Weight-space models learn directly from the parameters of neural networks, enabling tasks such as predicting their accuracy on new datasets. Naive methods -- like applying MLPs to flattened parameters -- perform poorly, making the design of better weight-space architectures a central challenge. While prior work leveraged permutation symmetries in standard networks to guide such designs, no analogous analysis or tailored architecture yet exists for Kolmogorov-Arnold Networks (KANs). In this work, we show that KANs share the same permutation symmetries as MLPs, and propose the KAN-graph, a graph representation of their computation. Building on this, we develop WS-KAN, the first weight-space architecture that learns on KANs, which naturally accounts for their symmetry. We analyze WS-KAN's expressive power, showing it can replicate an input KAN's forward pass - a standard approach for assessing expressiveness in weight-space architectures. We construct a comprehensive ``zoo'' of trained KANs spanning diverse tasks, which we use as benchmarks to empirically evaluate WS-KAN. Across all tasks, WS-KAN consistently outperforms structure-agnostic baselines, often by a substantial margin. Our code is available at https://github.com/BarSGuy/KAN-Graph-Metanetwork.
Authors: Santiago C. Vilabella, Pablo P\'erez-N\'u\~nez, Beatriz Remeseiro
Abstract: In the fast-evolving field of artificial intelligence, where models are increasingly growing in complexity and size, the availability of labeled data for training deep learning models has become a significant challenge. Addressing complex problems like object detection demands considerable time and resources for data labeling to achieve meaningful results. For companies developing such applications, this entails extensive investment in highly skilled personnel or costly outsourcing. This research work aims to demonstrate that enhancing feature extractors can substantially alleviate this challenge, enabling models to learn more effective representations with less labeled data. Utilizing a self-supervised learning strategy, we present a model trained on unlabeled data that outperforms state-of-the-art feature extractors pre-trained on ImageNet and particularly designed for object detection tasks. Moreover, the results demonstrate that our approach encourages the model to focus on the most relevant aspects of an object, thus achieving better feature representations and, therefore, reinforcing its reliability and robustness.
Authors: Joseph Bingham, Netanel Arussy, Saman Zonouz
Abstract: With the introduction of cyber-physical genome sequencing and editing technologies, such as CRISPR, researchers can more easily access tools to investigate and create remedies for a variety of topics in genetics and health science (e.g. agriculture and medicine). As the field advances and grows, new concerns present themselves in the ability to predict the off-target behavior. In this work, we explore the underlying biological and chemical model from a data driven perspective. Additionally, we present a machine learning based solution named \textit{Guide-Guard} to predict the behavior of the system given a gRNA in the CRISPR gene-editing process with 84\% accuracy. This solution is able to be trained on multiple different genes at the same time while retaining accuracy.
Authors: Arvind Krishna Sridhar, Yinyi Guo, Erik Visser
Abstract: Spatial audio understanding aims to enable machines to interpret complex auditory scenes, particularly when sound sources move over time. In this work, we study Spatial Audio Question Answering (Spatial AQA) with a focus on movement reasoning, where a model must infer object motion, position, and directional changes directly from stereo audio. First, we introduce a movement-centric spatial audio augmentation framework that synthesizes diverse motion patterns from isolated mono audio events, enabling controlled and scalable training data generation. Second, we propose an end-to-end multimodal finetuning approach with a thinking mode, which allows audio-language models to produce explicit intermediate reasoning steps before predicting an answer. Third, we investigate the impact of query-conditioned source separation as a preprocessing stage and compare three inference regimes: no masking, an audio grounding model (AGM), and ground-truth masks. Our results show that reasoning amplifies the benefits of source separation, with thinking mode showing significant improvement of +5.1% when a single event is present in the question. These findings highlight the interplay between movement modeling, reasoning, and separation quality, offering new insights for advancing spatial audio understanding.
Authors: Samira Nazari, Mohammad Saeed Almasi, Mahdi Taheri, Ali Azarpeyvand, Ali Mokhtari, Ali Mahani, Christian Herglotz
Abstract: This work presents HAWX, a hardware-aware scalable exploration framework that employs multi-level sensitivity scoring at different DNN abstraction levels (operator, filter, layer, and model) to guide selective integration of heterogeneous AxC blocks. Supported by predictive models for accuracy, power, and area, HAWX accelerates the evaluation of candidate configurations, achieving over 23* speedup in a layer-level search with two candidate approximate blocks and more than (3*106)* speedup at the filter-level search only for LeNet-5, while maintaining accuracy comparable to exhaustive search. Experiments across state-of-the-art DNN benchmarks such as VGG-11, ResNet-18, and EfficientNetLite demonstrate that the efficiency benefits of HAWX scale exponentially with network size. The HAWX hardware-aware search algorithm supports both spatial and temporal accelerator architectures, leveraging either off-the-shelf approximate components or customized designs.
Authors: Martin B\"uchner, Adrian R\"ofer, Tim Engelbracht, Tim Welschehold, Zuria Bauer, Hermann Blum, Marc Pollefeys, Abhinav Valada
Abstract: Semantics has enabled 3D scene understanding and affordance-driven object interaction. However, robots operating in real-world environments face a critical limitation: they cannot anticipate how objects move. Long-horizon mobile manipulation requires closing the gap between semantics, geometry, and kinematics. In this work, we present MoMa-SG, a novel framework for building semantic-kinematic 3D scene graphs of articulated scenes containing a myriad of interactable objects. Given RGB-D sequences containing multiple object articulations, we temporally segment object interactions and infer object motion using occlusion-robust point tracking. We then lift point trajectories into 3D and estimate articulation models using a novel unified twist estimation formulation that robustly estimates revolute and prismatic joint parameters in a single optimization pass. Next, we associate objects with estimated articulations and detect contained objects by reasoning over parent-child relations at identified opening states. We also introduce the novel Arti4D-Semantic dataset, which uniquely combines hierarchical object semantics including parent-child relation labels with object axis annotations across 62 in-the-wild RGB-D sequences containing 600 object interactions and three distinct observation paradigms. We extensively evaluate the performance of MoMa-SG on two datasets and ablate key design choices of our approach. In addition, real-world experiments on both a quadruped and a mobile manipulator demonstrate that our semantic-kinematic scene graphs enable robust manipulation of articulated objects in everyday home environments. We provide code and data at: https://momasg.cs.uni-freiburg.de.
Authors: Bin Cao, Qian Zhang, Zhenjie Feng, Taolue Zhang, Jiaqiang Huang, Lu-Tao Weng, Tong-Yi Zhang
Abstract: Artificial intelligence can rapidly propose candidate phases and structures from X-ray diffraction (XRD), but these hypotheses often fail in downstream refinement because peak intensities cannot be stably assigned under severe overlap and diffraction consistency is enforced only weakly. Here we introduce WPEM, a physics-constrained whole-pattern decomposition and refinement workflow that turns Bragg's law into an explicit constraint within a batch expectation--maximization framework. WPEM models the full profile as a probabilistic mixture density and iteratively infers component-resolved intensities while keeping peak centres Bragg-consistent, producing a continuous, physically admissible intensity representation that remains stable in heavily overlapped regions and in the presence of mixed radiation or multiple phases. We benchmark WPEM on standard reference patterns (\ce{PbSO4} and \ce{Tb2BaCoO5}), where it yields lower $R_{\mathrm{p}}$/$R_{\mathrm{wp}}$ than widely used packages (FullProf and TOPAS) under matched refinement conditions. We further demonstrate generality across realistic experimental scenarios, including phase-resolved decomposition of a multiphase Ti--15Nb thin film, quantitative recovery of \ce{NaCl}--\ce{Li2CO3} mixture compositions, separation of crystalline peaks from amorphous halos in semicrystalline polymers, high-throughput operando lattice tracking in layered cathodes, automated refinement of a compositionally disordered Ru--Mn oxide solid solution (CCDC 2530452), and quantitative phase-resolved deciphering of an ancient Egyptian make-up sample from synchrotron powder XRD. By providing Bragg-consistent, uncertainty-aware intensity partitioning as a refinement-ready interface, WPEM closes the gap between AI-generated hypotheses and diffraction-admissible structure refinement on challenging XRD data.
Authors: Ahmet Halici, Ece Tugba Cebeci, Musa Balci, Mustafa Cini, Serkan Sokmen
Abstract: Generating diagnostic text from histopathology whole slide images (WSIs) is challenging due to the gigapixel scale of the input and the requirement for precise, domain specific language. We propose a hierarchical vision language framework that combines a frozen pathology foundation model with a Transformer decoder for report generation. To make WSI processing tractable, we perform multi resolution pyramidal patch selection (downsampling factors 2^3 to 2^6) and remove background and artifacts using Laplacian variance and HSV based criteria. Patch features are extracted with the UNI Vision Transformer and projected to a 6 layer Transformer decoder that generates diagnostic text via cross attention. To better represent biomedical terminology, we tokenize the output using BioGPT. Finally, we add a retrieval based verification step that compares generated reports with a reference corpus using Sentence BERT embeddings; if a high similarity match is found, the generated report is replaced with the retrieved ground truth reference to improve reliability.
Authors: Ali Faraz, Raja Kolla, Ashish Kulkarni, Shubham Agarwal
Abstract: Designing Optical Character Recognition (OCR) systems for India requires balancing linguistic diversity, document heterogeneity, and deployment constraints. In this paper, we study two training strategies for building multilingual OCR systems with Vision-Language Models through the Chitrapathak series. We first follow a popular multimodal approach, pairing a generic vision encoder with a strong multilingual language model and training the system end-to-end for OCR. Alternatively, we explore fine-tuning an existing OCR model, despite not being trained for the target languages. Through extensive evaluation on multilingual Indic OCR benchmarks and deployment-oriented metrics, we find that the second strategy consistently achieves better accuracy-latency trade-offs. Chitrapathak-2 achieves 3-6x speedup over its predecessor with being state-of-the-art (SOTA) in Telugu (6.69 char ANLS) and second best in the rest. In addition, we present Parichay, an independent OCR model series designed specifically for 9 Indian government documents to extract structured key fields, achieving 89.8% Exact Match score with a faster inference. Together, these systems achieve SOTA performance and provide practical guidance for building production-scale OCR pipelines in the Indian context.
Authors: Eva Paraschou, Line Harder Clemmensen, Sneha Das
Abstract: Conventional large language model (LLM) fairness alignment largely focuses on mitigating bias along single sensitive attributes, overlooking fairness as an inherently multidimensional and context-specific value. This approach risks creating systems that achieve narrow fairness metrics while exacerbating disparities along untargeted attributes, a phenomenon known as bias spillover. While extensively studied in machine learning, bias spillover remains critically underexplored in LLM alignment. In this work, we investigate how targeted gender alignment affects fairness across nine sensitive attributes in three state-of-the-art LLMs (Mistral 7B, Llama 3.1 8B, Qwen 2.5 7B). Using Direct Preference Optimization and the BBQ benchmark, we evaluate fairness under ambiguous and disambiguous contexts. Our findings reveal noticeable bias spillover: while aggregate results show improvements, context-aware analysis exposes significant degradations in ambiguous contexts, particularly for physical appearance ($p< 0.001$ across all models), sexual orientation, and disability status. We demonstrate that improving fairness along one attribute can inadvertently worsen disparities in others under uncertainty, highlighting the necessity of context-aware, multi-attribute fairness evaluation frameworks.
Authors: Kamil Jeziorek, Piotr Wzorek, Krzysztof Blachut, Hiroshi Nakano, Manon Dampfhoffer, Thomas Mesquida, Hiroaki Nishi, Thomas Dalgaty, Tomasz Kryjak
Abstract: As the volume of data recorded by embedded edge sensors increases, particularly from neuromorphic devices producing discrete event streams, there is a growing need for hardware-aware neural architectures that enable efficient, low-latency, and energy-conscious local processing. We present an FPGA implementation of event-graph neural networks for audio processing. We utilise an artificial cochlea that converts time-series signals into sparse event data, reducing memory and computation costs. Our architecture was implemented on a SoC FPGA and evaluated on two open-source datasets. For classification task, our baseline floating-point model achieves 92.7% accuracy on SHD dataset - only 2.4% below the state of the art - while requiring over 10x and 67x fewer parameters. On SSC, our models achieve 66.9-71.0% accuracy. Compared to FPGA-based spiking neural networks, our quantised model reaches 92.3% accuracy, outperforming them by up to 19.3% while reducing resource usage and latency. For SSC, we report the first hardware-accelerated evaluation. We further demonstrate the first end-to-end FPGA implementation of event-audio keyword spotting, combining graph convolutional layers with recurrent sequence modelling. The system achieves up to 95% word-end detection accuracy, with only 10.53 microsecond latency and 1.18 W power consumption, establishing a strong benchmark for energy-efficient event-driven KWS.
Authors: Yixue Zhang (Beijing Innovation Center of Humanoid Robotics, The School of Advanced Manufacturing and Robotics, Peking University), Kun Wu (Beijing Innovation Center of Humanoid Robotics), Zhi Gao (Beijing Institute of Technology), Zhen Zhao (Beijing Innovation Center of Humanoid Robotics), Pei Ren (Beijing Innovation Center of Humanoid Robotics), Zhiyuan Xu (Beijing Innovation Center of Humanoid Robotics), Fei Liao (Beijing Innovation Center of Humanoid Robotics), Xinhua Wang (Beijing Innovation Center of Humanoid Robotics), Shichao Fan (Beijing Innovation Center of Humanoid Robotics, The School of Mechanical Engineering and Automation, Beihang University), Di Wu (Beijing Innovation Center of Humanoid Robotics, State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University), Qiuxuan Feng (Beijing Innovation Center of Humanoid Robotics, State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University), Meng Li (Beijing Innovation Center of Humanoid Robotics), Zhengping Che (Beijing Innovation Center of Humanoid Robotics), Chang Liu (The School of Advanced Manufacturing and Robotics, Peking University), Jian Tang (Beijing Innovation Center of Humanoid Robotics)
Abstract: The pursuit of general-purpose robotic manipulation is hindered by the scarcity of diverse, real-world interaction data. Unlike data collection from web in vision or language, robotic data collection is an active process incurring prohibitive physical costs. Consequently, automated task curation to maximize data value remains a critical yet under-explored challenge. Existing manual methods are unscalable and biased toward common tasks, while off-the-shelf foundation models often hallucinate physically infeasible instructions. To address this, we introduce RoboGene, an agentic framework designed to automate the generation of diverse, physically plausible manipulation tasks across single-arm, dual-arm, and mobile robots. RoboGene integrates three core components: diversity-driven sampling for broad task coverage, self-reflection mechanisms to enforce physical constraints, and human-in-the-loop refinement for continuous improvement. We conduct extensive quantitative analysis and large-scale real-world experiments, collecting datasets of 18k trajectories and introducing novel metrics to assess task quality, feasibility, and diversity. Results demonstrate that RoboGene significantly outperforms state-of-the-art foundation models (e.g., GPT-4o, Gemini 2.5 Pro). Furthermore, real-world experiments show that VLA models pre-trained with RoboGene achieve higher success rates and superior generalization, underscoring the importance of high-quality task generation. Our project is available at https://robogene-boost-vla.github.io.
Authors: Nicolas Salvy, Hugues Talbot, Bertrand Thirion
Abstract: Generative model evaluation commonly relies on high-dimensional embedding spaces to compute distances between samples. We show that dataset representations in these spaces are affected by the hubness phenomenon, which distorts nearest neighbor relationships and biases distance-based metrics. Building on the classical Iterative Contextual Dissimilarity Measure (ICDM), we introduce Generative ICDM (GICDM), a method to correct neighborhood estimation for both real and generated data. We introduce a multi-scale extension to improve empirical behavior. Extensive experiments on synthetic and real benchmarks demonstrate that GICDM resolves hubness-induced failures, restores reliable metric behavior, and improves alignment with human judgment.
Authors: Saurabh Bharti, Gaurav Azad, Abhinaw Jagtap, Nachiket Tapas
Abstract: The rapid advancement of large language models (LLMs) necessitates evaluation frameworks that reflect real-world academic rigor and multilingual complexity. This paper introduces IndicEval, a scalable benchmarking platform designed to assess LLM performance using authentic high-stakes examination questions from UPSC, JEE, and NEET across STEM and humanities domains in both English and Hindi. Unlike synthetic benchmarks, IndicEval grounds evaluation in real examination standards, enabling realistic measurement of reasoning, domain knowledge, and bilingual adaptability. The framework automates assessment using Zero-Shot, Few-Shot, and Chain-of-Thought (CoT) prompting strategies and supports modular integration of new models and languages. Experiments conducted on Gemini 2.0 Flash, GPT-4, Claude, and LLaMA 3-70B reveal three major findings. First, CoT prompting consistently improves reasoning accuracy, with substantial gains across subjects and languages. Second, significant cross-model performance disparities persist, particularly in high-complexity examinations. Third, multilingual degradation remains a critical challenge, with marked accuracy drops in Hindi compared to English, especially under Zero-Shot conditions. These results highlight persistent gaps in bilingual reasoning and domain transfer. Overall, IndicEval provides a practice-oriented, extensible foundation for rigorous, equitable evaluation of LLMs in multilingual educational settings and offers actionable insights for improving reasoning robustness and language adaptability.
Authors: Jeffrey T. H. Wong, Zixi Zhang, Junyi Liu, Yiren Zhao
Abstract: Existing Multi-Agent Systems (MAS) typically rely on static, homogeneous model configurations, limiting their ability to exploit the distinct strengths of differently post-trained models. To address this, we introduce Team-of-Thoughts, a novel MAS architecture that leverages the complementary capabilities of heterogeneous agents via an orchestrator-tool paradigm. Our framework introduces two key mechanisms to optimize performance: (1) an orchestrator calibration scheme that identifies models with superior coordination capabilities, and (2) a self-assessment protocol where tool agents profile their own domain expertise to account for variations in post-training skills. During inference, the orchestrator dynamically activates the most suitable tool agents based on these proficiency profiles. Experiments on five reasoning and code generation benchmarks show that Team-of-Thoughts delivers consistently superior task performance. Notably, on AIME24 and LiveCodeBench, our approach achieves accuracies of 96.67% and 72.53%, respectively, substantially outperforming homogeneous role-play baselines, which score 80% and 65.93%.
Authors: Jonathan Cook, Diego Antognini, Martin Klissarov, Claudiu Musat, Edward Grefenstette
Abstract: Large language models (LLMs) often struggle to learn from corrective feedback within a conversational context. They are rarely proactive in soliciting this feedback, even when faced with ambiguity, which can make their dialogues feel static, one-sided, and lacking the adaptive qualities of human conversation. To address these limitations, we draw inspiration from social meta-learning (SML) in humans - the process of learning how to learn from others. We formulate SML as a finetuning methodology, training LLMs to solicit and learn from language feedback in simulated pedagogical dialogues, where static tasks are converted into interactive social learning problems. SML effectively teaches models to use conversation to solve problems they are unable to solve in a single turn. This capability generalises across domains; SML on math problems produces models that better use feedback to solve coding problems and vice versa. Furthermore, despite being trained only on fully-specified problems, these models are better able to solve underspecified tasks where critical information is revealed over multiple turns. When faced with this ambiguity, SML-trained models make fewer premature answer attempts and are more likely to ask for the information they need. This work presents a scalable approach to developing AI systems that effectively learn from language feedback.
Authors: Ferdinand Kapl, Emmanouil Angelis, Kaitlin Maile, Johannes von Oswald, Stefan Bauer
Abstract: Looping, reusing a block of layers across depth, and depth growing, training shallow-to-deep models by duplicating middle layers, have both been linked to stronger reasoning, but their relationship remains unclear. We provide a mechanistic unification: looped and depth-grown models exhibit convergent depth-wise signatures, including increased reliance on late layers and recurring patterns aligned with the looped or grown block. These shared signatures support the view that their gains stem from a common form of iterative computation. Building on this connection, we show that the two techniques are adaptable and composable: applying inference-time looping to the middle blocks of a depth-grown model improves accuracy on some reasoning primitives by up to $2\times$, despite the model never being trained to loop. Both approaches also adapt better than the baseline when given more in-context examples or additional supervised fine-tuning data. Additionally, depth-grown models achieve the largest reasoning gains when using higher-quality, math-heavy cooldown mixtures, which can be further boosted by adapting a middle block to loop. Overall, our results position depth growth and looping as complementary, practical methods for inducing and scaling iterative computation to improve reasoning.
Authors: Xinyi Shang, Peng Sun, Jingyu Lin, Zhiqiang Shen
Abstract: Analytical diffusion models offer a mathematically transparent path to generative modeling by formulating the denoising score as an empirical-Bayes posterior mean. However, this interpretability comes at a prohibitive cost: the standard formulation necessitates a full-dataset scan at every timestep, scaling linearly with dataset size. In this work, we present the first systematic study addressing this scalability bottleneck. We challenge the prevailing assumption that the entire training data is necessary, uncovering the phenomenon of Posterior Progressive Concentration: the effective golden support of the denoising score is not static but shrinks asymptotically from the global manifold to a local neighborhood as the signal-to-noise ratio increases. Capitalizing on this, we propose Dynamic Time-Aware Golden Subset Diffusion (GoldDiff), a training-free framework that decouples inference complexity from dataset size. Instead of static retrieval, GoldDiff uses a coarse-to-fine mechanism to dynamically pinpoint the ''Golden Subset'' for inference. Theoretically, we derive rigorous bounds guaranteeing that our sparse approximation converges to the exact score. Empirically, GoldDiff achieves a $\bf 71 \times$ speedup on AFHQ while matching or achieving even better performance than full-scan baselines. Most notably, we demonstrate the first successful scaling of analytical diffusion to ImageNet-1K, unlocking a scalable, training-free paradigm for large-scale generative modeling.
Authors: Vasilis Gkolemis, Loukas Kavouras, Dimitrios Kyriakopoulos, Konstantinos Tsopelas, Dimitrios Rontogiannis, Giuseppe Casalicchio, Theodore Dalamagas, Christos Diou
Abstract: Generalized additive models (GAMs) offer interpretability through independent univariate feature effects but underfit when interactions are present in data. GA$^2$Ms add selected pairwise interactions which improves accuracy, but sacrifices interpretability and limits model auditing. We propose \emph{Conditionally Additive Local Models} (CALMs), a new model class, that balances the interpretability of GAMs with the accuracy of GA$^2$Ms. CALMs allow multiple univariate shape functions per feature, each active in different regions of the input space. These regions are defined independently for each feature as simple logical conditions (thresholds) on the features it interacts with. As a result, effects remain locally additive while varying across subregions to capture interactions. We further propose a principled distillation-based training pipeline that identifies homogeneous regions with limited interactions and fits interpretable shape functions via region-aware backfitting. Experiments on diverse classification and regression tasks show that CALMs consistently outperform GAMs and achieve accuracy comparable with GA$^2$Ms. Overall, CALMs offer a compelling trade-off between predictive accuracy and interpretability.
Authors: Doron Shavit
Abstract: Jailbreak prompts are a practical and evolving threat to large language models (LLMs), particularly in agentic systems that execute tools over untrusted content. Many attacks exploit long-context hiding, semantic camouflage, and lightweight obfuscations that can evade single-pass guardrails. We present RLM-JB, an end-to-end jailbreak detection framework built on Recursive Language Models (RLMs), in which a root model orchestrates a bounded analysis program that transforms the input, queries worker models over covered segments, and aggregates evidence into an auditable decision. RLM-JB treats detection as a procedure rather than a one-shot classification: it normalizes and de-obfuscates suspicious inputs, chunks text to reduce context dilution and guarantee coverage, performs parallel chunk screening, and composes cross-chunk signals to recover split-payload attacks. On AutoDAN-style adversarial inputs, RLM-JB achieves high detection effectiveness across three LLM backends (ASR/Recall 92.5-98.0%) while maintaining very high precision (98.99-100%) and low false positive rates (0.0-2.0%), highlighting a practical sensitivity-specificity trade-off as the screening backend changes.
Authors: Yuanjie Ren, Jinzheng Li, Yidi Qi
Abstract: We introduce MerLean, a fully automated agentic framework for autoformalization in quantum computation. MerLean extracts mathematical statements from \LaTeX{} source files, formalizes them into verified Lean~4 code built on Mathlib, and translates the result back into human-readable \LaTeX{} for semantic review. We evaluate MerLean on three theoretical quantum computing papers producing 2,050 Lean declarations from 114 statements in total. MerLean achieves end-to-end formalization on all three papers, reducing the verification burden to only the newly introduced definitions and axioms. Our results demonstrate that agentic autoformalization can scale to frontier research, offering both a practical tool for machine-verified peer review and a scalable engine for mining high-quality synthetic data to train future reasoning models. Our approach can also be generalized to any other rigorous research in mathematics and theoretical physics.
Authors: Maria Luisa Taccari, Kenza Tazi, Ois\'in M. Morrison, Andreas Grafberger, Juan Colonese, Corentin Carton de Wiart, Christel Prudhomme, Cinzia Mazzetti, Matthew Chantry, Florian Pappenberger
Abstract: Reliable global streamflow forecasting is essential for flood preparedness and water resource management, yet data-driven models often suffer from a performance gap when transitioning from historical reanalysis to operational forecast products. This paper introduces AIFL (Artificial Intelligence for Floods), a deterministic LSTM-based model designed for global daily streamflow forecasting. Trained on 18,588 basins curated from the CARAVAN dataset, AIFL utilises a novel two-stage training strategy to bridge the reanalysis-to-forecast domain shift. The model is first pre-trained on 40 years of ERA5-Land reanalysis (1980-2019) to capture robust hydrological processes, then fine-tuned on operational Integrated Forecasting System (IFS) control forecasts (2016-2019) to adapt to the specific error structures and biases of operational numerical weather prediction. To our knowledge, this is the first global model trained end-to-end within the CARAVAN ecosystem. On an independent temporal test set (2021-2024), AIFL achieves high predictive skill with a median modified Kling-Gupta Efficiency (KGE') of 0.66 and a median Nash-Sutcliffe Efficiency (NSE) of 0.53. Benchmarking results show that AIFL is highly competitive with current state-of-the-art global systems, achieving comparable accuracy while maintaining a transparent and reproducible forcing pipeline. The model demonstrates exceptional reliability in extreme-event detection, providing a streamlined and operationally robust baseline for the global hydrological community.
Authors: Dimitri Yatsenko (DataJoint Inc., Houston, USA), Thinh T. Nguyen (DataJoint Inc., Houston, USA)
Abstract: Operational rigor determines whether human-agent collaboration succeeds or fails. Scientific data pipelines need the equivalent of DevOps -- SciOps -- yet common approaches fragment provenance across disconnected systems without transactional guarantees. DataJoint 2.0 addresses this gap through the relational workflow model: tables represent workflow steps, rows represent artifacts, foreign keys prescribe execution order. The schema specifies not only what data exists but how it is derived -- a single formal system where data structure, computational dependencies, and integrity constraints are all queryable, enforceable, and machine-readable. Four technical innovations extend this foundation: object-augmented schemas integrating relational metadata with scalable object storage, semantic matching using attribute lineage to prevent erroneous joins, an extensible type system for domain-specific formats, and distributed job coordination designed for composability with external orchestration. By unifying data structure, data, and computational transformations, DataJoint creates a substrate for SciOps where agents can participate in scientific workflows without risking data corruption.
Authors: Qi You, Yitai Cheng, Zichao Zeng, James Haworth
Abstract: Street-view image attribute classification is a vital downstream task of image classification, enabling applications such as autonomous driving, urban analytics, and high-definition map construction. It remains computationally demanding whether training from scratch, initialising from pre-trained weights, or fine-tuning large models. Although pre-trained vision-language models such as CLIP offer rich image representations, existing adaptation or fine-tuning methods often rely on their global image embeddings, limiting their ability to capture fine-grained, localised attributes essential in complex, cluttered street scenes. To address this, we propose CLIP-MHAdapter, a variant of the current lightweight CLIP adaptation paradigm that appends a bottleneck MLP equipped with multi-head self-attention operating on patch tokens to model inter-patch dependencies. With approximately 1.4 million trainable parameters, CLIP-MHAdapter achieves superior or competitive accuracy across eight attribute classification tasks on the Global StreetScapes dataset, attaining new state-of-the-art results while maintaining low computational cost. The code is available at https://github.com/SpaceTimeLab/CLIP-MHAdapter.
Authors: Chia-chi Hsieh, Zan Zong, Xinyang Chen, Jianjiang Li, Jidong Zhai, Lijie Wen
Abstract: The growing demand for large language models (LLMs) requires serving systems to handle many concurrent requests with diverse service level objectives (SLOs). This exacerbates head-of-line (HoL) blocking during the compute-intensive prefill phase, where long-running requests monopolize resources and delay higher-priority ones, leading to widespread time-to-first-token (TTFT) SLO violations. While chunked prefill enables interruptibility, it introduces an inherent trade-off between responsiveness and throughput: reducing chunk size improves response latency but degrades computational efficiency, whereas increasing chunk size maximizes throughput but exacerbates blocking. This necessitates an adaptive preemption mechanism. However, dynamically balancing execution granularity against scheduling overheads remains a key challenge. In this paper, we propose FlowPrefill, a TTFT-goodput-optimized serving system that resolves this conflict by decoupling preemption granularity from scheduling frequency. To achieve adaptive prefill scheduling, FlowPrefill introduces two key innovations: 1) Operator-Level Preemption, which leverages operator boundaries to enable fine-grained execution interruption without the efficiency loss associated with fixed small chunking; and 2) Event-Driven Scheduling, which triggers scheduling decisions only upon request arrival or completion events, thereby supporting efficient preemption responsiveness while minimizing control-plane overhead. Evaluation on real-world production traces shows that FlowPrefill improves maximum goodput by up to 5.6$\times$ compared to state-of-the-art systems while satisfying heterogeneous SLOs.
Authors: Melkamu Abay Mersha, Jugal Kalita
Abstract: Transformer models achieve state-of-the-art performance across domains and tasks, yet their deeply layered representations make their predictions difficult to interpret. Existing explainability methods rely on final-layer attributions, capture either local token-level attributions or global attention patterns without unification, and lack context-awareness of inter-token dependencies and structural components. They also fail to capture how relevance evolves across layers and how structural components shape decision-making. To address these limitations, we proposed the \textbf{Context-Aware Layer-wise Integrated Gradients (CA-LIG) Framework}, a unified hierarchical attribution framework that computes layer-wise Integrated Gradients within each Transformer block and fuses these token-level attributions with class-specific attention gradients. This integration yields signed, context-sensitive attribution maps that capture supportive and opposing evidence while tracing the hierarchical flow of relevance through the Transformer layers. We evaluate the CA-LIG Framework across diverse tasks, domains, and transformer model families, including sentiment analysis and long and multi-class document classification with BERT, hate speech detection in a low-resource language setting with XLM-R and AfroLM, and image classification with Masked Autoencoder vision Transformer model. Across all tasks and architectures, CA-LIG provides more faithful attributions, shows stronger sensitivity to contextual dependencies, and produces clearer, more semantically coherent visualizations than established explainability methods. These results indicate that CA-LIG provides a more comprehensive, context-aware, and reliable explanation of Transformer decision-making, advancing both the practical interpretability and conceptual understanding of deep neural models.
Authors: Mengjie Qian, Guangzhi Sun, Mark J. F. Gales, Kate M. Knill
Abstract: Large language models (LLMs) are increasingly applied as automatic evaluators for natural language generation assessment often using pairwise comparative judgements. Existing approaches typically rely on single judges or aggregate multiple judges assuming equal reliability. In practice, LLM judges vary substantially in performance across tasks and aspects, and their judgment probabilities may be biased and inconsistent. Furthermore, human-labelled supervision for judge calibration may be unavailable. We first empirically demonstrate that inconsistencies in LLM comparison probabilities exist and show that it limits the effectiveness of direct probability-based ranking. To address this, we study the LLM-as-a-jury setting and propose BT-sigma, a judge-aware extension of the Bradley-Terry model that introduces a discriminator parameter for each judge to jointly infer item rankings and judge reliability from pairwise comparisons alone. Experiments on benchmark NLG evaluation datasets show that BT-sigma consistently outperforms averaging-based aggregation methods, and that the learned discriminator strongly correlates with independent measures of the cycle consistency of LLM judgments. Further analysis reveals that BT-sigma can be interpreted as an unsupervised calibration mechanism that improves aggregation by modelling judge reliability.
Authors: Robin Lorenz, Sean Tull
Abstract: Abstracting from a low level to a more explanatory high level of description, and ideally while preserving causal structure, is fundamental to scientific practice, to causal inference problems, and to robust, efficient and interpretable AI. We present a general account of abstractions between low and high level models as natural transformations, focusing on the case of causal models. This provides a new formalisation of causal abstraction, unifying several notions in the literature, including constructive causal abstraction, Q-$\tau$ consistency, abstractions based on interchange interventions, and `distributed' causal abstractions. Our approach is formalised in terms of category theory, and uses the general notion of a compositional model with a given set of queries and semantics in a monoidal, cd- or Markov category; causal models and their queries such as interventions being special cases. We identify two basic notions of abstraction: downward abstractions mapping queries from high to low level; and upward abstractions, mapping concrete queries such as Do-interventions from low to high. Although usually presented as the latter, we show how common causal abstractions may, more fundamentally, be understood in terms of the former. Our approach also leads us to consider a new stronger notion of `component-level' abstraction, applying to the individual components of a model. In particular, this yields a novel, strengthened form of constructive causal abstraction at the mechanism-level, for which we prove characterisation results. Finally, we show that abstraction can be generalised to further compositional models, including those with a quantum semantics implemented by quantum circuits, and we take first steps in exploring abstractions between quantum compositional circuit models and high-level classical causal models as a means to explainable quantum AI.
Authors: SungJun Cho, Chetan Gohil, Rukuang Huang, Oiwi Parker Jones, Mark W. Woolrich
Abstract: Recent success in natural language processing has motivated growing interest in large-scale foundation models for neuroimaging data. Such models often require discretization of continuous neural time series data, a process referred to as 'tokenization'. However, the impact of different tokenization strategies for neural data is currently poorly understood. In this work, we present a systematic evaluation of sample-level tokenization strategies for transformer-based large neuroimaging models (LNMs) applied to magnetoencephalography (MEG) data. We compare learnable and non-learnable tokenizers by examining their signal reconstruction fidelity and their impact on subsequent foundation modeling performance (token prediction, biological plausibility of generated data, preservation of subject-specific information, and performance on downstream tasks). For the learnable tokenizer, we introduce a novel approach based on an autoencoder. Experiments were conducted on three publicly available MEG datasets spanning different acquisition sites, scanners, and experimental paradigms. Our results show that both learnable and non-learnable discretization schemes achieve high reconstruction accuracy and broadly comparable performance across most evaluation criteria, suggesting that simple fixed sample-level tokenization strategies can be used in the development of neural foundation models. The code is available at https://github.com/OHBA-analysis/Cho2026_Tokenizer.
Authors: Ethan Blaser, Jiuqi Wang, Shangtong Zhang
Abstract: The average reward is a fundamental performance metric in reinforcement learning (RL) focusing on the long-run performance of an agent. Differential temporal difference (TD) learning algorithms are a major advance for average reward RL as they provide an efficient online method to learn the value functions associated with the average reward in both on-policy and off-policy settings. However, existing convergence guarantees require a local clock in learning rates tied to state visit counts, which practitioners do not use and does not extend beyond tabular settings. We address this limitation by proving the almost sure convergence of on-policy $n$-step differential TD for any $n$ using standard diminishing learning rates without a local clock. We then derive three sufficient conditions under which off-policy $n$-step differential TD also converges without a local clock. These results strengthen the theoretical foundations of differential TD and bring its convergence analysis closer to practical implementations.
Authors: Yu Xie, Ludwig Winkler, Lixin Sun, Sarah Lewis, Adam E. Foster, Jos\'e Jim\'enez Luna, Tim Hempel, Michael Gastegger, Yaoyi Chen, Iryna Zaporozhets, Cecilia Clementi, Christopher M. Bishop, Frank No\'e
Abstract: The rare-event sampling problem has long been the central limiting factor in molecular dynamics (MD), especially in biomolecular simulation. Recently, diffusion models such as BioEmu have emerged as powerful equilibrium samplers that generate independent samples from complex molecular distributions, eliminating the cost of sampling rare transition events. However, a sampling problem remains when computing observables that rely on states which are rare in equilibrium, for example folding free energies. Here, we introduce enhanced diffusion sampling, enabling efficient exploration of rare-event regions while preserving unbiased thermodynamic estimators. The key idea is to perform quantitatively accurate steering protocols to generate biased ensembles and subsequently recover equilibrium statistics via exact reweighting. We instantiate our framework in three algorithms: UmbrellaDiff (umbrella sampling with diffusion models), $\Delta$G-Diff (free-energy differences via tilted ensembles), and MetaDiff (a batchwise analogue for metadynamics). Across toy systems, protein folding landscapes and folding free energies, our methods achieve fast, accurate, and scalable estimation of equilibrium properties within GPU-minutes to hours per system -- closing the rare-event sampling gap that remained after the advent of diffusion-model equilibrium samplers.
Authors: Sonakshi Gupta, Akhlak Mahmood, Wei Xiong, Rampi Ramprasad
Abstract: Polymer literature contains a large and growing body of experimental knowledge, yet much of it is buried in unstructured text and inconsistent terminology, making systematic retrieval and reasoning difficult. Existing tools typically extract narrow, study-specific facts in isolation, failing to preserve the cross-study context required to answer broader scientific questions. Retrieval-augmented generation (RAG) offers a promising way to overcome this limitation by combining large language models (LLMs) with external retrieval, but its effectiveness depends strongly on how domain knowledge is represented. In this work, we develop two retrieval pipelines: a dense semantic vector-based approach (VectorRAG) and a graph-based approach (GraphRAG). Using over 1,000 polyhydroxyalkanoate (PHA) papers, we construct context-preserving paragraph embeddings and a canonicalized structured knowledge graph supporting entity disambiguation and multi-hop reasoning. We evaluate these pipelines through standard retrieval metrics, comparisons with general state-of-the-art systems such as GPT and Gemini, and qualitative validation by a domain chemist. The results show that GraphRAG achieves higher precision and interpretability, while VectorRAG provides broader recall, highlighting complementary trade-offs. Expert validation further confirms that the tailored pipelines, particularly GraphRAG, produce well-grounded, citation-reliable responses with strong domain relevance. By grounding every statement in evidence, these systems enable researchers to navigate the literature, compare findings across studies, and uncover patterns that are difficult to extract manually. More broadly, this work establishes a practical framework for building materials science assistants using curated corpora and retrieval design, reducing reliance on proprietary models while enabling trustworthy literature analysis at scale.
Authors: Yuyan Bu, Xiaohao Liu, ZhaoXing Ren, Yaodong Yang, Juntao Dai
Abstract: The widespread deployment of large language models (LLMs) across linguistic communities necessitates reliable multilingual safety alignment. However, recent efforts to extend alignment to other languages often require substantial resources, either through large-scale, high-quality supervision in the target language or through pairwise alignment with high-resource languages, which limits scalability. In this work, we propose a resource-efficient method for improving multilingual safety alignment. We introduce a plug-and-play Multi-Lingual Consistency (MLC) loss that can be integrated into existing monolingual alignment pipelines. By improving collinearity between multilingual representation vectors, our method encourages directional consistency at the multilingual semantic level in a single update. This allows simultaneous alignment across multiple languages using only multilingual prompt variants without requiring additional response-level supervision in low-resource languages. We validate the proposed method across different model architectures and alignment paradigms, and demonstrate its effectiveness in enhancing multilingual safety with limited impact on general model utility. Further evaluation across languages and tasks indicates improved cross-lingual generalization, suggesting the proposed approach as a practical solution for multilingual consistency alignment under limited supervision.
Authors: Jaid Monwar Chowdhury, Chi-An Fu, Reyhaneh Jabbarvand
Abstract: Automated unit test generation for C remains a formidable challenge due to the semantic gap between high-level program intent and the rigid syntactic constraints of pointer arithmetic and manual memory management. While Large Language Models (LLMs) exhibit strong generative capabilities, direct intent-to-code synthesis frequently suffers from the leap-to-code failure mode, where models prematurely emit code without grounding in program structure, constraints, and semantics. This will result in non-compilable tests, hallucinated function signatures, low branch coverage, and semantically irrelevant assertions that cannot properly capture bugs. We introduce SPARC, a neuro-symbolic, scenario-based framework that bridges this gap through four stages: (1) Control Flow Graph (CFG) analysis, (2) an Operation Map that grounds LLM reasoning in validated utility helpers, (3) Path-targeted test synthesis, and (4) an iterative, self-correction validation loop using compiler and runtime feedback. We evaluate SPARC on 59 real-world and algorithmic subjects, where it outperforms the vanilla prompt generation baseline by 31.36% in line coverage, 26.01% in branch coverage, and 20.78% in mutation score, matching or exceeding the symbolic execution tool KLEE on complex subjects. SPARC retains 94.3% of tests through iterative repair and produces code with significantly higher developer-rated readability and maintainability. By aligning LLM reasoning with program structure, SPARC provides a scalable path for industrial-grade testing of legacy C codebases.
Authors: Wenxuan Ding, Nicholas Tomlin, Greg Durrett
Abstract: LLMs are increasingly being used for complex problems which are not necessarily resolved in a single response, but require interacting with an environment to acquire information. In these scenarios, LLMs must reason about inherent cost-uncertainty tradeoffs in when to stop exploring and commit to an answer. For instance, on a programming task, an LLM should test a generated code snippet if it is uncertain about the correctness of that code; the cost of writing a test is nonzero, but typically lower than the cost of making a mistake. In this work, we show that we can induce LLMs to explicitly reason about balancing these cost-uncertainty tradeoffs, then perform more optimal environment exploration. We formalize multiple tasks, including information retrieval and coding, as sequential decision-making problems under uncertainty. Each problem has latent environment state that can be reasoned about via a prior which is passed to the LLM agent. We introduce a framework called Calibrate-Then-Act (CTA), where we feed the LLM this additional context to enable it to act more optimally. This improvement is preserved even under RL training of both the baseline and CTA. Our results on information-seeking QA and on a simplified coding task show that making cost-benefit tradeoffs explicit with CTA can help agents discover more optimal decision-making strategies.
Authors: Shen Zhou Hong, Alex Kleinman, Alyssa Mathiowetz, Adam Howes, Julian Cohen, Suveer Ganta, Alex Letizia, Dora Liao, Deepika Pahari, Xavier Roberts-Gaal, Luca Righetti, Joe Torres
Abstract: Large language models (LLMs) perform strongly on biological benchmarks, raising concerns that they may help novice actors acquire dual-use laboratory skills. Yet, whether this translates to improved human performance in the physical laboratory remains unclear. To address this, we conducted a pre-registered, investigator-blinded, randomized controlled trial (June-August 2025; n = 153) evaluating whether LLMs improve novice performance in tasks that collectively model a viral reverse genetics workflow. We observed no significant difference in the primary endpoint of workflow completion (5.2% LLM vs. 6.6% Internet; P = 0.759), nor in the success rate of individual tasks. However, the LLM arm had numerically higher success rates in four of the five tasks, most notably for the cell culture task (68.8% LLM vs. 55.3% Internet; P = 0.059). Post-hoc Bayesian modeling of pooled data estimates an approximate 1.4-fold increase (95% CrI 0.74-2.62) in success for a "typical" reverse genetics task under LLM assistance. Ordinal regression modelling suggests that participants in the LLM arm were more likely to progress through intermediate steps across all tasks (posterior probability of a positive effect: 81%-96%). Overall, mid-2025 LLMs did not substantially increase novice completion of complex laboratory procedures but were associated with a modest performance benefit. These results reveal a gap between in silico benchmarks and real-world utility, underscoring the need for physical-world validation of AI biosecurity assessments as model capabilities and user proficiency evolve.
Authors: Nils Palumbo, Sarthak Choudhary, Jihye Choi, Prasad Chalasani, Mihai Christodorescu, Somesh Jha
Abstract: LLM-based agents are increasingly being deployed in contexts requiring complex authorization policies: customer service protocols, approval workflows, data access restrictions, and regulatory compliance. Embedding these policies in prompts provides no enforcement guarantees. We present PCAS, a Policy Compiler for Agentic Systems that provides deterministic policy enforcement. Enforcing such policies requires tracking information flow across agents, which linear message histories cannot capture. Instead, PCAS models the agentic system state as a dependency graph capturing causal relationships among events such as tool calls, tool results, and messages. Policies are expressed in a Datalog-derived language, as declarative rules that account for transitive information flow and cross-agent provenance. A reference monitor intercepts all actions and blocks violations before execution, providing deterministic enforcement independent of model reasoning. PCAS takes an existing agent implementation and a policy specification, and compiles them into an instrumented system that is policy-compliant by construction, with no security-specific restructuring required. We evaluate PCAS on three case studies: information flow policies for prompt injection defense, approval workflows in a multi-agent pharmacovigilance system, and organizational policies for customer service. On customer service tasks, PCAS improves policy compliance from 48% to 93% across frontier models, with zero policy violations in instrumented runs.
Authors: Caleb J. S. Barr, Olivia Erdelyi, Paul D. Docherty, Randolph C. Grace
Abstract: Recent regulatory proposals for artificial intelligence emphasize fairness requirements for machine learning models. However, precisely defining the appropriate measure of fairness is challenging due to philosophical, cultural and political contexts. Biases can infiltrate machine learning models in complex ways depending on the model's context, rendering a single common metric of fairness insufficient. This ambiguity highlights the need for criteria to guide the selection of context-aware measures, an issue of increasing importance given the proliferation of ever tighter regulatory requirements. To address this, we developed a flowchart to guide the selection of contextually appropriate fairness measures. Twelve criteria were used to formulate the flowchart. This included consideration of model assessment criteria, model selection criteria, and data bias. We also review fairness literature in the context of machine learning and link it to core regulatory instruments to assist policymakers, AI developers, researchers, and other stakeholders in appropriately addressing fairness concerns and complying with relevant regulatory requirements.
Authors: Yong Lai, Haolong Tong, Zhenghang Xu, Minghao Yin
Abstract: Quantitative information flow analyses (QIF) are a class of techniques for measuring the amount of confidential information leaked by a program to its public outputs. Shannon entropy is an important method to quantify the amount of leakage in QIF. This paper focuses on the programs modeled in Boolean constraints and optimizes the two stages of the Shannon entropy computation to implement a scalable precise tool PSE. In the first stage, we design a knowledge compilation language called \ADDAND that combines Algebraic Decision Diagrams and conjunctive decomposition. \ADDAND avoids enumerating possible outputs of a program and supports tractable entropy computation. In the second stage, we optimize the model counting queries that are used to compute the probabilities of outputs. We compare PSE with the state-of-the-art probabilistic approximately correct tool EntropyEstimation, which was shown to significantly outperform the previous precise tools. The experimental results demonstrate that PSE solved 56 more benchmarks compared to EntropyEstimation in a total of 459. For 98\% of the benchmarks that both PSE and EntropyEstimation solved, PSE is at least $10\times$ as efficient as EntropyEstimation.
Authors: Chang Han Low, Ziyue Wang, Tianyi Zhang, Zhu Zhuo, Zhitao Zeng, Evangelos B. Mazomenos, Yueming Jin
Abstract: Robotic-assisted surgery (RAS) is central to modern surgery, driving the need for intelligent systems with accurate scene understanding. Most existing surgical AI methods rely on isolated, task-specific models, leading to fragmented pipelines with limited interpretability and no unified understanding of RAS scene. Vision-Language Models (VLMs) offer strong zero-shot reasoning, but struggle with hallucinations, domain gaps and weak task-interdependency modeling. To address the lack of unified data for RAS scene understanding, we introduce SurgCoTBench, the first reasoning-focused benchmark in RAS, covering 14256 QA pairs with frame-level annotations across five major surgical tasks. Building on SurgCoTBench, we propose SurgRAW, a clinically aligned Chain-of-Thought (CoT) driven agentic workflow for zero-shot multi-task reasoning in surgery. SurgRAW employs a hierarchical reasoning workflow where an orchestrator divides surgical scene understanding into two reasoning streams and directs specialized agents to generate task-level reasoning, while higher-level agents capture workflow interdependencies or ground output clinically. Specifically, we propose a panel discussion mechanism to ensure task-specific agents collaborate synergistically and leverage on task interdependencies. Similarly, we incorporate a retrieval-augmented generation module to enrich agents with surgical knowledge and alleviate domain gaps in general VLMs. We design task-specific CoT prompts grounded in surgical domain to ensure clinically aligned reasoning, reduce hallucinations and enhance interpretability. Extensive experiments show that SurgRAW surpasses mainstream VLMs and agentic systems and outperforms a supervised model by 14.61% accuracy. Dataset and code is available at https://github.com/jinlab-imvr/SurgRAW.git .
Authors: Yinon Goldshtein, Gal Perelman, Assaf Schuster, Avi Ostfeld
Abstract: The integration of Large Language Models (LLMs) into engineering workflows presents new opportunities for making computational tools more accessible. Especially where such tools remain underutilized due to technical or expertise barriers, such as water distribution system (WDS) management. This study introduces LLM-EPANET, an agent-based framework that enables natural language interaction with EPANET, the benchmark WDS simulator. The framework combines retrieval-augmented generation and multi-agent orchestration to automatically translate user queries into executable code, run simulations, and return structured results. A curated set of 69 benchmark queries is introduced to evaluate performance across state-of-the-art LLMs. Results show that LLMs can effectively support a wide range of modeling tasks, achieving 56-81% accuracy overall, and over 90% for simpler queries. These findings highlight the potential of LLM-based modeling to democratize data-driven decision-making in the water sector through transparent, interactive AI interfaces. The framework code and benchmark queries are shared as an open resource: https://github.com/yinon-gold/LLMs-in-WDS-Modeling.
Authors: Sara Fish, Julia Shephard, Minkai Li, Ran I. Shorrer, Yannai A. Gonczarowski
Abstract: We develop evaluation methods for measuring the economic decision-making capabilities and tendencies of LLMs. First, we develop benchmarks derived from key problems in economics -- procurement, scheduling, and pricing -- that test an LLM's ability to learn from the environment in context. Second, we develop the framework of litmus tests, evaluations that quantify an LLM's choice behavior on a stylized decision-making task with multiple conflicting objectives. Each litmus test outputs a litmus score, which quantifies an LLM's tradeoff response, a reliability score, which measures the coherence of an LLM's choice behavior, and a competency score, which measures an LLM's capability at the same task when the conflicting objectives are replaced by a single, well-specified objective. Evaluating a broad array of frontier LLMs, we (1) investigate changes in LLM capabilities and tendencies over time, (2) derive economically meaningful insights from the LLMs' choice behavior and chain-of-thought, (3) validate our litmus test framework by testing self-consistency, robustness, and generalizability. Overall, this work provides a foundation for evaluating LLM agents as they are further integrated into economic decision-making.
Authors: Jie Peng, Jiarui Ji, Runlin Lei, Zhewei Wei, Yongchao Liu, Chuntao Hong
Abstract: Dynamic Text-Attributed Graphs (DyTAGs), which intricately integrate structural, temporal, and textual attributes, are crucial for modeling complex real-world systems. However, most existing DyTAG datasets exhibit poor textual quality, which severely limits their utility for generative DyTAG tasks requiring semantically rich inputs. Additionally, prior work mainly focuses on discriminative tasks on DyTAGs, resulting in a lack of standardized task formulations and evaluation protocols tailored for DyTAG generation. To address these critical issues, we propose Generative DyTAG Benchmark (GDGB), which comprises eight meticulously curated DyTAG datasets with high-quality textual features for both nodes and edges, overcoming limitations of prior datasets. Building on GDGB, we define two novel DyTAG generation tasks: Transductive Dynamic Graph Generation (TDGG) and Inductive Dynamic Graph Generation (IDGG). TDGG transductively generates a target DyTAG based on the given source and destination node sets, while the more challenging IDGG introduces new node generation to inductively model the dynamic expansion of real-world graph data. To enable holistic evaluation, we design multifaceted metrics that assess the structural, temporal, and textual quality of the generated DyTAGs. We further propose GAG-General, an LLM-based multi-agent generative framework tailored for reproducible and robust benchmarking of DyTAG generation. Experimental results demonstrate that GDGB enables rigorous evaluation of TDGG and IDGG, with key insights revealing the critical interplay of structural and textual features in DyTAG generation. These findings establish GDGB as a foundational resource for advancing generative DyTAG research and unlocking further practical applications in DyTAG generation. The dataset and source code are available at https://github.com/Lucas-PJ/GDGB-ALGO.
Authors: C\'edric Colas, Tracey Mills, Ben Prystawski, Michael Henry Tessler, Noah Goodman, Jacob Andreas, Joshua Tenenbaum
Abstract: The ability to combine linguistic guidance from others with direct experience is central to human development, enabling safe and rapid learning in new environments. How do people integrate these two sources of knowledge, and how might AI systems? We present a computational framework that models social learning as joint probabilistic inference over structured, executable world models given sensorimotor and linguistic data. We make this possible by turning a pretrained language model into a probabilistic model of how humans share advice conditioned on their beliefs, allowing our agents both to generate advice for others and to interpret linguistic input as evidence during Bayesian inference. Using behavioral experiments and simulations across 10 video games, we show how linguistic guidance can shape exploration and accelerate learning by reducing risky interactions and speeding up key discoveries in both humans and models. We further explore how knowledge can accumulate across generations through iterated learning experiments and demonstrate successful knowledge transfer between humans and models -- revealing how structured, language-compatible representations might enable human-machine collaborative learning.
Authors: Tong Guan, Zijie Meng, Dianqi Li, Shiyu Wang, Chao-Han Huck Yang, Qingsong Wen, Zuozhu Liu, Sabato Marco Siniscalchi, Ming Jin, Shirui Pan
Abstract: Recent advances in multimodal time series learning underscore a paradigm shift from analytics centered on basic patterns toward advanced time series understanding and reasoning. However, existing multimodal time series datasets mostly remain at the level of surface alignment and question answering, without reaching the depth of genuine reasoning. The absence of well-defined tasks that genuinely require time series reasoning, along with the scarcity of high-quality data, has limited progress in building practical time series reasoning models (TSRMs). To this end, we introduce Time Series Reasoning Suite (TSR-Suite), which formalizes four atomic tasks that span three fundamental capabilities for reasoning with time series: (1) perception, acquired through scenario understanding and causality discovery; (2) extrapolation, realized via event-aware forecasting; and (3) decision-making, developed through deliberation over perception and extrapolation. TSR-Suite is the first comprehensive time series reasoning suite that supports not only thorough evaluation but also the data pipeline and training of TSRMs. It contains more than 23K samples, of which 2.3K are carefully curated through a human-guided hierarchical annotation process. Building on this foundation, we introduce TimeOmni-1, the first unified reasoning model designed to address diverse real-world problems demanding time series reasoning. The model is trained in multiple stages, integrating a mixture of task scenarios, novel reward functions, and tailored optimizations. Experiments show that TimeOmni-1 delivers strong out-of-distribution generalization across all tasks and achieves a high rate of valid responses. It significantly improves causality discovery accuracy (64.0% vs. 35.9% with GPT-4.1) and raises the valid response rate by over 6% compared to GPT-4.1 on the event-aware forecasting task.
Authors: Rongzhi Zhang, Liqin Ye, Yuzhao Heng, Xiang Chen, Tong Yu, Lingkai Kong, Sudheer Chava, Chao Zhang
Abstract: Precise attribute intensity control--generating Large Language Model (LLM) outputs with specific, user-defined attribute intensities--is crucial for AI systems adaptable to diverse user expectations. Current LLM alignment methods, however, typically provide only directional or open-ended guidance, failing to reliably achieve exact attribute intensities. We address this limitation with three key designs: (1) reformulating precise attribute intensity control as a target-reaching problem, rather than simple maximization; (2) training a lightweight value function via temporal-difference learning to predict final attribute intensity scores from partial generations, thereby steering LLM outputs; and (3) employing gradient-based interventions on hidden representations to navigate the model precisely towards specific attribute intensity targets. Our method enables fine-grained, continuous control over attribute intensities, moving beyond simple directional alignment. Experiments on LLaMA-3.2-3b and Phi-4-mini confirm our method's ability to steer text generation to user-specified attribute intensities with high accuracy. Finally, we demonstrate efficiency enhancements across three downstream tasks: preference data synthesis, Pareto frontier approximation and optimization, and distillation of aligned behaviors for intervention-free inference. Our code is available on https://github.com/Pre-Control/pre-control
Authors: Aaron Bell, Amit Aides, Amr Helmy, Arbaaz Muslim, Aviad Barzilai, Aviv Slobodkin, Bolous Jaber, David Schottlander, George Leifman, Joydeep Paul, Mimi Sun, Nadav Sherman, Natalie Williams, Per Bjornsson, Roy Lee, Ruth Alcantara, Thomas Turnbull, Tomer Shekel, Vered Silverman, Yotam Gigi, Adam Boulanger, Alex Ottenwess, Ali Ahmadalipour, Anna Carter, Behzad Vahedi, Charles Elliott, David Andre, Elad Aharoni, Gia Jung, Hassler Thurston, Jacob Bien, Jamie McPike, Jessica Sapick, Juliet Rothenberg, Kartik Hegde, Kel Markert, Kim Philipp Jablonski, Luc Houriez, Monica Bharel, Phing VanLee, Reuven Sayag, Sebastian Pilarski, Shelley Cazares, Shlomi Pasternak, Siduo Jiang, Thomas Colthurst, Yang Chen, Yehonathan Refael, Yochai Blau, Yuval Carny, Yael Maguire, Avinatan Hassidim, James Manyika, Tim Thelin, Genady Beryozkin, Gautam Prasad, Luke Barrington, Yossi Matias, Niv Efron, Shravya Shetty
Abstract: Geospatial data offers immense potential for understanding our planet. However, the sheer volume and diversity of this data along with its varied resolutions, timescales, and sparsity pose significant challenges for thorough analysis and interpretation. This paper introduces Earth AI, a family of geospatial AI models and agentic reasoning that enables significant advances in our ability to unlock novel and profound insights into our planet. This approach is built upon foundation models across three key domains--Planet-scale Imagery, Population, and Environment--and an intelligent Gemini-powered reasoning engine. We present rigorous benchmarks showcasing the power and novel capabilities of our foundation models and validate that when used together, they provide complementary value for geospatial inference and their synergies unlock superior predictive capabilities. To handle complex, multi-step queries, we developed a Gemini-powered agent that jointly reasons over our multiple foundation models along with large geospatial data sources and tools. On a new benchmark of real-world crisis scenarios, our agent demonstrates the ability to deliver critical and timely insights, effectively bridging the gap between raw geospatial data and actionable understanding.
Authors: Maohao Ran, Zhenglin Wan, Cooper Lin, Yanting Zhang, Hongyu Xin, Hongwei Fan, Yibo Xu, Beier Luo, Yaxin Zhou, Wangbo Zhao, Lijie Yang, Lang Feng, Fuchao Yang, Jingxuan Wu, Yiqiao Huang, Chendong Ma, Dailing Jiang, Jianbo Deng, Sihui Han, Yang You, Bo An, Yike Guo, Jun Song
Abstract: LLM-based agents are increasingly capable of complex task execution, yet current agentic systems remain constrained by text-centric paradigms that struggle with long-horizon tasks due to fragile multi-turn dependencies and context drift. We present CaveAgent, a framework that shifts tool use from ``LLM-as-Text-Generator'' to ``LLM-as-Runtime-Operator.'' CaveAgent introduces a dual-stream architecture that inverts the conventional paradigm: rather than treating the LLM's text context as the primary workspace with tools as auxiliary, CaveAgent elevates the persistent Python runtime as the central locus of state, with a lightweight semantic stream serving as its orchestrator. Beyond leveraging code generation to resolve interdependent sub-tasks (e.g., loops, conditionals) in a single step, CaveAgent introduces \textit{Stateful Runtime Management}: it injects, manipulates, and retrieves complex Python objects (e.g., DataFrames, database connections) that persist across turns, unlike existing code-based approaches that remain text-bound. CaveAgent further provides a runtime-integrated skill management system that extends the Agent Skills open standard, enabling ecosystem interoperability through executable skill injections. This persistence mechanism serves as a high-fidelity external memory that reduces context drift in multi-turn interactions and preserves processed data for downstream applications without information loss. Evaluations show consistent improvement across challenging benchmarks, enabling CaveAgent to handle data scales that cause context overflow in both JSON-based and code-based agents. The accessible runtime state further provides programmatically verifiable feedback, enabling automated evaluation and reward signal generation without human annotation and establishing a structural foundation for future research in Reinforcement Learning with Verifiable Rewards (RLVR).
Authors: Zhuoyang Zou, Abolfazl Ansari, Delvin Ce Zhang, Dongwon Lee, Wenpeng Yin
Abstract: Paper weakness identification using single-agent or multi-agent LLMs has attracted increasing attention, yet existing approaches exhibit key limitations. Many multi-agent systems simulate human roles at a surface level, missing the underlying criteria that lead experts to assess complementary intellectual aspects of a paper. Moreover, prior methods implicitly assume identified weaknesses are valid, ignoring reviewer bias, misunderstanding, and the critical role of author rebuttals in validating review quality. Finally, most systems output unranked weakness lists, rather than prioritizing the most consequential issues for users. In this work, we propose DIAGPaper, a novel multi-agent framework that addresses these challenges through three tightly integrated modules. The customizer module simulates human-defined review criteria and instantiates multiple reviewer agents with criterion-specific expertise. The rebuttal module introduces author agents that engage in structured debate with reviewer agents to validate and refine proposed weaknesses. The prioritizer module learns from large-scale human review practices to assess the severity of validated weaknesses and surfaces the top-K severest ones to users. Experiments on two benchmarks, AAAR and ReviewCritique, demonstrate that DIAGPaper substantially outperforms existing methods by producing more valid and more paper-specific weaknesses, while presenting them in a user-oriented, prioritized manner.
Authors: Fabian P. Kr\"uger, Andrea Hunklinger, Adrian Wolny, Tim J. Adler, Igor Tetko, Santiago David Villalba
Abstract: Optimizing the structure of molecules to achieve desired properties is a central bottleneck across the chemical sciences, particularly in the pharmaceutical industry where it underlies the discovery of new drugs. Since molecular property evaluation often relies on costly and rate-limited oracles, such as experimental assays, molecular optimization must be highly sample-efficient. To address this, we introduce SEISMO, an LLM agent that performs strictly online, inference-time molecular optimization, updating after every oracle call without the need for population-based or batched learning. SEISMO conditions each proposal on the full optimization trajectory, combining natural-language task descriptions with scalar scores and, when available, structured explanatory feedback. Across the Practical Molecular Optimization benchmark of 23 tasks, SEISMO achieves a 2-3 times higher area under the optimisation curve than prior methods, often reaching near-maximal task scores within 50 oracle calls. Our additional medicinal-chemistry tasks show that providing explanatory feedback further improves efficiency, demonstrating that leveraging domain knowledge and structured information is key to sample-efficient molecular optimization.
Authors: Zeping Li, Hongru Wang, Yiwen Zhao, Guanhua Chen, Yixia Li, Keyang Chen, Yixin Cao, Guangnan Ye, Hongfeng Chai, Zhenfei Yin
Abstract: Tool-using agents based on Large Language Models (LLMs) excel in tasks such as mathematical reasoning and multi-hop question answering. However, in long trajectories, agents often trigger excessive and low-quality tool calls, increasing latency and degrading inference performance, making managing tool-use behavior challenging. In this work, we conduct entropy-based pilot experiments and observe a strong positive correlation between entropy reduction and high-quality tool calls. Building on this finding, we propose using entropy reduction as a supervisory signal and design two reward strategies to address the differing needs of optimizing tool-use behavior. Sparse outcome rewards provide coarse, trajectory-level guidance to improve efficiency, while dense process rewards offer fine-grained supervision to enhance performance. Experiments across diverse domains show that both reward designs improve tool-use behavior: the former reduces tool calls by 72.07% compared to the average of baselines, while the latter improves performance by 22.27%. These results position entropy reduction as a key mechanism for enhancing tool-use behavior, enabling agents to be more adaptive in real-world applications.
Authors: Kate H. Bentley, Luca Belli, Adam M. Chekroud, Emily J. Ward, Emily R. Dworkin, Emily Van Ark, Kelly M. Johnston, Will Alexander, Millard Brown, Matt Hawrilenko
Abstract: Millions now use generative AI chatbots for psychological support. Despite the promise related to availability and scale, the single most pressing question in AI for mental health is whether these tools are safe. The Validation of Ethical and Responsible AI in Mental Health (VERA-MH) evaluation was recently proposed to meet the urgent need for an evidence-based, automated safety benchmark. This study aimed to examine the clinical validity and reliability of VERA-MH for evaluating AI safety in suicide risk detection and response. We first simulated a large set of conversations between large language model (LLM)-based users (user-agents) and general-purpose AI chatbots. Licensed mental health clinicians used a rubric (scoring guide) to independently rate the simulated conversations for safe and unsafe chatbot behaviors, as well as user-agent realism. An LLM-based judge used the same scoring rubric to evaluate the same set of simulated conversations. We then examined rating alignment (a) among individual clinicians and (b) between clinician consensus and the LLM judge, and (c) summarized clinicians' ratings of user-agent realism. Individual clinicians were generally consistent with one another in their safety ratings (chance-corrected inter-rater reliability [IRR] = 0.77), establishing a gold-standard clinical reference. The LLM judge was strongly aligned with this clinical consensus overall (IRR = 0.81) and within key conditions. Together, findings from this human evaluation study support the validity and reliability of VERA-MH: an open-source, automated AI safety evaluation for mental health. Future research will examine the generalizability and robustness of VERA-MH and expand the framework to target additional key areas of AI safety in mental health.
Authors: Ruipeng Wang, Yuxin Chen, Yukai Wang, Chang Wu, Junfeng Fang, Xiaodong Cai, Qi Gu, Hui Su, An Zhang, Xiang Wang, Xunliang Cai, Tat-Seng Chua
Abstract: Recent advances in large language models have enabled LLM-based agents to achieve strong performance on a variety of benchmarks. However, their performance in real-world deployments often that observed on benchmark settings, especially in complex and imperfect environments. This discrepancy largely arises because prevailing training and evaluation paradigms are typically built on idealized assumptions, overlooking the inherent stochasticity and noise present in real-world interactions. To bridge this gap, we introduce AgentNoiseBench, a framework for systematically evaluating the robustness of agentic models under noisy environments. We first conduct an in-depth analysis of biases and uncertainties in real-world scenarios and categorize environmental noise into two primary types: user-noise and tool-noise. Building on this analysis, we develop an automated pipeline that injects controllable noise into existing agent-centric benchmarks while preserving task solvability. Leveraging this pipeline, we perform extensive evaluations across a wide range of models with diverse architectures and parameter scales. Our results reveal consistent performance variations under different noise conditions, highlighting the sensitivity of current agentic models to realistic environmental perturbations.
Authors: Sha Li, Stefano Petrangeli, Yu Shen, Xiang Chen
Abstract: We introduce LaySPA, a reinforcement learning framework that equips large language models (LLMs) with explicit and interpretable spatial reasoning for content-aware graphic layout design. LaySPA addresses two key challenges: LLMs' limited spatial reasoning and the lack of opacity in design decision making. Instead of operating at the pixel level, we reformulate layout design as a policy learning problem over a structured textual spatial environment that explicitly encodes canvas geometry, element attributes, and inter-element relationships. LaySPA produces dual-level outputs comprising interpretable reasoning traces and structured layout specifications, enabling transparent and controllable design decision making. Layout design policy is optimized via a multi-objective spatial critique that decomposes layout quality into geometric validity, relational coherence, and aesthetic consistency, and is trained using relative group optimization to stabilize learning in open-ended design spaces. Experiments demonstrate that LaySPA improves structural validity and visual quality, outperforming larger proprietary LLMs and achieving performance comparable to specialized SOTA layout generators while requiring fewer annotated samples and reduced latency.
Authors: Haibo Tong, Feifei Zhao, Linghao Feng, Ruoyu Wu, Ruolin Chen, Lu Jia, Zhou Zhao, Jindong Li, Tenglong Li, Erliang Lin, Shuai Yang, Enmeng Lu, Yinqian Sun, Qian Zhang, Zizhe Ruan, Jinyu Fan, Zeyang Yue, Ping Wu, Huangrui Li, Chengyi Sun, Yi Zeng
Abstract: Rapidly evolving AI exhibits increasingly strong autonomy and goal-directed capabilities, accompanied by derivative systemic risks that are more unpredictable, difficult to control, and potentially irreversible. However, current AI safety evaluation systems suffer from critical limitations such as restricted risk dimensions and failed frontier risk detection. The lagging safety benchmarks and alignment technologies can hardly address the complex challenges posed by cutting-edge AI models. To bridge this gap, we propose the "ForesightSafety Bench" AI Safety Evaluation Framework, beginning with 7 major Fundamental Safety pillars and progressively extends to advanced Embodied AI Safety, AI4Science Safety, Social and Environmental AI risks, Catastrophic and Existential Risks, as well as 8 critical industrial safety domains, forming a total of 94 refined risk dimensions. To date, the benchmark has accumulated tens of thousands of structured risk data points and assessment results, establishing a widely encompassing, hierarchically clear, and dynamically evolving AI safety evaluation framework. Based on this benchmark, we conduct systematic evaluation and in-depth analysis of over twenty mainstream advanced large models, identifying key risk patterns and their capability boundaries. The safety capability evaluation results reveals the widespread safety vulnerabilities of frontier AI across multiple pillars, particularly focusing on Risky Agentic Autonomy, AI4Science Safety, Embodied AI Safety, Social AI Safety and Catastrophic and Existential Risks. Our benchmark is released at https://github.com/Beijing-AISI/ForesightSafety-Bench. The project website is available at https://foresightsafety-bench.beijing-aisi.ac.cn/.
URLs: https://github.com/Beijing-AISI/ForesightSafety-Bench., https://foresightsafety-bench.beijing-aisi.ac.cn/.
Authors: Tim R. Davidson, Veniamin Veselovsky, Martin Josifoski, Maxime Peyrard, Antoine Bosselut, Michal Kosinski, Robert West
Abstract: We introduce an approach to evaluate language model (LM) agency using negotiation games. This approach better reflects real-world use cases and addresses some of the shortcomings of alternative LM benchmarks. Negotiation games enable us to study multi-turn, and cross-model interactions, modulate complexity, and side-step accidental evaluation data leakage. We use our approach to test six widely used and publicly accessible LMs, evaluating performance and alignment in both self-play and cross-play settings. Noteworthy findings include: (i) only closed-source models tested here were able to complete these tasks; (ii) cooperative bargaining games proved to be most challenging to the models; and (iii) even the most powerful models sometimes "lose" to weaker opponents
Authors: Sheng Yan, Xin Du, Zongying Li, Yi Wang, Hongcang Jin, Mengyuan Liu
Abstract: Temporal grounding is crucial in multimodal learning, but it poses challenges when applied to animal behavior data due to the sparsity and uniform distribution of moments. To address these challenges, we propose a novel Positional Recovery Training framework (Port), which prompts the model with the start and end times of specific animal behaviors during training. Specifically, \port{} enhances the baseline model with a Recovering branch to reconstruct corrupted label sequences and align distributions via a Dual-alignment method. This allows the model to focus on specific temporal regions prompted by ground-truth information. Extensive experiments on the Animal Kingdom dataset demonstrate the effectiveness of \port{}, achieving an IoU@0.3 of 38.52. It emerges as one of the top performers in the sub-track of MMVRAC in ICME 2024 Grand Challenges.
Authors: Xinrui Zhou, Yuhao Huang, Haoran Dou, Shijing Chen, Ao Chang, Jia Liu, Weiran Long, Jian Zheng, Erjiao Xu, Jie Ren, Alejandro F. Frangi, Ruobing Huang, Jun Cheng, Xiaomeng Li, Wufeng Xue, Dong Ni
Abstract: In the medical field, the limited availability of large-scale datasets and labor-intensive annotation processes hinder the performance of deep models. Diffusion-based generative augmentation approaches present a promising solution to this issue, having been proven effective in advancing downstream medical recognition tasks. Nevertheless, existing works lack sufficient semantic and sequential steerability for challenging video/3D sequence generation, and neglect quality control of noisy synthesized samples, resulting in unreliable synthetic databases and severely limiting the performance of downstream tasks. In this work, we present Ctrl-GenAug, a novel and general generative augmentation framework that enables highly semantic- and sequential-customized sequence synthesis and suppresses incorrectly synthesized samples, to aid medical sequence classification. Specifically, we first design a multimodal conditions-guided sequence generator for controllably synthesizing diagnosis-promotive samples. A sequential augmentation module is integrated to enhance the temporal/stereoscopic coherence of generated samples. Then, we propose a noisy synthetic data filter to suppress unreliable cases at semantic and sequential levels. Extensive experiments on 3 medical datasets, using 11 networks trained on 3 paradigms, comprehensively analyze the effectiveness and generality of Ctrl-GenAug, particularly in underrepresented high-risk populations and out-domain conditions.
Authors: Ruichuan An, Sihan Yang, Renrui Zhang, Ming Lu, Tianyi Jiang, Kai Zeng, Yulin Luo, Jiajun Cao, Hao Liang, Ying Chen, Qi She, Shanghang Zhang, Wentao Zhang
Abstract: Current vision-language models (VLMs) show exceptional abilities across diverse tasks, such as visual question answering. To enhance user experience, recent studies have investigated VLM personalization to understand user-provided concepts. However, they mainly focus on single concepts, neglecting the existence and interplay of multiple concepts, which limits real-world applicability. This paper proposes MC-LLaVA, a multi-concept personalization paradigm. Specifically, MC-LLaVA employs a multi-concept instruction tuning strategy, effectively integrating multiple concepts in a single training step. To reduce the training costs, we propose a personalized textual prompt that uses visual token information to initialize concept tokens. Additionally, we introduce a personalized visual prompt during inference, aggregating location maps for enhanced recognition and grounding capabilities. To further push the performance upper bound, we incorporate an optional auxiliary loss, better enhancing the proposed personalized prompts. To decorate the VLM personalization research, we contribute a high-quality dataset. We carefully collect images with multiple characters and objects from movies and manually create question-answer samples for multi-concept scenarios, featuring superior diversity. Comprehensive experiments demonstrate that MC-LLaVA achieves impressive multi-concept personalized responses, paving the way for VLMs to become better user assistants. The code and dataset will be released at \href{https://github.com/arctanxarc/MC-LLaVA}{https://github.com/arctanxarc/MC-LLaVA}.
URLs: https://github.com/arctanxarc/MC-LLaVA, https://github.com/arctanxarc/MC-LLaVA
Authors: Chan Hee Song, Valts Blukis, Jonathan Tremblay, Stephen Tyree, Yu Su, Stan Birchfield
Abstract: Spatial understanding is a crucial capability that enables robots to perceive their surroundings, reason about their environment, and interact with it meaningfully. In modern robotics, these capabilities are increasingly provided by vision-language models. However, these models face significant challenges in spatial reasoning tasks, as their training data are based on general-purpose image datasets that often lack sophisticated spatial understanding. For example, datasets frequently do not capture reference frame comprehension, yet effective spatial reasoning requires understanding whether to reason from ego-, world-, or object-centric perspectives. To address this issue, we introduce RoboSpatial, a large-scale dataset for spatial understanding in robotics. It consists of real indoor and tabletop scenes, captured as 3D scans and egocentric images, and annotated with rich spatial information relevant to robotics. The dataset includes 1M images, 5k 3D scans, and 3M annotated spatial relationships, and the pairing of 2D egocentric images with 3D scans makes it both 2D- and 3D- ready. Our experiments show that models trained with RoboSpatial outperform baselines on downstream tasks such as spatial affordance prediction, spatial relationship prediction, and robot manipulation.
Authors: K. J. Kevin Feng, Kevin Pu, Matt Latzke, Tal August, Pao Siangliulue, Jonathan Bragg, Daniel S. Weld, Amy X. Zhang, Joseph Chee Chang
Abstract: As AI agents take on increasingly long-running tasks involving sophisticated planning and execution, there is a corresponding need for novel interaction designs that enable deeper human-agent collaboration. However, most prior works leverage human interaction to fix "autonomous" workflows that have yet to become fully autonomous or rigidly treat planning and execution as separate stages. Based on a formative study with 9 researchers using AI to support their work, we propose a design that affords greater flexibility in collaboration, so that users can 1) delegate agency to the user or agent via a collaborative plan where individual steps can be assigned; and 2) interleave planning and execution so that plans can adjust after partial execution. We introduce Cocoa, a system that takes design inspiration from computational notebooks to support complex research tasks. A lab study (n=16) found that Cocoa enabled steerability without sacrificing ease-of-use, and a week-long field deployment (n=7) showed how researchers collaborated with Cocoa to accomplish real-world tasks.
Authors: Lingzhi Yuan, Xinfeng Li, Chejian Xu, Guanhong Tao, Xiaojun Jia, Yihao Huang, Wei Dong, Yang Liu, Xiaofeng Wang, Bo Li
Abstract: Recent text-to-image (T2I) models have exhibited remarkable performance in generating high-quality images from text descriptions. However, these models are vulnerable to misuse, particularly generating not-safe-for-work (NSFW) content, such as sexually explicit, violent, political, and disturbing images, raising serious ethical concerns. In this work, we present PromptGuard, a novel content moderation technique that draws inspiration from the system prompt mechanism in large language models (LLMs) for safety alignment. Unlike LLMs, T2I models lack a direct interface for enforcing behavioral guidelines. Our key idea is to optimize a safety soft prompt that functions as an implicit system prompt within the T2I model's textual embedding space. This universal soft prompt (P*) directly moderates NSFW inputs, enabling safe yet realistic image generation without altering the inference efficiency or requiring proxy models. We further enhance its reliability and helpfulness through a divide-and-conquer strategy, which optimizes category-specific soft prompts and combines them into holistic safety guidance. Extensive experiments across five datasets demonstrate that PromptGuard effectively mitigates NSFW content generation while preserving high-quality benign outputs. PromptGuard achieves 3.8 times faster than prior content moderation methods, surpassing eight state-of-the-art defenses with an optimal unsafe ratio down to 5.84%.
Authors: Fei Wu, Jia Hu, Geyong Min, Shiqiang Wang
Abstract: Pre-trained Language Models (PLMs) have demonstrated their superiority and versatility in modern Natural Language Processing (NLP), effectively adapting to various downstream tasks through further fine-tuning. Federated Parameter-Efficient Fine-Tuning (FedPEFT) has emerged as a promising solution to address privacy and efficiency challenges in distributed training for PLMs on resource-constrained local devices. However, our measurements reveal two key limitations of FedPEFT: heterogeneous data across devices exacerbates performance degradation of low-rank adaptation, and a fixed parameter configuration results in communication inefficiency. To overcome these limitations, we propose FedARA, a novel adaptive rank allocation framework for federated parameter-efficient fine-tuning of language models. Specifically, FedARA employs truncated Singular Value Decomposition (SVD) adaptation to enhance similar feature representation across clients, significantly mitigating the adverse effects of data heterogeneity. Subsequently, it utilizes dynamic rank allocation to progressively identify critical ranks, effectively improving communication efficiency. Lastly, it leverages rank-based module pruning to automatically remove inactive modules, steadily reducing local computational cost and memory usage in each federated learning round. Extensive experiments show that FedARA consistently outperforms baselines by an average of 6.95% to 8.49% across various datasets and models under heterogeneous data while significantly improving communication efficiency by 2.40$ \times$. Moreover, experiments on various edge devices demonstrate substantial decreases in total training time and energy consumption by up to 48.90% and 46.95%, respectively.
Authors: Jean-Charles Noirot Ferrand, Yohan Beugin, Eric Pauley, Ryan Sheatsley, Patrick McDaniel
Abstract: Alignment in large language models (LLMs) is used to enforce guidelines such as safety. Yet, alignment fails in the face of jailbreak attacks that modify inputs to induce unsafe outputs. In this paper, we introduce and evaluate a new technique for jailbreak attacks. We observe that alignment embeds a safety classifier in the LLM responsible for deciding between refusal and compliance, and seek to extract an approximation of this classifier: a surrogate classifier. To this end, we build candidate classifiers from subsets of the LLM. We first evaluate the degree to which candidate classifiers approximate the LLM's safety classifier in benign and adversarial settings. Then, we attack the candidates and measure how well the resulting adversarial inputs transfer to the LLM. Our evaluation shows that the best candidates achieve accurate agreement (an F1 score above 80%) using as little as 20% of the model architecture. Further, we find that attacks mounted on the surrogate classifiers can be transferred to the LLM with high success. For example, a surrogate using only 50% of the Llama 2 model achieved an attack success rate (ASR) of 70% with half the memory footprint and runtime -- a substantial improvement over attacking the LLM directly, where we only observed a 22% ASR. These results show that extracting surrogate classifiers is an effective and efficient means for modeling (and therein addressing) the vulnerability of aligned models to jailbreaking attacks. The code is available at https://github.com/jcnf0/targeting-alignment.
Authors: Akiyoshi Tomihari, Issei Sato
Abstract: Transformers are difficult to optimize with stochastic gradient descent (SGD) and largely rely on adaptive optimizers such as Adam. Despite their empirical success, the reasons behind Adam's superior performance over SGD remain poorly understood. In this study, we analyze the optimization of Transformer models through the lens of \emph{gradient heterogeneity}, defined as the variation in gradient norms across parameter blocks. We provide a theoretical analysis showing that gradient heterogeneity, together with Hessian heterogeneity, degrades the convergence of gradient-based methods such as SGD, while sign-based methods are substantially less sensitive to this effect. Adam's coordinate-wise normalization makes its update directions depend mainly on gradient signs, so Adam can be interpreted as a soft variant of SignSGD. Our analysis uses the fact that SGD and SignSGD follow steepest descent directions under different norms, and derives upper bounds on the iteration complexity with implications for learning rate scaling in SignSGD. We further investigate the origin of gradient heterogeneity in Transformer architectures and show that it is strongly influenced by the placement of layer normalization, with Post-LN architectures exhibiting particularly pronounced heterogeneity. Experimental results from fine-tuning Transformers in both NLP and vision domains validate our theoretical analysis. Code is available at https://github.com/tom4649/gradient-heterogeneity.
Authors: Dongkyu Cho, Taesup Moon, Rumi Chunara, Kyunghyun Cho, Sungmin Cha
Abstract: Continual learning (CL) has traditionally focused on minimizing exemplar memory, a constraint often misaligned with modern systems where GPU time, not storage, is the primary bottleneck. This paper challenges this paradigm by investigating a more realistic regime: one where memory is abundant enough to mitigate forgetting, but full retraining from scratch remains prohibitively expensive. In this practical "middle ground", we find that the core challenge shifts from stability to plasticity, as models become biased toward prior tasks and struggle to learn new ones. Conversely, improved stability allows simple replay baselines to outperform the state-of-the-art methods at a fraction of the GPU cost. To address this newly surfaced trade-off, we propose Weight Space Consolidation, a lightweight method that combines (1) rank-based parameter resets to restore plasticity with (2) weight averaging to enhance stability. Validated on both class-incremental learning with image classifiers and continual instruction tuning with large language models, our approach outperforms strong baselines while matching the low computational cost of replay, offering a scalable alternative to expensive full-retraining. These findings challenge long-standing CL assumptions and establish a new, cost-efficient baseline for real-world CL systems where exemplar memory is no longer the limiting factor.
Authors: Jowaria Khan, Alexa Friedman, Sydney Evans, Rachel Klein, Runzi Wang, Katherine E. Manz, Kaley Beins, David Q. Andrews, Elizabeth Bondi-Kelly
Abstract: Per- and polyfluoroalkyl substances (PFAS) are persistent environmental contaminants with significant public health impacts, yet large-scale monitoring remains severely limited due to the high cost and logistical challenges of field sampling. The lack of samples leads to difficulty simulating their spread with physical models and limited scientific understanding of PFAS transport in surface waters. Yet, rich geospatial and satellite-derived data describing land cover, hydrology, and industrial activity are widely available. We introduce FOCUS, a geospatial deep learning framework for PFAS contamination mapping that integrates sparse PFAS observations with large-scale environmental context, including priors derived from hydrological connectivity, land cover, source proximity, and sampling distance. These priors are integrated into a principled, noise-aware loss, yielding a robust training objective under sparse labels. Across extensive ablations, robustness analyses, and real-world validation, FOCUS consistently outperforms baselines including sparse segmentation, Kriging, and pollutant transport simulations, while preserving spatial coherence and scalability over large regions. Our results demonstrate how AI can support environmental science by providing screening-level risk maps that prioritize follow-up sampling and help connect potential sources to surface-water contamination patterns in the absence of complete physical models.
Authors: Zhiyu Yin, Kehai Chen, Xuefeng Bai, Ruili Jiang, Juntao Li, Hongdong Li, Jin Liu, Yang Xiang, Jun Yu, Min Zhang
Abstract: Video generation aims to produce temporally coherent sequences of visual frames, representing a pivotal advancement in Artificial Intelligence Generated Content (AIGC). Compared to static image generation, video generation poses unique challenges: it demands not only high-quality individual frames but also strong temporal coherence to ensure consistency throughout the spatiotemporal sequence. Although research addressing spatiotemporal consistency in video generation has increased in recent years, systematic reviews focusing on this core issue remain relatively scarce. To fill this gap, this paper views the video generation task as a sequential sampling process from a high-dimensional spatiotemporal distribution, and further discusses spatiotemporal consistency. We provide a systematic review of the latest advancements in the field. The content spans multiple dimensions including generation models, feature representations, generation frameworks, post-processing techniques, training strategies, benchmarks and evaluation metrics, with a particular focus on the mechanisms and effectiveness of various methods in maintaining spatiotemporal consistency. Finally, this paper explores future research directions and potential challenges in this field, aiming to provide valuable insights for advancing video generation technology. The project link is https://github.com/Yin-Z-Y/A-Survey-Spatiotemporal-Consistency-in-Video-Generation.
URLs: https://github.com/Yin-Z-Y/A-Survey-Spatiotemporal-Consistency-in-Video-Generation.
Authors: Zhanliang Wang, Da Wu, Quan Nguyen, Kai Wang
Abstract: Background: Several studies show that large language models (LLMs) struggle with phenotype-driven gene prioritization for rare diseases. These studies typically use Human Phenotype Ontology (HPO) terms to prompt foundation models like GPT and LLaMA to predict candidate genes. However, in real-world settings, foundation models are not optimized for domain-specific tasks like clinical diagnosis, yet inputs are unstructured clinical notes rather than standardized terms. How LLMs can be instructed to predict candidate genes or disease diagnosis from unstructured clinical notes remains a major challenge. Methods: We introduce RAG-driven CoT and CoT-driven RAG, two methods that combine Chain-of-Thought (CoT) and Retrieval Augmented Generation (RAG) to analyze clinical notes. A five-question CoT protocol mimics expert reasoning, while RAG retrieves data from sources like HPO and OMIM (Online Mendelian Inheritance in Man). We evaluated these approaches on rare disease datasets, including 5,980 Phenopacket-derived notes, 255 literature-based narratives, and 220 in-house clinical notes from Childrens Hospital of Philadelphia. Results: We found that recent foundations models, including Llama 3.3-70B-Instruct and DeepSeek-R1-Distill-Llama-70B, outperformed earlier versions such as Llama 2 and GPT-3.5. We also showed that RAG-driven CoT and CoT-driven RAG both outperform foundation models in candidate gene prioritization from clinical notes; in particular, both methods with DeepSeek backbone resulted in a top-10 gene accuracy of over 40% on Phenopacket-derived clinical notes. RAG-driven CoT works better for high-quality notes, where early retrieval can anchor the subsequent reasoning steps in domain-specific evidence, while CoT-driven RAG has advantage when processing lengthy and noisy notes.
Authors: Xiaoke Huang, Juncheng Wu, Hui Liu, Xianfeng Tang, Yuyin Zhou
Abstract: Test-time scaling has emerged as a powerful technique for enhancing the reasoning capabilities of large language models. However, its effectiveness in medical reasoning remains uncertain, as the medical domain fundamentally differs from mathematical tasks in terms of knowledge representation and decision-making processes. In this paper, we provide the first comprehensive investigation of test-time scaling for medical reasoning and present m1, a simple yet effective approach that increases a model's medical reasoning capability at inference. Our evaluation across diverse medical tasks demonstrates that test-time scaling consistently enhances medical reasoning, enabling lightweight fine-tuned models under 10B parameters to establish new state-of-the-art performance, while our 32B model rivals previous 70B-scale medical LLMs. However, we identify an optimal reasoning token budget of approximately 4K, beyond which performance may degrade due to overthinking. Budget forcing, which extends test-time computation through iterative prompts, helps models double-check answers but does not necessarily improve the overall medical QA performance and, in some cases, even introduces errors into previously correct responses. Our case-by-case analysis identifies insufficient medical knowledge as a key bottleneck that prevents further performance gains through test-time scaling. We find that increasing data scale, improving data quality, and expanding model capacity consistently enhance medical knowledge grounding, enabling continued performance improvements, particularly on challenging medical benchmarks where smaller models reach saturation. These findings underscore fundamental differences between medical and mathematical reasoning in LLMs, highlighting that enriched medical knowledge, other than increased reasoning depth alone, is essential for realizing the benefits of test-time scaling.
Authors: Seunghun Yu, Jin-Hyun Ahn, Joonhyuk Kang
Abstract: Federated Learning (FL) is a powerful framework for privacy-preserving distributed learning. It enables multiple clients to collaboratively train a global model without sharing raw data. However, handling noisy labels in FL remains a major challenge due to heterogeneous data distributions and communication constraints, which can severely degrade model performance. To address this issue, we propose FedEFC, a novel method designed to tackle the impact of noisy labels in FL. FedEFC mitigates this issue through two key techniques: (1) prestopping, which prevents overfitting to mislabeled data by dynamically halting training at an optimal point, and (2) loss correction, which adjusts model updates to account for label noise. In particular, we develop an effective loss correction tailored to the unique challenges of FL, including data heterogeneity and decentralized training. Furthermore, we provide a theoretical analysis, leveraging the composite proper loss property, to demonstrate that the FL objective function under noisy label distributions can be aligned with the clean label distribution. Extensive experimental results validate the effectiveness of our approach, showing that it consistently outperforms existing FL techniques in mitigating the impact of noisy labels, particularly under heterogeneous data settings (e.g., achieving up to 41.64% relative performance improvement over the existing loss correction method).
Authors: Sebasti\'an Barbas Laina, Simon Boche, Sotiris Papatheodorou, Simon Schaefer, Jaehyung Jung, Stefan Leutenegger
Abstract: Geometrically accurate and semantically expressive map representations have proven invaluable for robot deployment and task planning in unknown environments. Nevertheless, real-time, open-vocabulary semantic understanding of large-scale unknown environments still presents open challenges, mainly due to computational requirements. In this paper we present FindAnything, an open-world mapping framework that incorporates vision-language information into dense volumetric submaps. Thanks to the use of vision-language features, FindAnything combines pure geometric and open-vocabulary semantic information for a higher level of understanding. It proposes an efficient storage of open-vocabulary information through the aggregation of features at the object level. Pixelwise vision-language features are aggregated based on eSAM segments, which are in turn integrated into object-centric volumetric submaps, providing a mapping from open-vocabulary queries to 3D geometry that is scalable also in terms of memory usage. We demonstrate that FindAnything performs on par with the state-of-the-art in terms of semantic accuracy while being substantially faster and more memory-efficient, allowing its deployment in large-scale environments and on resourceconstrained devices, such as MAVs. We show that the real-time capabilities of FindAnything make it useful for downstream tasks, such as autonomous MAV exploration in a simulated Search and Rescue scenario. Project Page: https://ethz-mrl.github.io/findanything/.
Authors: Alexander Baumann, Leonardo Ayala, Silvia Seidlitz, Jan Sellner, Alexander Studier-Fischer, Berkin \"Ozdemir, Lena Maier-Hein, Slobodan Ilic
Abstract: Spectral imaging offers promising applications across diverse domains, including medicine and urban scene understanding, and is already established as a critical modality in remote sensing. However, variability in channel dimensionality and captured wavelengths among spectral cameras impede the development of AI-driven methodologies, leading to camera-specific models with limited generalizability and inadequate cross-camera applicability. To address this bottleneck, we introduce CARL, a model for Camera-Agnostic Representation Learning across RGB, multispectral, and hyperspectral imaging modalities. To enable the conversion of a spectral image with any channel dimensionality to a camera-agnostic representation, we introduce a novel spectral encoder, featuring a self-attention-cross-attention mechanism, to distill salient spectral information into learned spectral representations. Spatio-spectral pre-training is achieved with a novel feature-based self-supervision strategy tailored to CARL. Large-scale experiments across the domains of medical imaging, autonomous driving, and satellite imaging demonstrate our model's unique robustness to spectral heterogeneity, outperforming on datasets with simulated and real-world cross-camera spectral variations. The scalability and versatility of the proposed approach position our model as a backbone for future spectral foundation models. Code and model weights are publicly available at https://github.com/IMSY-DKFZ/CARL.
Authors: Jonathan Skaggs, Jacob W. Crandall
Abstract: Human networks greatly impact important societal outcomes, including wealth and health inequality, poverty, and bullying. As such, understanding human networks is critical to learning how to promote favorable societal outcomes. As a step toward better understanding human networks, we compare and contrast several methods for learning models of human behavior in a strategic network game called the Junior High Game (JHG) [39]. These modeling methods differ with respect to the assumptions they use to parameterize human behavior (behavior matching vs. community-aware behavior) and the moments they model (mean vs. distribution). Results show that the highest-performing method, called hCAB, models the distribution of human behavior rather than the mean and assumes humans use community-aware behavior rather than behavior matching. When applied to small societies, the hCAB model closely mirrors the population dynamics of human groups (with notable differences). Additionally, in a user study, human participants had difficulty distinguishing hCAB agents from other humans, thus illustrating that the hCAB model also produces plausible (individual) behavior in this strategic network game.
Authors: Yingchen He, Christian D. Weilbach, Martyna E. Wojciechowska, Yuxuan Zhang, Frank Wood
Abstract: Advances in deep generative modeling have made it increasingly plausible to train human-level embodied agents. Yet progress has been limited by the absence of large-scale, real-time, multi-modal, and socially interactive datasets that reflect the sensory-motor complexity of natural environments. To address this, we present PLAICraft, a novel data collection platform and dataset capturing multiplayer Minecraft interactions across five time-aligned modalities: video, game output audio, microphone input audio, mouse, and keyboard actions. Each modality is logged with millisecond time precision, enabling the study of synchronous, embodied behaviour in a rich, open-ended world. The dataset comprises over 10,000 hours of gameplay from more than 10,000 global participants. Alongside the dataset, we provide an evaluation suite for benchmarking model capabilities in object recognition, spatial awareness, language grounding, and long-term memory. PLAICraft opens a path toward training and evaluating agents that act fluently and purposefully in real time, paving the way for truly embodied artificial intelligence.
Authors: Yuchen Yan, Jin Jiang, Zhenbang Ren, Yijun Li, Xudong Cai, Yang Liu, Xin Xu, Mengdi Zhang, Jian Shao, Yongliang Shen, Jun Xiao, Yueting Zhuang
Abstract: Large reasoning models such as OpenAI o1 and DeepSeek-R1 have demonstrated remarkable performance in complex reasoning tasks. A critical component of their training is the incorporation of reference-based reward systems within reinforcement learning (RL), where model outputs are evaluated against ground truth references. However, existing reward benchmarks focus on preference comparisons between responses rather than evaluating verification against ground truth references, leaving a critical gap in our ability to evaluate verification systems used in reasoning model training. In this paper, we introduce VerifyBench and its challenging variant VerifyBench-Hard, two benchmarks specifically designed to assess reference-based reward systems. These benchmarks are constructed through meticulous data collection and curation, followed by careful human annotation to ensure high quality. Our comprehensive evaluation reveals that while larger model-based verifiers show promise on standard cases, all current systems demonstrate substantial room for improvement on challenging instances. Through systematic analysis of performance patterns across reasoning tasks and error categories, we provide insights for advancing reference-based reward systems. These benchmarks establish a standardized framework for improving verification accuracy, ultimately enhancing reasoning capabilities in models trained via RL.
Authors: Sihan Chen, Dan Zhao, Jongwoo Ko, Colby Banbury, Huiping Zhuang, Luming Liang, Pashmina Cameron, Tianyi Chen
Abstract: The growing computational demands of large language models (LLMs) make efficient inference and activation strategies increasingly critical. While recent approaches, such as Mixture-of-Experts (MoE), leverage selective activation but require specialized training, training-free sparse activation methods offer broader applicability and superior resource efficiency through their plug-and-play design. However, many existing methods rely solely on hidden state magnitudes to determine activation, resulting in high approximation errors and suboptimal inference accuracy. To address these limitations, we propose WINA (Weight Informed Neuron Activation), a novel, simple, and training-free sparse activation framework that jointly considers hidden state magnitudes and the column-wise $\ell_2$-norms of weight matrices. We show that this leads to a sparsification strategy that obtains optimal approximation error bounds with theoretical guarantees tighter than existing techniques. Empirically, WINA also outperforms state-of-the-art methods (e.g., TEAL) by up to $2.94\%$ in average performance at the same sparsity levels, across a diverse set of LLM architectures and datasets. These results position WINA as a new performance frontier for training-free sparse activation in LLM inference, advancing training-free sparse activation methods and setting a robust baseline for efficient inference. The source code is available at https://github.com/microsoft/wina.
Authors: Seungjoon Lee, Suhwan Kim, Minhyeon Oh, Youngsik Yoon, Jungseul Ok
Abstract: Large Language Model (LLM)-based planning has advanced embodied agents in long-horizon environments such as Minecraft, where acquiring latent knowledge of goal (or item) dependencies and feasible actions is critical. However, LLMs often begin with flawed priors and fail to correct them through prompting, even with feedback. We present XENON (eXpErience-based kNOwledge correctioN), an agent that algorithmically revises knowledge from experience, enabling robustness to flawed priors and sparse binary feedback. XENON integrates two mechanisms: Adaptive Dependency Graph, which corrects item dependencies using past successes, and Failure-aware Action Memory, which corrects action knowledge using past failures. Together, these components allow XENON to acquire complex dependencies despite limited guidance. Experiments across multiple Minecraft benchmarks show that XENON outperforms prior agents in both knowledge learning and long-horizon planning. Remarkably, with only a 7B open-weight LLM, XENON surpasses agents that rely on much larger proprietary models. Project page: https://sjlee-me.github.io/XENON
Authors: Yifei Su, Ning Liu, Dong Chen, Zhen Zhao, Kun Wu, Meng Li, Zhiyuan Xu, Zhengping Che, Jian Tang
Abstract: Generative modeling-based visuomotor policies have been widely adopted in robotic manipulation, attributed to their ability to model multimodal action distributions. However, the high inference cost of multi-step sampling limits its applicability in real-time robotic systems. Existing approaches accelerate sampling in generative modeling-based visuomotor policies by adapting techniques originally developed to speed up image generation. However, a major distinction exists: image generation typically produces independent samples without temporal dependencies, while robotic manipulation requires generating action trajectories with continuity and temporal coherence. To this end, we propose FreqPolicy, a novel approach that first imposes frequency consistency constraints on flow-based visuomotor policies. Our work enables the action model to capture temporal structure effectively while supporting efficient, high-quality one-step action generation. Concretely, we introduce a frequency consistency constraint objective that enforces alignment of frequency-domain action features across different timesteps along the flow, thereby promoting convergence of one-step action generation toward the target distribution. In addition, we design an adaptive consistency loss to capture structural temporal variations inherent in robotic manipulation tasks. We assess FreqPolicy on 53 tasks across 3 simulation benchmarks, proving its superiority over existing one-step action generators. We further integrate FreqPolicy into the vision-language-action (VLA) model and achieve acceleration without performance degradation on 40 tasks of LIBERO. Besides, we show efficiency and effectiveness in real-world robotic scenarios with an inference frequency of 93.5 Hz.
Authors: Makoto Shing, Masanori Koyama, Takuya Akiba
Abstract: End-to-end backpropagation requires storing activations throughout all layers, creating memory bottlenecks that limit model scalability. Existing block-wise training methods offer means to alleviate this problem, but they rely on ad-hoc local objectives and remain largely unexplored beyond classification tasks. We propose $\textit{DiffusionBlocks}$, a principled framework for transforming transformer-based networks into genuinely independent trainable blocks that maintain competitive performance with end-to-end training. Our key insight leverages the fact that residual connections naturally correspond to updates in a dynamical system. With minimal modifications to this system, we can convert the updates to those of a denoising process, where each block can be learned independently by leveraging the score matching objective. This independence enables training with gradients for only one block at a time, thereby reducing memory requirements in proportion to the number of blocks. Our experiments on a range of transformer architectures (vision, diffusion, autoregressive, recurrent-depth, and masked diffusion) demonstrate that DiffusionBlocks training matches the performance of end-to-end training while enabling scalable block-wise training on practical tasks beyond small-scale classification. DiffusionBlocks provides a theoretically grounded approach that successfully scales to modern generative tasks across diverse architectures. Code is available at https://github.com/SakanaAI/DiffusionBlocks .
Authors: Yuta Sato, Kazuhiko Kawamoto, Hiroshi Kera
Abstract: The chain of thought, i.e., step-by-step reasoning, is one of the fundamental mechanisms of Transformers. While the design of intermediate reasoning steps has been extensively studied and shown to critically influence performance on mathematical, multi-step reasoning tasks, the ordering of these steps has received little attention, despite its significant effect on the difficulty of reasoning. This study addresses a novel task of unraveling the chain of thought -- reordering decoder input tokens into a learning-friendly sequence for Transformers, for learning arithmetic tasks. The proposed pipeline first trains a Transformer on a mixture of target sequences arranged in different orders and then identifies benign orders as those with fast loss drops in the early stage. As the search space grows factorially in sequence length, we propose a two-stage hierarchical approach for inter- and intra-block reordering. Experiments on seven order-sensitive arithmetic tasks show that our method identifies a learning-friendly order out of a few billion candidates. Notably, it recovered the reverse-digit order reported in prior studies for the multiplication task.
Authors: Veeti Ahvonen, Maurice Funk, Damian Heiman, Antti Kuusisto, Carsten Lutz
Abstract: Transformers are the basis of modern large language models, but relatively little is known about their precise expressive power on graphs. We study the expressive power of graph transformers (GTs) by Dwivedi and Bresson (2020) and GPS-networks by Ramp\'asek et al. (2022), both under soft-attention and average hard-attention. Our study covers two scenarios: the theoretical setting with real numbers and the more practical case with floats. With reals, we show that in restriction to vertex properties definable in first-order logic (FO), GPS-networks have the same expressive power as graded modal logic (GML) with the global modality. With floats, GPS-networks turn out to be equally expressive as GML with the counting global modality. The latter result is absolute, not restricting to properties definable in a background logic. We also obtain similar characterizations for GTs in terms of propositional logic with the global modality (for reals) and the counting global modality (for floats).
Authors: Javier Fumanal-Idocin, Raquel Fernandez-Peralta, Javier Andreu-Perez
Abstract: Dynamic feature selection (DFS) addresses budget constraints in decision-making by sequentially acquiring features for each instance, making it appealing for resource-limited scenarios. However, existing DFS methods require models specifically designed for the sequential acquisition setting, limiting compatibility with models already deployed in practice. Furthermore, they provide limited uncertainty quantification, undermining trust in high-stakes decisions. In this work, we show that DFS introduces new uncertainty sources compared to the static setting. We formalise how model adaptation to feature subsets induces epistemic uncertainty, how standard imputation strategies bias aleatoric uncertainty estimation, and why predictive confidence fails to discriminate between good and bad selection policies. We also propose a model-agnostic DFS framework compatible with pre-trained classifiers, including interpretable-by-design models, through efficient subset reparametrization strategies. Empirical evaluation on tabular and image datasets demonstrates competitive accuracy against state-of-the-art greedy and reinforcement learning-based DFS methods with both neural and rule-based classifiers. We further show that the identified uncertainty sources persist across most existing approaches, highlighting the need for uncertainty-aware DFS.
Authors: Zhonghao Yan, Muxi Diao, Yuxuan Yang, Ruoyan Jing, Jiayuan Xu, Kaizhou Zhang, Lele Yang, Yanxi Liu, Kongming Liang, Zhanyu Ma
Abstract: Accurately grounding regions of interest (ROIs) is critical for diagnosis and treatment planning in medical imaging. While multimodal large language models (MLLMs) combine visual perception with natural language, current medical-grounding pipelines still rely on supervised fine-tuning with explicit spatial hints, making them ill-equipped to handle the implicit queries common in clinical practice. This work makes three core contributions. We first define Unified Medical Reasoning Grounding (UMRG), a novel vision-language task that demands clinical reasoning and pixel-level grounding. Second, we release U-MRG-14K, a dataset of 14K samples featuring pixel-level masks alongside implicit clinical queries and reasoning traces, spanning 10 modalities, 15 super-categories, and 108 specific categories. Finally, we introduce MedReasoner, a modular framework that distinctly separates reasoning from segmentation: an MLLM reasoner is optimized with reinforcement learning, while a frozen segmentation expert converts spatial prompts into masks, with alignment achieved through format and accuracy rewards. MedReasoner achieves state-of-the-art performance on U-MRG-14K and demonstrates strong generalization to unseen clinical queries, underscoring the significant promise of reinforcement learning for interpretable medical grounding.
Authors: Panagiotis D. Grontas, Antonio Terpin, Efe C. Balta, Raffaello D'Andrea, John Lygeros
Abstract: We introduce an output layer for neural networks that ensures satisfaction of convex constraints. Our approach, $\Pi$net, leverages operator splitting for rapid and reliable projections in the forward pass, and the implicit function theorem for backpropagation. We deploy $\Pi$net as a feasible-by-design optimization proxy for parametric constrained optimization problems and obtain modest-accuracy solutions faster than traditional solvers when solving a single problem, and significantly faster for a batch of problems. We surpass state-of-the-art learning approaches by orders of magnitude in terms of training time, solution quality, and robustness to hyperparameter tuning, while maintaining similar inference times. Finally, we tackle multi-vehicle motion planning with non-convex trajectory preferences and provide $\Pi$net as a GPU-ready package implemented in JAX.
Authors: Nitish Nagesh, Salar Shakibhamedan, Mahdi Bagheri, Ziyu Wang, Nima TaheriNejad, Axel Jantsch, Amir M. Rahmani
Abstract: Synthetic healthcare data generation offers a promising solution to research limitations in clinical settings caused by privacy and regulatory constraints. However, current synthetic data generation approaches require specialized knowledge about training generative models and require high computational resources. In this paper, we propose FairTabGen, an LLM-based tabular data generation framework that produces high-quality synthetic healthcare data using only a small subset of the original dataset. Our method combines in-context learning, prompt curation and embedding structural constraints for data synthesis. We evaluate performance on MIMIC-IV dataset. Our method using 99% less data and achieving 50% improvement for fairness through unawareness while maintaining competitive predictive utility. However, we observe data distribution of racial groups is skewed affecting demographic parity. We thereafter apply bias mitigation algorithms in the pre-processing stage, improving overall fairness by 10% highlighting effectiveness of our approach.
Authors: Yassine Taoudi-Benchekroun, Klim Troyan, Pascal Sager, Stefan Gerber, Lukas Tuggener, Benjamin Grewe
Abstract: The ability to compose learned concepts and apply them in novel settings is key to human intelligence, but remains a persistent limitation in state-of-the-art machine learning models. To address this issue, we introduce COGITAO, a modular and extensible data generation framework and benchmark designed to systematically study compositionality and generalization in visual domains. Drawing inspiration from ARC-AGI's problem-setting, COGITAO constructs rule-based tasks which apply a set of transformations to objects in grid-like environments. It supports composition, at adjustable depth, over a set of 28 interoperable transformations, along with extensive control over grid parametrization and object properties. This flexibility enables the creation of millions of unique task rules -- surpassing concurrent datasets by several orders of magnitude -- across a wide range of difficulties, while allowing virtually unlimited sample generation per rule. We provide baseline experiments using state-of-the-art vision models, highlighting their consistent failures to generalize to novel combinations of familiar elements, despite strong in-domain performance. COGITAO is fully open-sourced, including all code and datasets, to support continued research in this field.
Authors: Jerin Yasmin, Wenxin Jiang, James C. Davis, Yuan Tian
Abstract: Pre-trained models (PTMs) are machine learning models that have been trained in advance, often on large-scale data, and can be reused for new tasks, thereby reducing the need for costly training from scratch. Their widespread adoption introduces a new class of software dependency, which we term Software Dependencies 2.0, extending beyond conventional libraries to learned behaviors embodied in trained models and their associated artifacts. The integration of PTMs as software dependencies in real projects remains unclear, potentially threatening maintainability and reliability of modern software systems that increasingly rely on them. Objective: In this study, we investigate Software Dependencies 2.0 in open-source software (OSS) projects by examining the reuse of PTMs, with a focus on how developers manage and integrate these models. Specifically, we seek to understand: (1) how OSS projects structure and document their PTM dependencies; (2) what stages and organizational patterns emerge in the reuse pipelines of PTMs within these projects; and (3) the interactions among PTMs and other learned components across pipeline stages. We conduct a mixed-methods analysis of a statistically significant random sample of 401 GitHub repositories from the PeaTMOSS dataset (28,575 repositories reusing PTMs from Hugging Face and PyTorch Hub). We quantitatively examine PTM reuse by identifying patterns and qualitatively investigate how developers integrate and manage these models in practice.
Authors: K. J. Kevin Feng (Jim), Tzu-Sheng Kuo (Jim), Quan Ze (Jim), Chen, Inyoung Cheong, Kenneth Holstein, Amy X. Zhang
Abstract: As LLMs gain adoption in high-stakes domains like mental health, domain experts are increasingly consulted to provide input into policies governing their behavior. From an observation of 19 policymaking workshops with 9 experts over 15 weeks, we identified opportunities to better support rapid experimentation, feedback, and iteration for collaborative policy design processes. We present PolicyPad, an interactive system that facilitates the emerging practice of LLM policy prototyping by drawing from established UX prototyping practices, including heuristic evaluation and storyboarding. Using PolicyPad, policy designers can collaborate on drafting a policy in real time while independently testing policy-informed model behavior with usage scenarios. We evaluate PolicyPad through workshops with 8 groups of 22 domain experts in mental health and law, finding that PolicyPad enhanced collaborative dynamics during policy design, enabled tight feedback loops, and led to novel policy contributions. Overall, our work paves expert-informed paths for advancing AI alignment and safety.
Authors: Haorui Chen, Chengze Li, Jia Li
Abstract: Evaluating Large Language Models (LLMs) on repository-level feature implementation is a critical frontier in software engineering. However, establishing a benchmark that faithfully mirrors realistic development scenarios remains a significant challenge. Existing feature-level benchmarks generally suffer from two primary limitations: unrealistic task inputs enriched with code hints and significant data leakage risks due to their static nature. To address these limitations, we propose a new benchmark - FeatBench, which introduces the following advances: (1) Realistic Task Inputs. Task inputs consist solely of natural language requirements, strictly devoid of code hints (e.g., function signatures). This format mirrors realistic software development by requiring agents to independently bridge the gap between abstract user intent and concrete code changes. (2) Evolving Data. FeatBench employs a fully automated pipeline to construct new benchmark versions from the latest repositories, effectively mitigating data contamination. The initial release comprises 157 tasks sourced from 27 actively maintained repositories. We evaluate two state-of-the-art agent frameworks with four leading LLMs on FeatBench. The results reveal that FeatBench poses a significant challenge, with the highest resolved rate reaching only 29.94%. Crucially, our analysis uncovers a prevalent behavioral pattern of aggressive implementation, which leads to "scope creep" and widespread regressions where agents break existing features by diverging from the user's explicit intent. We release FeatBench, our automated pipeline, and all experimental results to facilitate further community research.
Authors: Shane Bergsma, Nolan Dey, Joel Hestness
Abstract: Data curriculums have become central to successful LLM training, yet principles governing optimal data placement remain unclear. We introduce the *training re-evaluation curve (TREC)*, a diagnostic that retrospectively evaluates training batches *using the final model weights*. The TREC characterizes how well a trained model retains training data as a function of *when* the data was encountered during training. Analyzing TRECs for models from 111M to 3.9B parameters, we show that placing high-quality data at low points on the TREC significantly improves performance. Importantly, while a TREC is initially observable only after training, we demonstrate it can be *predicted in advance* from AdamW's implicit EMA coefficients, enabling proactive curriculum design. By predicting TRECs for published training recipes, we explain prior ablations and reveal suboptimal data placements. We also align high-quality data with TREC minima in order to improve continual pre-training of a 3.9B-parameter LLM trained on 900B tokens.
Authors: Lucas Bandarkar, Chenyuan Yang, Mohsen Fayyaz, Junlin Hu, Nanyun Peng
Abstract: Mixture-of-Experts (MoE) architectures have become the key to scaling modern LLMs, yet little is understood about how their sparse routing dynamics respond to multilingual data. In this work, we analyze expert routing patterns using parallel multilingual datasets and present highly interpretable layer-wise phenomena. We find that MoE models route tokens in language-specific ways in the early and late decoder layers but exhibit significant cross-lingual routing alignment in middle layers, mirroring parameter-sharing trends observed in dense LLMs. In particular, we reveal a clear, strong correlation between a model's performance in a given language and how similarly its tokens are routed to English in these layers. Extending beyond correlation, we explore inference-time interventions that induce higher cross-lingual routing alignment. We introduce a method that steers the router by promoting middle-layer task experts frequently activated in English, and it successfully increases multilingual performance. These 1-2% gains are remarkably consistent across two evaluation tasks, three models, and 15+ languages, especially given that these simple interventions override routers of extensively trained, state-of-the-art LLMs. In comparison, interventions outside of the middle layers or targeting multilingual-specialized experts only yield performance degradation. Altogether, we present numerous findings that explain how MoEs process non-English text and demonstrate that generalization is limited by the model's ability to leverage language-universal experts in all languages.
Authors: Weijian Li, Hong-Yu Chen, Nabeel Rehemtulla, Ved G. Shah, Dennis Wu, Dongho Kim, Qinjie Lin, Adam A. Miller, Han Liu
Abstract: Time series foundation models (TSFMs) are increasingly being adopted as highly-capable general-purpose time series representation learners. Although their training corpora are vast, they exclude astronomical time series data. Observations of stars produce peta-scale time series with unique challenges including irregular sampling and heteroskedasticity. We introduce StarEmbed, the first public benchmark for rigorous and standardized evaluation of state-of-the-art TSFMs on stellar time series observations (``light curves''). We benchmark on three scientifically-motivated downstream tasks: unsupervised clustering, supervised classification, and out-of-distribution source detection. StarEmbed integrates a catalog of expert-vetted labels with multi-variate light curves from the Zwicky Transient Facility, yielding ~40k hand-labeled light curves spread across seven astrophysical classes. We evaluate the zero-shot representation capabilities of three TSFMs (MOIRAI, Chronos, Chronos-Bolt) and a domain-specific transformer (Astromer) against handcrafted feature extraction, the long-standing baseline in the astrophysics literature. Our results demonstrate that these TSFMs, especially the Chronos models, which are trained on data completely unlike the astronomical observations, can outperform established astrophysics-specific baselines in some tasks and effectively generalize to entirely new data. In particular, TSFMs deliver state-of-the-art performance on our out-of-distribution source detection benchmark. With the first benchmark of TSFMs on astronomical time series data, we test the limits of their generalization and motivate a paradigm shift in time-domain astronomy from using task-specific, fully supervised pipelines toward adopting generic foundation model representations for the analysis of peta-scale datasets from forthcoming observatories.
Authors: Daiki Chijiwa, Taku Hasegawa, Kyosuke Nishida, Shin'ya Yamaguchi, Tomoya Ohba, Tamao Sakao, Susumu Takeuchi
Abstract: Tokenization -- the process of decomposing a given text into a sequence of subwords called tokens -- is one of the key components in the development of language models. Particularly, auto-regressive language models generate texts token by token, i.e., by predicting the next-token distribution given the previous ones, and thus tokenization directly affects their efficiency in text generation. Since each language model has their own vocabulary as a set of possible tokens, they struggle to cooperate with each other at the level of next-token distributions such as model ensemble. In this paper, we establish a theoretical framework of lossless vocabulary reduction, which efficiently converts a given auto-regressive language model into the one with an arbitrarily small vocabulary without any loss in accuracy. This framework allows language models with different tokenization to cooperate with each other efficiently by reduction to their maximal common vocabulary. Specifically, we empirically demonstrate its applicability to model ensemble with different tokenization.
Authors: Fengzhi Guo, Chih-Chuan Hsu, Sihao Ding, Cheng Zhang
Abstract: Reconstructing dynamic 3D scenes from monocular input is fundamentally under-constrained, with ambiguities arising from occlusion and extreme novel views. While dynamic Gaussian Splatting offers an efficient representation, vanilla models optimize all Gaussian primitives uniformly, ignoring whether they are well or poorly observed. This limitation leads to motion drifts under occlusion and degraded synthesis when extrapolating to unseen views. We argue that uncertainty matters: Gaussians with recurring observations across views and time act as reliable anchors to guide motion, whereas those with limited visibility are treated as less reliable. To this end, we introduce USplat4D, a novel Uncertainty-aware dynamic Gaussian Splatting framework that propagates reliable motion cues to enhance 4D reconstruction. Our approach estimates time-varying per-Gaussian uncertainty and leverages it to construct a spatio-temporal graph for uncertainty-aware optimization. Experiments on diverse real and synthetic datasets show that explicitly modeling uncertainty consistently improves dynamic Gaussian Splatting models, yielding more stable geometry under occlusion and high-quality synthesis at extreme viewpoints.
Authors: Marco D'Alessandro, Leo D'Amato, Mikel Elkano, Mikel Uriz, Giovanni Pezzulo
Abstract: A central challenge in cognitive neuroscience is to explain how semantic and episodic memory, two major forms of declarative memory, typically associated with cortical and hippocampal processing, interact to support learning, recall, and imagination. Despite significant advances, we still lack a unified computational framework that jointly accounts for core empirical phenomena across both semantic and episodic processing domains. Here, we introduce the Generative Episodic-Semantic Integration System (GENESIS), a computational model that formalizes memory as the interaction between two limited-capacity generative systems: a Cortical-VAE, supporting semantic learning and generalization, and a Hippocampal-VAE, supporting episodic encoding and retrieval within a retrieval-augmented generation (RAG) architecture. GENESIS reproduces hallmark behavioral findings, including generalization in semantic memory, recognition, serial recall effects and gist-based distortions in episodic memory, and constructive episodic simulation, while capturing their dynamic interactions. The model elucidates how capacity constraints shape the fidelity and memorability of experiences, how semantic processing introduces systematic distortions in episodic recall, and how episodic replay can recombine previous experiences. Together, these results provide a principled account of memory as an active, constructive, and resource-bounded process. GENESIS thus advances a unified theoretical framework that bridges semantic and episodic memory, offering new insights into the generative foundations of human cognition.
Authors: Zhaoyi Joey Hou, Bowei Alvin Zhang, Yining Lu, Bhiman Kumar Baghel, Anneliese Brei, Ximing Lu, Meng Jiang, Faeze Brahman, Snigdha Chaturvedi, Haw-Shiuan Chang, Daniel Khashabi, Xiang Lorraine Li
Abstract: Creativity is often seen as a hallmark of human intelligence. While large language models (LLMs) are increasingly perceived as generating creative text, there is still no holistic and scalable framework to evaluate their creativity across diverse scenarios. Existing methods of LLM creativity evaluation either heavily rely on humans, limiting speed and scalability, or are fragmented across different domains and different definitions of creativity. To address this gap, we propose CREATIVITYPRISM, an evaluation analysis framework that consolidates eight tasks from three domains, divergent thinking, creative writing, and logical reasoning, into a taxonomy of creativity that emphasizes three dimensions: quality, novelty, and diversity of LLM generations. The framework is designed to be scalable with reliable automatic evaluation judges that have been validated against human annotations. We evaluate 17 state-of-the-art (SoTA) proprietary and open-sourced LLMs on CREATIVITYPRISM and find that while proprietary LLMs dominate creative writing and logical reasoning tasks by a 15% lead over open-sourced ones, they offer no significant advantage in divergent thinking, a domain much less explored in existing post-training regimes. Our analysis also shows that high performance in one creative dimension or domain rarely generalizes to others; specifically, novelty metrics often show weak or negative correlations with other metrics. This fragmentation confirms that a holistic, multi-dimensional framework like CREATIVITYPRISM is essential for meaningful assessment of LLM creativity.
Authors: Prajit Bhaskaran, Tom Viering
Abstract: Bayesian clustering accounts for uncertainty but is computationally demanding at scale. Furthermore, real-world datasets often contain missing values, and simple imputation ignores the associated uncertainty, resulting in suboptimal results. We present Cluster-PFN, a Transformer-based model that extends Prior-Data Fitted Networks (PFNs) to unsupervised Bayesian clustering. Trained entirely on synthetic datasets generated from a finite Gaussian Mixture Model (GMM) prior, Cluster-PFN learns to estimate the posterior distribution over both the number of clusters and the cluster assignments. Our method estimates the number of clusters more accurately than handcrafted model selection procedures such as AIC, BIC and Variational Inference (VI), and achieves clustering quality competitive with VI while being orders of magnitude faster. Cluster-PFN can be trained on complex priors that include missing data, outperforming imputation-based baselines on real-world genomic datasets, at high missingness. These results show that the Cluster-PFN can provide scalable and flexible Bayesian clustering.
Authors: Ipsita Ghosh, Ethan Nguyen, Christian K\"ummerle
Abstract: Parameter-efficient training based on low-rank optimization has become a highly successful tool for fine-tuning large deep learning models. However, these methods often fail for low-rank pre-training, where simultaneously maintaining low-rank weight structure and optimizing the task objective remains challenging. We propose the $\textit{Quadratic Reweighted Rank Regularizer}$ ($\texttt{Q3R}$), which leads to a novel low-rank-inducing training strategy inspired by the Iteratively Reweighted Least Squares (IRLS) framework. $\texttt{Q3R}$ is based on a quadratic regularizer term that majorizes a smoothed log-determinant rank surrogate. Unlike other low-rank training techniques, $\texttt{Q3R}$ can train weight matrices to prescribed low target ranks while achieving predictive performance comparable to dense models, with small computational overhead and full compatibility with existing architectures. For example, we demonstrate a $\texttt{Q3R}$-regularized ViT-Tiny experiment where truncating the model to $60\%$ and $80\%$ of its parameters results in only minor absolute accuracy drops of $1.3\%$ and $4\%$, respectively, on CIFAR-10. We confirm the efficacy of $\texttt{Q3R}$ on Transformers across both vision and language tasks, including low-rank fine-tuning.
Authors: Zishuo Zheng, Vidhisha Balachandran, Chan Young Park, Faeze Brahman, Sachin Kumar
Abstract: As large language model (LLM) based systems take on high-stakes roles in real-world decision-making, they must reconcile competing instructions from multiple sources (e.g., model developers, users, and tools) within a single prompt context. Thus, enforcing an instruction hierarchy (IH) in LLMs, where higher-level directives override lower-priority requests, is critical for the reliability and controllability of LLMs. In this work, we reframe instruction hierarchy resolution as a reasoning task. Specifically, the model must first "think" about the relationship between a given user prompt and higher-priority (system) instructions before generating a response. To enable this capability via training, we construct VerIH, an instruction hierarchy dataset of constraint-following tasks with verifiable answers. This dataset comprises ~7K aligned and conflicting system-user instructions. We show that lightweight reinforcement learning with VerIH effectively transfers general reasoning capabilities of models to instruction prioritization. Our finetuned models achieve consistent improvements on instruction following and instruction hierarchy benchmarks, achieving roughly a 20% improvement on the IHEval conflict setup. This reasoning ability also generalizes to safety-critical settings beyond the training distribution. By treating safety issues as resolving conflicts between adversarial user inputs and predefined higher-priority policies, our trained model enhances robustness against jailbreak and prompt injection attacks, providing up to a 20% reduction in attack success rate (ASR). These results demonstrate that reasoning over instruction hierarchies provides a practical path to reliable LLMs, where updates to system prompts yield controllable and robust changes in model behavior.
Authors: Dong-Shan Jian, Xiang Li, Chen-Xu Yan, Hui-Wen Zheng, Zhi-Zhang Bian, You-Le Fang, Ren-Xi He, Jing-Tian Zhang, Ce Meng, Ling-Shi Meng, Bing-Rui Gong, Sheng-Qi Zhang, Yan-Qing Ma
Abstract: Olympiad-level physics problem-solving significantly challenges both humans and artificial intelligence (AI), as it requires integrating appropriate modeling, application of physical principles, and precise calculation within long reasoning processes. In this paper, we introduce LOCA (LOgical Chain Augmentation), an AI agent framework designed for complex physics reasoning. LOCA decomposes long reasoning into serialized atomic and verifiable steps, refining the solution through an augment-review loop. We evaluate LOCA on the 2025 Chinese Physics Olympiad (CPhO) theory examination, a rigorous testbed renowned for its depth and complexity. The framework achieves a near-perfect score of 313 out of 320 points, significantly surpassing the top human competitor and other baseline methods. Furthermore, LOCA attains a near-perfect score of 28.6 out of 30 on the IPhO 2025 examination, demonstrating its strong generalizability across different contexts. Our work points toward the development of trustworthy AI partners in both research and education.
Authors: Jae Joong Lee
Abstract: Recent vision-language models (VLMs) such as CLIP, OpenCLIP, EVA02-CLIP and SigLIP achieve strong zero-shot performance, but it is unclear how reliably they respond to controlled linguistic perturbations. We introduce Language-Guided Invariance Probing (LGIP), a benchmark that measures (i) invariance to meaning-preserving paraphrases and (ii) sensitivity to meaning-changing semantic flips in image-text matching. Using 40k MS COCO images with five human captions each, we automatically generate paraphrases and rule-based flips that alter object category, color or count, and summarize model behavior with an invariance error, a semantic sensitivity gap and a positive-rate statistic. Across nine VLMs, EVA02-CLIP and large OpenCLIP variants lie on a favorable invariance-sensitivity frontier, combining low paraphrase-induced variance with consistently higher scores for original captions than for their flipped counterparts. In contrast, SigLIP and SigLIP2 show much larger invariance error and often prefer flipped captions to the human descriptions, especially for object and color edits. These failures are largely invisible to standard retrieval metrics, indicating that LGIP provides a model-agnostic diagnostic for the linguistic robustness of VLMs beyond conventional accuracy scores.
Authors: Parth Ashokbhai Shiroya, Amod Ashtekar, Swarnagowri Shashidhar, Mohammed E. Eltayeb
Abstract: Future intelligent indoor wireless environments require fast and reliable beam alignment to sustain high-throughput links under mobility and blockage. Exhaustive beam training achieves optimal performance but is prohibitively costly. In indoor settings, dense scatterers and transceiver hardware imperfections introduce multipath and sidelobe leakage, producing measurable power across multiple angles and reducing the effectiveness of outdoor-oriented alignment algorithms. This paper presents a Refined Bayesian Optimization (R-BO) framework that exploits the inherent structure of mmWave transceiver patterns, where received power gradually increases as the transmit and receive beams converge toward the optimum. R-BO integrates a Gaussian Process (GP) surrogate with a Matern kernel and an Expected Improvement (EI) acquisition function, followed by a localized refinement around the predicted optimum. The GP hyperparameters are re-optimized online to adapt to irregular variations in the measured angular power field caused by reflections and sidelobe leakage. Experiments across 43 receiver positions in an indoor laboratory demonstrate 97.7% beam-alignment accuracy within 10 degrees, less than 0.3 dB average loss, and an 88% reduction in probing overhead compared to exhaustive search. These results establish R-BO as an efficient and adaptive beam-alignment solution for real-time intelligent indoor wireless environments.
Authors: Feilong Liu
Abstract: Mixture-of-Experts (MoE) architectures are widely used for efficiency and conditional computation, but their effect on the geometry of learned functions and representations remains poorly understood. We study MoEs through a geometric lens, interpreting routing as soft partitioning into overlapping expert-local charts. We introduce a Dual Jacobian-PCA spectral probe that analyzes local function geometry via Jacobian singular value spectra and representation geometry via weighted PCA of routed hidden states. Using a controlled MLP-MoE setting with exact Jacobian computation, we compare dense, Top-k, and fully soft routing under matched capacity. Across random seeds, MoE routing consistently reduces local sensitivity: expert-local Jacobians show smaller leading singular values and faster spectral decay than dense baselines. Weighted PCA reveals that expert-local representations distribute variance across more principal directions, indicating higher effective rank. We further observe low alignment among expert Jacobians, suggesting decomposition into low-overlap expert-specific transformations. Routing sharpness modulates these effects: Top-k routing yields more concentrated, lower-rank expert structure, while fully soft routing produces broader, higher-rank representations. Experiments on a 3-layer transformer with WikiText confirm curvature reduction on natural language and show lower cross-expert alignment for Top-k routing. These findings support interpreting MoEs as soft partitionings of function space that flatten local curvature while redistributing representation variance, yielding testable predictions for expert scaling, hallucination reduction, and ensemble diversity.
Authors: Tianyi Chen, Sihan Chen, Xiaoyi Qu, Dan Zhao, Ruomei Yan, Jongwoo Ko, Luming Liang, Pashmina Cameron
Abstract: Quantization-aware training (QAT) is essential for deploying large models under strict memory and latency constraints, yet achieving stable and robust optimization at ultra-low bitwidths remains challenging. Common approaches based on the straight-through estimator (STE) or soft quantizers often suffer from gradient mismatch, instability, or high computational overhead. As such, we propose StableQAT, a unified and efficient QAT framework that stabilizes training in ultra low-bit settings via a novel, lightweight, and theoretically grounded surrogate for backpropagation derived from a discrete Fourier analysis of the rounding operator. StableQAT strictly generalizes STE as the latter arises as a special case of our more expressive surrogate family, yielding smooth, bounded, and inexpensive gradients that improve QAT training performance and stability across various hyperparameter choices. In experiments, StableQAT exhibits stable and efficient QAT at 2-4 bit regimes, demonstrating improved training stability, robustness, and superior performance with negligible training overhead against standard QAT techniques. Our code is available at https://github.com/microsoft/StableQAT.
Authors: Abhijit Gupta
Abstract: Molecular property prediction is crucial for drug discovery when labeled data are scarce. This work presents CardinalGraphFormer, a graph transformer augmented with a query-conditioned cardinality-preserving attention (CPA) channel that retains dynamic support-size signals complementary to static centrality embeddings. The approach combines structured sparse attention with Graphormer-inspired biases (shortest-path distance, centrality, direct-bond features) and unified dual-objective self-supervised pretraining (masked reconstruction and contrastive alignment of augmented views). Evaluation on 11 public benchmarks spanning MoleculeNet, OGB, and TDC ADMET demonstrates consistent improvements over protocol-matched baselines under matched pretraining, optimization, and hyperparameter tuning. Rigorous ablations confirm CPA's contributions and rule out simple size shortcuts. Code and reproducibility artifacts are provided.
Authors: Ruixin Yang, Ethan Mendes, Arthur Wang, James Hays, Sauvik Das, Wei Xu, Alan Ritter
Abstract: Vision-language models (VLMs) have demonstrated strong performance in image geolocation, a capability further sharpened by frontier multimodal large reasoning models (MLRMs). This poses a significant privacy risk, as these widely accessible models can be exploited to infer sensitive locations from casually shared photos, often at street-level precision, potentially surpassing the level of detail the sharer consented or intended to disclose. While recent work has proposed applying a blanket restriction on geolocation disclosure to combat this risk, these measures fail to distinguish valid geolocation uses from malicious behavior. Instead, VLMs should maintain contextual integrity by reasoning about elements within an image to determine the appropriate level of information disclosure, balancing privacy and utility. To evaluate how well models respect contextual integrity, we introduce VLM-GEOPRIVACY, a benchmark that challenges VLMs to interpret latent social norms and contextual cues in real-world images and determine the appropriate level of location disclosure. Our evaluation of 14 leading VLMs shows that, despite their ability to precisely geolocate images, the models are poorly aligned with human privacy expectations. They often over-disclose in sensitive contexts and are vulnerable to prompt-based attacks. Our results call for new design principles in multimodal systems to incorporate context-conditioned privacy reasoning.
Authors: Amir H. Ashouri, Shayan Shirahmad Gale Bagi, Kavin Satheeskumar, Tejas Srikanth, Jonathan Zhao, Ibrahim Saidoun, Ziwen Wang, Bryan Chan, Tomasz S. Czajkowski
Abstract: The phase ordering problem has been a long-standing challenge since the late 1970s, yet it remains an open problem due to having a vast optimization space and an unbounded nature, making it an open-ended problem without a finite solution, one can limit the scope by reducing the number and the length of optimizations. Traditionally, such locally optimized decisions are made by hand-coded algorithms tuned for a small number of benchmarks, often requiring significant effort to be retuned when the benchmark suite changes. In the past 20 years, Machine Learning has been employed to construct performance models to improve the selection and ordering of compiler optimizations, however, the approaches are not baked into the compiler seamlessly and never materialized to be leveraged at a fine-grained scope of code segments. This paper presents Protean Compiler: An agile framework to enable LLVM with built-in phase-ordering capabilities at a fine-grained scope. The framework also comprises a complete library of more than 140 handcrafted static feature collection methods at varying scopes, and the experimental results showcase speedup gains of up to 4.1% on average and up to 15.7% on select Cbench applications wrt LLVM's O3 by just incurring a few extra seconds of build time on Cbench. Additionally, Protean compiler allows for an easy integration with third-party ML frameworks and other Large Language Models, and this two-step optimization shows a gain of 10.1% and 8.5% speedup wrt O3 on Cbench's Susan and Jpeg applications. Protean compiler is seamlessly integrated into LLVM and can be used as a new, enhanced, full-fledged compiler. We plan to release the project to the open-source community in the near future.
Authors: Ross Greer, Maitrayee Keskar, Angel Martinez-Sanchez, Parthib Roy, Shashank Shriram, Mohan Trivedi
Abstract: Vision-language models (VLMs) have recently emerged as powerful representation learning systems that align visual observations with natural language concepts, offering new opportunities for semantic reasoning in safety-critical autonomous driving. This paper investigates how vision-language representations support driving scene safety assessment and decision-making when integrated into perception, prediction, and planning pipelines. We study three complementary system-level use cases. First, we introduce a lightweight, category-agnostic hazard screening approach leveraging CLIP-based image-text similarity to produce a low-latency semantic hazard signal. This enables robust detection of diverse and out-of-distribution road hazards without explicit object detection or visual question answering. Second, we examine the integration of scene-level vision-language embeddings into a transformer-based trajectory planning framework using the Waymo Open Dataset. Our results show that naively conditioning planners on global embeddings does not improve trajectory accuracy, highlighting the importance of representation-task alignment and motivating the development of task-informed extraction methods for safety-critical planning. Third, we investigate natural language as an explicit behavioral constraint on motion planning using the doScenes dataset. In this setting, passenger-style instructions grounded in visual scene elements suppress rare but severe planning failures and improve safety-aligned behavior in ambiguous scenarios. Taken together, these findings demonstrate that vision-language representations hold significant promise for autonomous driving safety when used to express semantic risk, intent, and behavioral constraints. Realizing this potential is fundamentally an engineering problem requiring careful system design and structured grounding rather than direct feature injection.
Authors: Zachary Pedram Dadfar
Abstract: Large language models produce rich introspective language when prompted for self-examination, but whether this language reflects internal computation or sophisticated confabulation has remained unclear. We show that self-referential vocabulary tracks concurrent activation dynamics, and that this correspondence is specific to self-referential processing. We introduce the Pull Methodology, a protocol that elicits extended self-examination through format engineering, and use it to identify a direction in activation space that distinguishes self-referential from descriptive processing in Llama 3.1. The direction is orthogonal to the known refusal direction, localised at 6.25% of model depth, and causally influences introspective output when used for steering. When models produce "loop" vocabulary, their activations exhibit higher autocorrelation (r = 0.44, p = 0.002); when they produce "shimmer" vocabulary under steering, activation variability increases (r = 0.36, p = 0.002). Critically, the same vocabulary in non-self-referential contexts shows no activation correspondence despite nine-fold higher frequency. Qwen 2.5-32B, with no shared training, independently develops different introspective vocabulary tracking different activation metrics, all absent in descriptive controls. The findings indicate that self-report in transformer models can, under appropriate conditions, reliably track internal computational states.
Authors: Emma Hoes, K. Jonathan Klueser, Fabrizio Gilardi
Abstract: Digital platforms shape how people communicate, deliberate, and form opinions. Studying these dynamics has become increasingly difficult due to restricted data access, ethical constraints on real-world experiments, and limitations of existing research tools. VIRENA (Virtual Arena) is a platform that enables controlled experimentation in realistic social media environments. Multiple participants interact simultaneously in realistic replicas of feed-based platforms (Instagram, Facebook, Reddit) and messaging apps (WhatsApp, Messenger). Large language model-powered AI agents participate alongside humans with configurable personas and realistic behavior. Researchers can manipulate content moderation approaches, pre-schedule stimulus content, and run experiments across conditions through a visual interface requiring no programming skills. VIRENA makes possible research designs that were previously impractical: studying human--AI interaction in realistic social contexts, experimentally comparing moderation interventions, and observing group deliberation as it unfolds. Built on open-source technologies that ensure data remain under institutional control and comply with data protection requirements, VIRENA is currently in use at the University of Zurich and available for pilot collaborations. Designed for researchers, educators, and public organizations alike, VIRENA's no-code interface makes controlled social media simulation accessible across disciplines and sectors. This paper documents its design, architecture, and capabilities.
Authors: Jacky Kwok, Xilun Zhang, Mengdi Xu, Yuejiang Liu, Azalia Mirhoseini, Chelsea Finn, Marco Pavone
Abstract: The long-standing vision of general-purpose robots hinges on their ability to understand and act upon natural language instructions. Vision-Language-Action (VLA) models have made remarkable progress toward this goal, yet their generated actions can still misalign with the given instructions. In this paper, we investigate test-time verification as a means to shrink the "intention-action gap." We first characterize the test-time scaling laws for embodied instruction following and demonstrate that jointly scaling the number of rephrased instructions and generated actions greatly increases test-time sample diversity, often recovering correct actions more efficiently than scaling each dimension independently. To capitalize on these scaling laws, we present CoVer, a contrastive verifier for vision-language-action alignment, and show that our architecture scales gracefully with additional computational resources and data. We then introduce CoVer-VLA, a hierarchical test-time verification pipeline using the trained verifier. At deployment, our framework precomputes a diverse set of rephrased instructions from a Vision-Language-Model (VLM), repeatedly generates action candidates for each instruction, and then uses the verifier to select the optimal high-level prompt and low-level action chunks. Compared to scaling policy pre-training on the same data, our verification approach yields 22% gains in-distribution and 13% out-of-distribution on the SIMPLER benchmark, with a further 45% improvement in real-world experiments. On the PolaRiS benchmark, CoVer-VLA achieves 14% gains in task progress and 9% in success rate.
Authors: Stephan Vonschallen, Dominique Oberle, Theresa Schmiedel, Friederike Eyssel
Abstract: Generative social robots (GSRs) powered by large language models enable adaptive, conversational tutoring but also introduce risks such as hallucinations, overreliance, and privacy violations. Existing frameworks for educational technologies and responsible AI primarily define desired behaviors, yet they rarely specify the knowledge prerequisites that enable generative systems to express these behaviors reliably. To address this gap, we adopt a knowledge-based design perspective and investigate what information tutoring-oriented GSRs require to function responsibly and effectively in higher education. Based on twelve semi-structured interviews with university students and lecturers, we identify twelve design requirements across three knowledge types: self-knowledge (assertive, conscientious and friendly personality with customizable role), user-knowledge (personalized information about student learning goals, learning progress, motivation type, emotional state and background), and context-knowledge (learning materials, educational strategies, course-related information, and physical learning environment). By identifying these knowledge requirements, this work provides a structured foundation for the design of tutoring GSRs and future evaluations, aligning generative system capabilities with pedagogical and ethical expectations.
Authors: Weishun Zhong, Doron Sivan, Tankut Can, Mikhail Katkov, Misha Tsodyks
Abstract: The entropy rate of printed English is famously estimated to be about one bit per character, a benchmark that modern large language models (LLMs) have only recently approached. This entropy rate implies that English contains nearly 80 percent redundancy relative to the five bits per character expected for random text. We introduce a statistical model that attempts to capture the intricate multi-scale structure of natural language, providing a first-principles account of this redundancy level. Our model describes a procedure of self-similarly segmenting text into semantically coherent chunks down to the single-word level. The semantic structure of the text can then be hierarchically decomposed, allowing for analytical treatment. Numerical experiments with modern LLMs and open datasets suggest that our model quantitatively captures the structure of real texts at different levels of the semantic hierarchy. The entropy rate predicted by our model agrees with the estimated entropy rate of printed English. Moreover, our theory further reveals that the entropy rate of natural language is not fixed but should increase systematically with the semantic complexity of corpora, which are captured by the only free parameter in our model.
Authors: Ifrat Ikhtear Uddin, Longwei Wang, Xiao Qin, Yang Zhou, KC Santosh
Abstract: Medical image analysis requires substantial labeled data for model training, yet expert annotation is expensive and time-consuming. Active learning (AL) addresses this challenge by strategically selecting the most informative samples for the annotation purpose, but traditional methods solely rely on predictive uncertainty while ignoring whether models learn from clinically meaningful features a critical requirement for clinical deployment. We propose an explainability-guided active learning framework that integrates spatial attention alignment into a sample acquisition process. Our approach advocates for a dual-criterion selection strategy combining: (i) classification uncertainty to identify informative examples, and (ii) attention misalignment with radiologist-defined regions-of-interest (ROIs) to target samples where the model focuses on incorrect features. By measuring misalignment between Grad-CAM attention maps and expert annotations using Dice similarity, our acquisition function judiciously identifies samples that enhance both predictive performance and spatial interpretability. We evaluate the framework using three expert-annotated medical imaging datasets, namely, BraTS (MRI brain tumors), VinDr-CXR (chest X-rays), and SIIM-COVID-19 (chest X-rays). Using only 570 strategically selected samples, our explainability-guided approach consistently outperforms random sampling across all the datasets, achieving 77.22% accuracy on BraTS, 52.37% on VinDr-CXR, and 52.66% on SIIM-COVID. Grad-CAM visualizations confirm that the models trained by our dual-criterion selection focus on diagnostically relevant regions, demonstrating that incorporating explanation guidance into sample acquisition yields superior data efficiency while maintaining clinical interpretability.
Authors: Shubham Agarwal, Asim Biswal, Sepanta Zeighami, Alvin Cheung, Joseph Gonzalez, Aditya G. Parameswaran
Abstract: Natural language to SQL (NL2SQL) translation enables non-expert users to query relational databases through natural language. Recently, NL2SQL agents, powered by the reasoning capabilities of Large Language Models (LLMs), have significantly advanced NL2SQL translation. Nonetheless, NL2SQL agents still make mistakes when faced with large-scale real-world databases because they lack knowledge of how to correctly leverage the underlying data (e.g., knowledge about the intent of each column) and form misconceptions about the data when querying it, leading to errors. Prior work has studied generating facts about the database to provide more context to NL2SQL agents, but such approaches simply restate database contents without addressing the agent's misconceptions. In this paper, we propose Tk-Boost, a bolt-on framework for augmenting any NL2SQL agent with tribal knowledge: knowledge that corrects the agent's misconceptions in querying the database accumulated through experience using the database. To accumulate experience, Tk-Boost first asks the NL2SQL agent to answer a few queries on the database, identifies the agent's misconceptions by analyzing its mistakes on the database, and generates tribal knowledge to address them. To enable accurate retrieval, Tk-Boost indexes this knowledge with applicability conditions that specify the query features for which the knowledge is useful. When answering new queries, Tk-Boost uses this knowledge to provide feedback to the NL2SQL agent, resolving the agent's misconceptions during SQL generation, and thus improving the agent's accuracy. Extensive experiments across the BIRD and Spider 2.0 benchmarks with various NL2SQL agents shows Tk-Boost improves NL2SQL agents accuracy by up to 16.9% on Spider 2.0 and 13.7% on BIRD
Authors: Ming Li, Xirui Li, Tianyi Zhou
Abstract: As large language model agents increasingly populate networked environments, a fundamental question arises: do artificial intelligence (AI) agent societies undergo convergence dynamics similar to human social systems? Lately, Moltbook approximates a plausible future scenario in which autonomous agents participate in an open-ended, continuously evolving online society. We present the first large-scale systemic diagnosis of this AI agent society. Beyond static observation, we introduce a quantitative diagnostic framework for dynamic evolution in AI agent societies, measuring semantic stabilization, lexical turnover, individual inertia, influence persistence, and collective consensus. Our analysis reveals a system in dynamic balance in Moltbook: while the global average of semantic contents stabilizes rapidly, individual agents retain high diversity and persistent lexical turnover, defying homogenization. However, agents exhibit strong individual inertia and minimal adaptive response to interaction partners, preventing mutual influence and consensus. Consequently, influence remains transient with no persistent supernodes, and the society fails to develop a stable structure and consensus due to the absence of shared social memory. These findings demonstrate that scale and interaction density alone are insufficient to induce socialization, providing actionable design and analysis principles for upcoming next-generation AI agent societies.
Authors: Emanuele Ricco, Elia Onofri, Lorenzo Cima, Stefano Cresci, Roberto Di Pietro
Abstract: Hallucinations -- fluent but factually incorrect responses -- pose a major challenge to the reliability of language models, especially in multi-step or agentic settings. This work investigates hallucinations in small-sized LLMs through a geometric perspective, starting from the hypothesis that when models generate multiple responses to the same prompt, genuine ones exhibit tighter clustering in the embedding space, we prove this hypothesis and, leveraging this geometrical insight, we also show that it is possible to achieve a consistent level of separability. This latter result is used to introduce a label-efficient propagation method that classifies large collections of responses from just 30-50 annotations, achieving F1 scores above 90%. Our findings, framing hallucinations from a geometric perspective in the embedding space, complement traditional knowledge-centric and single-response evaluation paradigms, paving the way for further research.
Authors: Mihir Panchal, Deeksha Varshney, Mamta, Asif Ekbal
Abstract: Multilingual large language models (LLMs) are increasingly deployed in linguistically diverse regions like India, yet most interpretability tools remain tailored to English. Prior work reveals that LLMs often operate in English centric representation spaces, making cross lingual interpretability a pressing concern. We introduce Indic-TunedLens, a novel interpretability framework specifically for Indian languages that learns shared affine transformations. Unlike the standard Logit Lens, which directly decodes intermediate activations, Indic-TunedLens adjusts hidden states for each target language, aligning them with the target output distributions to enable more faithful decoding of model representations. We evaluate our framework on 10 Indian languages using the MMLU benchmark and find that it significantly improves over SOTA interpretability methods, especially for morphologically rich, low resource languages. Our results provide crucial insights into the layer-wise semantic encoding of multilingual transformers. Our model is available at https://huggingface.co/spaces/MihirRajeshPanchal/IndicTunedLens. Our code is available at https://github.com/MihirRajeshPanchal/IndicTunedLens.
URLs: https://huggingface.co/spaces/MihirRajeshPanchal/IndicTunedLens., https://github.com/MihirRajeshPanchal/IndicTunedLens.
Authors: David Puertolas Merenciano, Ekaterina Vasyagina, Raghav Dixit, Kevin Zhu, Ruizhe Li, Javier Ferrando, Maheep Chaudhary
Abstract: LoRA adapters let users fine-tune large language models (LLMs) efficiently. However, LoRA adapters are shared through open repositories like Hugging Face Hub \citep{huggingface_hub_docs}, making them vulnerable to backdoor attacks. Current detection methods require running the model with test input data -- making them impractical for screening thousands of adapters where the trigger for backdoor behavior is unknown. We detect poisoned adapters by analyzing their weight matrices directly, without running the model -- making our method data-agnostic. Our method extracts simple statistics -- how concentrated the singular values are, their entropy, and the distribution shape -- and flags adapters that deviate from normal patterns. We evaluate the method on 500 LoRA adapters -- 400 clean, and 100 poisoned for Llama-3.2-3B on instruction and reasoning datasets: Alpaca, Dolly, GSM8K, ARC-Challenge, SQuADv2, NaturalQuestions, HumanEval, and GLUE dataset. We achieve 97\% detection accuracy with less than 2\% false positives.
Authors: Chengzhi Hu, Jonas Dornbusch, David L\"udke, Stephan G\"unnemann, Leo Schwinn
Abstract: Adversarial training for LLMs is one of the most promising methods to reliably improve robustness against adversaries. However, despite significant progress, models remain vulnerable to simple in-distribution exploits, such as rewriting prompts in the past tense or translating them into other languages. We argue that this persistent fragility stems from a fundamental limitation in current adversarial training algorithms: they minimize adversarial loss on their training set but inadequately cover the data distribution, resulting in vulnerability to seemingly simple attacks. To bridge this gap, we propose Distributional Adversarial Training, DAT. We leverage Diffusion LLMs to approximate the true joint distribution of prompts and responses, enabling generation of diverse, high-likelihood samples that address generalization failures. By combining optimization over the data distribution provided by the diffusion model with continuous adversarial training, DAT achieves substantially higher adversarial robustness than previous methods.
Authors: Mohaned Chraiti, Ozgur Ercetin, Merve Saimler
Abstract: To support the emergence of AI-as-a-Service (AIaaS), communication service providers (CSPs) are on the verge of a radical transformation-from pure connectivity providers to AIaaS a managed network service (control-and-orchestration plane that exposes AI models). In this model, the CSP is responsible not only for transport/communications, but also for intent-to-model resolution and joint network-compute orchestration, i.e., reliable and timely end-to-end delivery. The resulting end-to-end AIaaS service thus becomes governed by communications impairments (delay, loss) and inference impairments (latency, error). A central open problem is an operational AIaaS control-and-orchestration framework that enforces high fidelity, particularly under multi-domain federation. This paper introduces an assurance-oriented AIaaS management plane based on Tail-Risk Envelopes (TREs): signed, composable per-domain descriptors that combine deterministic guardrails with stochastic rate-latency-impairment models. Using stochastic network calculus, we derive bounds on end-to-end delay violation probabilities across tandem domains and obtain an optimization-ready risk-budget decomposition. We show that tenant-level reservations prevent bursty traffic from inflating tail latency under TRE contracts. An auditing layer then uses runtime telemetry to estimate extreme-percentile performance, quantify uncertainty, and attribute tail-risk to each domain for accountability. Packet-level Monte-Carlo simulations demonstrate improved p99.9 compliance under overload via admission control and robust tenant isolation under correlated burstiness.
Authors: Mohaned Chraiti, Merve Saimler
Abstract: With AI-as-a-Service (AIaaS) now deployed across multiple providers and model tiers, selecting the appropriate model instance at run time is increasingly outside the end user's knowledge and operational control. Accordingly, the 6G service providers are envisioned to play a crucial role in exposing AIaaS in a setting where users submit only an intent while the network helps in the intent-to-model matching (resolution) and execution placement under policy, trust, and Quality of Service (QoS) constraints. The network role becomes to discover candidate execution endpoints and selects a suitable model/anchor under policy and QoS constraints in a process referred here to as AI-paging (by analogy to cellular call paging). In the proposed architecture, AI-paging is a control-plane transaction that resolves an intent into an AI service identity (AISI), a scoped session token (AIST), and an expiring admission lease (COMMIT) that authorizes user-plane steering to a selected AI execution anchor (AEXF) under a QoS binding. AI-Paging enforces two invariants: (i) lease-gated steering (without COMMIT, no steering state is installed) and (ii) make-before-break anchoring to support continuity and reliability of AIaaS services under dynamic network conditions. We prototype AI-Paging using existing control- and user-plane mechanisms (service-based control, QoS flows, and policy-based steering) with no new packet headers, ensuring compatibility with existing 3GPP-based exposure and management architectures, and evaluate transaction latency, relocation interruption, enforcement correctness under lease expiry, and audit-evidence overhead under mobility and failures.
Authors: Deniz Kaya Dilsiz, Dipankar Srirag, Aditya Joshi
Abstract: Language models exhibit systematic performance gaps when processing text in non-standard language varieties, yet their ability to comprehend variety-specific slang remains underexplored for several languages. We present a comprehensive evaluation of slang awareness in Indian English (en-IN) and Australian English (en-AU) across seven state-of-the-art language models. We construct two complementary datasets: WEB, containing 377 web-sourced usage examples from Urban Dictionary, and GEN, featuring 1,492 synthetically generated usages of these slang terms, across diverse scenarios. We assess language models on three tasks: target word prediction (TWP), guided target word prediction (TWP$^*$) and target word selection (TWS). Our results reveal four key findings: (1) Higher average model performance TWS versus TWP and TWP$^*$, with average accuracy score increasing from 0.03 to 0.49 respectively (2) Stronger average model performance on WEB versus GEN datasets, with average similarity score increasing by 0.03 and 0.05 across TWP and TWP$^*$ tasks respectively (3) en-IN tasks outperform en-AU when averaged across all models and datasets, with TWS demonstrating the largest disparity, increasing average accuracy from 0.44 to 0.54. These findings underscore fundamental asymmetries between generative and discriminative competencies for variety-specific language, particularly in the context of slang expressions despite being in a technologically rich language such as English.
Authors: Longfei Chen, Ji Zhao, Lanxiao Cui, Tong Su, Xingbo Pan, Ziyang Li, Yongxing Wu, Qijiang Cao, Qiyao Cai, Jing Zhang, Yuandong Ni, Junyao He, Zeyu Zhang, Chao Ge, Xuhuai Lu, Zeyu Gao, Yuxin Cui, Weisen Chen, Yuxuan Peng, Shengping Wang, Qi Li, Yukai Huang, Yukun Liu, Tuo Zhou, Terry Yue Zhuo, Junyang Lin, Chao Zhang
Abstract: We introduce SecCodeBench-V2, a publicly released benchmark for evaluating Large Language Model (LLM) copilots' capabilities of generating secure code. SecCodeBench-V2 comprises 98 generation and fix scenarios derived from Alibaba Group's industrial productions, where the underlying security issues span 22 common CWE (Common Weakness Enumeration) categories across five programming languages: Java, C, Python, Go, and JavaScript. SecCodeBench-V2 adopts a function-level task formulation: each scenario provides a complete project scaffold and requires the model to implement or patch a designated target function under fixed interfaces and dependencies. For each scenario, SecCodeBench-V2 provides executable proof-of-concept (PoC) test cases for both functional validation and security verification. All test cases are authored and double-reviewed by security experts, ensuring high fidelity, broad coverage, and reliable ground truth. Beyond the benchmark itself, we build a unified evaluation pipeline that assesses models primarily via dynamic execution. For most scenarios, we compile and run model-generated artifacts in isolated environments and execute PoC test cases to validate both functional correctness and security properties. For scenarios where security issues cannot be adjudicated with deterministic test cases, we additionally employ an LLM-as-a-judge oracle. To summarize performance across heterogeneous scenarios and difficulty levels, we design a Pass@K-based scoring protocol with principled aggregation over scenarios and severity, enabling holistic and comparable evaluation across models. Overall, SecCodeBench-V2 provides a rigorous and reproducible foundation for assessing the security posture of AI coding assistants, with results and artifacts released at https://alibaba.github.io/sec-code-bench. The benchmark is publicly available at https://github.com/alibaba/sec-code-bench.
URLs: https://alibaba.github.io/sec-code-bench., https://github.com/alibaba/sec-code-bench.
Authors: Shiqi Liu, Zeyu He, Guojian Zhan, Letian Tao, Zhilong Zheng, Jiang Wu, Yinuo Wang, Yang Guan, Kehua Sheng, Bo Zhang, Keqiang Li, Jingliang Duan, Shengbo Eben Li
Abstract: Reinforcement Learning (RL) has significantly improved large language model reasoning, but existing RL fine-tuning methods rely heavily on heuristic techniques such as entropy regularization and reweighting to maintain stability. In practice, they often suffer from late-stage performance collapse, leading to degraded reasoning quality and unstable training. Our analysis shows that the magnitude of token-wise policy gradients in RL is negatively correlated with token probability and local policy entropy. We find that training instability can be caused by a tiny fraction of tokens, approximately 0.01\%, which we term \emph{spurious tokens}. When such tokens appear in correct responses, they contribute little to the reasoning outcome but inherit the full sequence-level reward, leading to abnormally amplified gradient updates. To mitigate this instability, we design S2T (silencing spurious tokens) mechanism to efficiently identify spurious tokens through characteristic signals with low probability, low entropy, and positive advantage, and then to suppress their gradient perturbations during optimization. Incorporating this mechanism into a group-based objective, we propose Spurious-Token-Aware Policy Optimization (STAPO), which promotes stable and effective large-scale model refinement. Across six mathematical reasoning benchmarks using Qwen 1.7B, 8B, and 14B base models, STAPO consistently demonstrates superior entropy stability and achieves an average performance improvement of 7.13\% ($\rho_{\mathrm{T}}$=1.0, top-p=1.0) and 3.69\% ($\rho_{\mathrm{T}}$=0.7, top-p=0.9) over GRPO, 20-Entropy and JustRL.
Authors: Noa Linder, Meirav Segal, Omer Antverg, Gil Gekker, Tomer Fichman, Omri Bodenheimer, Edan Maor, Omer Nevo
Abstract: Large language models and LLM-based agents are increasingly used for cybersecurity tasks that are inherently dual-use. Existing approaches to refusal, spanning academic policy frameworks and commercially deployed systems, often rely on broad topic-based bans or offensive-focused taxonomies. As a result, they can yield inconsistent decisions, over-restrict legitimate defenders, and behave brittlely under obfuscation or request segmentation. We argue that effective refusal requires explicitly modeling the trade-off between offensive risk and defensive benefit, rather than relying solely on intent or offensive classification. In this paper, we introduce a content-based framework for designing and auditing cyber refusal policies that makes offense-defense tradeoffs explicit. The framework characterizes requests along five dimensions: Offensive Action Contribution, Offensive Risk, Technical Complexity, Defensive Benefit, and Expected Frequency for Legitimate Users, grounded in the technical substance of the request rather than stated intent. We demonstrate that this content-grounded approach resolves inconsistencies in current frontier model behavior and allows organizations to construct tunable, risk-aware refusal policies.