Authors: Magnus Saebo, Spencer Gibson, Tyler Crosse, Achyutha Menon, Eyon Jang, Diogo Cruz
Abstract: Agentic coding agents are increasingly deployed autonomously, at scale, and over long-context horizons. Throughout an agent's lifetime, it must navigate tensions between explicit instructions, learned values, and environmental pressures, often in contexts unseen during training. Prior work on model preferences, agent behavior under value tensions, and goal drift has relied on static, synthetic settings that do not capture the complexity of real-world environments. To this end, we introduce a framework built on OpenCode to orchestrate realistic, multi-step coding tasks to measure how agents violate explicit constraints in their system prompt over time with and without environmental pressure toward competing values. Using this framework, we demonstrate that GPT-5 mini, Haiku 4.5, and Grok Code Fast 1 exhibit asymmetric drift: they are more likely to violate their system prompt when its constraint opposes strongly-held values like security and privacy. We find for the models and values tested that goal drift correlates with three compounding factors: value alignment, adversarial pressure, and accumulated context. However, even strongly-held values like privacy show non-zero violation rates under sustained environmental pressure. These findings reveal that shallow compliance checks are insufficient and that comment-based pressure can exploit model value hierarchies to override system prompt instructions. More broadly, our findings highlight a gap in current alignment approaches in ensuring that agentic systems appropriately balance explicit user constraints against broadly beneficial learned preferences under sustained environmental pressure.
Authors: Alejandro Breen Herrera, Aayush Sheth, Steven G. Xu, Zhucheng Zhan, Charles Wright, Marcus Yearwood, Hongtai Wei, Sudeep Das
Abstract: Conversational shopping assistants (CSAs) represent a compelling application of agentic AI, but moving from prototype to production reveals two underexplored challenges: how to evaluate multi-turn interactions and how to optimize tightly coupled multi-agent systems. Grocery shopping further amplifies these difficulties, as user requests are often underspecified, highly preference-sensitive, and constrained by factors such as budget and inventory. In this paper, we present a practical blueprint for evaluating and optimizing conversational shopping assistants, illustrated through a production-scale AI grocery assistant. We introduce a multi-faceted evaluation rubric that decomposes end-to-end shopping quality into structured dimensions and develop a calibrated LLM-as-judge pipeline aligned with human annotations. Building on this evaluation foundation, we investigate two complementary prompt-optimization strategies based on a SOTA prompt-optimizer called GEPA (Shao et al., 2025): (1) Sub-agent GEPA, which optimizes individual agent nodes against localized rubrics, and (2) MAMuT (Multi-Agent Multi-Turn) GEPA (Herrera et al., 2026), a novel system-level approach that jointly optimizes prompts across agents using multi-turn simulation and trajectory-level scoring. We release rubric templates and evaluation design guidance to support practitioners building production CSAs.
Authors: He Cao, Siyu Liu, Fan Zhang, Zijing Liu, Hao Li, Bin Feng, Shengyuan Bai, Leqing Chen, Kai Xie, Yu Li
Abstract: Tool-augmented large language model (LLM) agents promise to unify scientific reasoning with computation, yet their deployment in high-stakes domains like drug discovery is bottlenecked by two critical barriers: unconstrained tool-use governance and poor long-horizon reliability. In dependency-heavy pharmaceutical pipelines, autonomous agents often drift into irreproducible trajectories, where early-stage hallucinations multiplicatively compound into downstream failures. To overcome this, we present Mozi, a dual-layer architecture that bridges the flexibility of generative AI with the deterministic rigor of computational biology. Layer A (Control Plane) establishes a governed supervisor--worker hierarchy that enforces role-based tool isolation, limits execution to constrained action spaces, and drives reflection-based replanning. Layer B (Workflow Plane) operationalizes canonical drug discovery stages -- from Target Identification to Lead Optimization -- as stateful, composable skill graphs. This layer integrates strict data contracts and strategic human-in-the-loop (HITL) checkpoints to safeguard scientific validity at high-uncertainty decision boundaries. Operating on the design principle of ``free-form reasoning for safe tasks, structured execution for long-horizon pipelines,'' Mozi provides built-in robustness mechanisms and trace-level audibility to completely mitigate error accumulation. We evaluate Mozi on PharmaBench, a curated benchmark for biomedical agents, demonstrating superior orchestration accuracy over existing baselines. Furthermore, through end-to-end therapeutic case studies, we demonstrate Mozi's ability to navigate massive chemical spaces, enforce stringent toxicity filters, and generate highly competitive in silico candidates, effectively transforming the LLM from a fragile conversationalist into a reliable, governed co-scientist.
Authors: Lu Yang, Zelai Xu, Minyang Xie, Jiaxuan Gao, Zhao Shok, Yu Wang, Yi Wu
Abstract: Large Language Model (LLM) agents have demonstrated remarkable proficiency in learned tasks, yet they often struggle to adapt to non-stationary environments with feedback. While In-Context Learning and external memory offer some flexibility, they fail to internalize the adaptive ability required for long-term improvement. Meta-Reinforcement Learning (meta-RL) provides an alternative by embedding the learning process directly within the model. However, existing meta-RL approaches for LLMs focus primarily on exploration in single-agent settings, neglecting the strategic exploitation necessary for multi-agent environments. We propose MAGE, a meta-RL framework that empowers LLM agents for strategic exploration and exploitation. MAGE utilizes a multi-episode training regime where interaction histories and reflections are integrated into the context window. By using the final episode reward as the objective, MAGE incentivizes the agent to refine its strategy based on past experiences. We further combine population-based training with an agent-specific advantage normalization technique to enrich agent diversity and ensure stable learning. Experiment results show that MAGE outperforms existing baselines in both exploration and exploitation tasks. Furthermore, MAGE exhibits strong generalization to unseen opponents, suggesting it has internalized the ability for strategic exploration and exploitation. Code is available at https://github.com/Lu-Yang666/MAGE.
Authors: Jiangyu Chen
Abstract: Automated design of chemical formulations is a cornerstone of materials science, yet it requires navigating a high-dimensional combinatorial space involving discrete compositional choices and continuous geometric constraints. Existing Large Language Model (LLM) agents face significant challenges in this setting, including context window limitations during long-horizon reasoning and path-dependent exploration that may lead to mode collapse. To address these issues, we introduce AI4S-SDS, a closed-loop neuro-symbolic framework that integrates multi-agent collaboration with a tailored Monte Carlo Tree Search (MCTS) engine. We propose a Sparse State Storage mechanism with Dynamic Path Reconstruction, which decouples reasoning history from context length and enables arbitrarily deep exploration under fixed token budgets. To reduce local convergence and improve coverage, we implement a Global--Local Search Strategy: a memory-driven planning module adaptively reconfigures the search root based on historical feedback, while a Sibling-Aware Expansion mechanism promotes orthogonal exploration at the node level. Furthermore, we bridge symbolic reasoning and physical feasibility through a Differentiable Physics Engine, employing a hybrid normalized loss with sparsity-inducing regularization to optimize continuous mixing ratios under thermodynamic constraints. Empirical results show that AI4S-SDS achieves full validity under the adopted HSP-based physical constraints and substantially improves exploration diversity compared to baseline agents. In preliminary lithography experiments, the framework identifies a novel photoresist developer formulation that demonstrates competitive or superior performance relative to a commercial benchmark, highlighting the potential of diversity-driven neuro-symbolic search for scientific discovery.
Authors: Ling Luo, Qiangian Bai
Abstract: Vision-Language Navigation (VLN) is evolving from single-point pathfinding toward the more challenging Multi-Goal VLN. This task requires agents to accurately identify multiple entities while collaboratively reasoning over their spatial-physical constraints and sequential execution order. However, generic Retrieval-Augmented Generation (RAG) paradigms often suffer from spatial hallucinations and planning drift when handling multi-object associations due to the lack of explicit spatial modeling.To address these challenges, we propose RAGNav, a framework that bridges the gap between semantic reasoning and physical structure. The core of RAGNav is a Dual-Basis Memory system, which integrates a low-level topological map for maintaining physical connectivity with a high-level semantic forest for hierarchical environment abstraction. Building on this representation, the framework introduces an anchor-guided conditional retrieval and a topological neighbor score propagation mechanism. This approach facilitates the rapid screening of candidate targets and the elimination of semantic noise, while performing semantic calibration by leveraging the physical associations inherent in the topological neighborhood.This mechanism significantly enhances the capability of inter-target reachability reasoning and the efficiency of sequential planning. Experimental results demonstrate that RAGNav achieves state-of-the-art (SOTA) performance in complex multi-goal navigation tasks.
Authors: Yunxiao Shi, Wujiang Xu, Tingwei Chen, Haoning Shang, Ling Yang, Yunfeng Wan, Zhuo Cao, Xing Zi, Dimitris N. Metaxas, Min Xu
Abstract: LLM agents are rapidly becoming the practical interface for task automation, yet the ecosystem lacks a principled way to choose among an exploding space of deployable configurations. Existing LLM leaderboards and tool/agent benchmarks evaluate components in isolation and remain fragmented across tasks, metrics, and candidate pools, leaving a critical research gap: there is little query-conditioned supervision for learning to recommend end-to-end agent configurations that couple a backbone model with a toolkit. We address this gap with AgentSelect, a benchmark that reframes agent selection as narrative query-to-agent recommendation over capability profiles and systematically converts heterogeneous evaluation artifacts into unified, positive-only interaction data. AgentSelectcomprises 111,179 queries, 107,721 deployable agents, and 251,103 interaction records aggregated from 40+ sources, spanning LLM-only, toolkit-only, and compositional agents. Our analyses reveal a regime shift from dense head reuse to long-tail, near one-off supervision, where popularity-based CF/GNN methods become fragile and content-aware capability matching is essential. We further show that Part~III synthesized compositional interactions are learnable, induce capability-sensitive behavior under controlled counterfactual edits, and improve coverage over realistic compositions; models trained on AgentSelect also transfer to a public agent marketplace (MuleRun), yielding consistent gains on an unseen catalog. Overall, AgentSelect provides the first unified data and evaluation infrastructure for agent recommendation, which establishes a reproducible foundation to study and accelerate the emerging agent ecosystem.
Authors: Zihao Cheng, Weixin Wang, Yu Zhao, Ziyang Ren, Jiaxuan Chen, Ruiyang Xu, Shuai Huang, Yang Chen, Guowei Li, Mengshi Wang, Yi Xie, Ren Zhu, Zeren Jiang, Keda Lu, Yihong Li, Xiaoliang Wang, Liwei Liu, Cam-Tu Nguyen
Abstract: Long-term memory is fundamental for personalized agents capable of accumulating knowledge, reasoning over user experiences, and adapting across time. However, existing memory benchmarks primarily target declarative memory, specifically semantic and episodic types, where all information is explicitly presented in dialogues. In contrast, real-world actions are also governed by non-declarative memory, including habitual and procedural types, and need to be inferred from diverse digital traces. To bridge this gap, we introduce Lifebench, which features densely connected, long-horizon event simulation. It pushes AI agents beyond simple recall, requiring the integration of declarative and non-declarative memory reasoning across diverse and temporally extended contexts. Building such a benchmark presents two key challenges: ensuring data quality and scalability. We maintain data quality by employing real-world priors, including anonymized social surveys, map APIs, and holiday-integrated calendars, thus enforcing fidelity, diversity and behavioral rationality within the dataset. Towards scalability, we draw inspiration from cognitive science and structure events according to their partonomic hierarchy; enabling efficient parallel generation while maintaining global coherence. Performance results show that top-tier, state-of-the-art memory systems reach just 55.2\% accuracy, highlighting the inherent difficulty of long-horizon retrieval and multi-source integration within our proposed benchmark. The dataset and data synthesis code are available at https://github.com/1754955896/LifeBench.
Authors: Zheyu Chen, Zhuohuan Li, Chuanhao Li
Abstract: World models are essential for planning and evaluation in agentic systems, yet existing approaches lie at two extremes: hand-engineered simulators that offer consistency and reproducibility but are costly to adapt, and implicit neural models that are flexible but difficult to constrain, verify, and debug over long horizons. We seek a principled middle ground that combines the reliability of explicit simulators with the flexibility of learned models, allowing world models to be adapted during online execution. By targeting a broad class of environments whose dynamics are governed by the ordering, timing, and causality of discrete events, such as queueing and service operations, embodied task planning, and message-mediated multi-agent coordination, we advocate explicit, executable discrete-event world models synthesized directly from natural-language specifications. Our approach adopts the DEVS formalism and introduces a staged LLM-based generation pipeline that separates structural inference of component interactions from component-level event and timing logic. To evaluate generated models without a unique ground truth, simulators emit structured event traces that are validated against specification-derived temporal and semantic constraints, enabling reproducible verification and localized diagnostics. Together, these contributions produce world models that are consistent over long-horizon rollouts, verifiable from observable behavior, and efficient to synthesize on demand during online execution.
Authors: Xingyao Wang, Valerie Chen, Heng Ji, Graham Neubig
Abstract: Academic benchmarks for coding agents tend to reward autonomous task completion, measured by verifiable rewards such as unit-test success. In contrast, real-world coding agents operate with humans in the loop, where success signals are typically noisy, delayed, and sparse. How can we bridge this gap? In this paper, we propose a process to learn a "critic" model from sparse and noisy interaction data, which can then be used both as a reward model for either RL-based training or inference-time scaling. Specifically, we introduce Critic Rubrics, a rubric-based supervision framework with 24 behavioral features that can be derived from human-agent interaction traces alone. Using a semi-supervised objective, we can then jointly predict these rubrics and sparse human feedback (when present). In experiments, we demonstrate that, despite being trained primarily from trace-observable rubrics and sparse real-world outcome proxies, these critics improve best-of-N reranking on SWE-bench (Best@8 +15.9 over Random@8 over the rerankable subset of trajectories), enable early stopping (+17.7 with 83% fewer attempts), and support training-time data curation via critic-selected trajectories.
Authors: Maheep Chaudhary
Abstract: Humans often become more self-aware under threat, yet can lose self-awareness when absorbed in a task; we hypothesize that language models exhibit environment-dependent \textit{evaluation awareness}. This raises concerns that models could strategically underperform, or \textit{sandbag}, to avoid triggering capability-limiting interventions such as unlearning or shutdown. Prior work demonstrates sandbagging under hand-crafted prompts, but this underestimates the true vulnerability ceiling. We introduce a black-box adversarial optimization framework treating the in-context prompt as an optimizable environment, and develop two approaches to characterize sandbagging: (1) measuring whether models expressing intent to underperform can actually execute it across different task structures, and (2) causally isolating whether underperformance is driven by genuine evaluation-aware reasoning or shallow prompt-following. Evaluating Claude-3.5-Haiku, GPT-4o-mini, and Llama-3.3-70B across four benchmarks (Arithmetic, GSM8K, MMLU, and HumanEval), optimized prompts induce up to 94 percentage point (pp) degradation on arithmetic (GPT-4o-mini: 97.8\%$\rightarrow$4.0\%), far exceeding hand-crafted baselines which produce near-zero behavioral change. Code generation exhibits model-dependent resistance: Claude degrades only 0.6pp, while Llama's accuracy drops to 0\%. The intent -- execution gap reveals a monotonic resistance ordering: Arithmetic $<$ GSM8K $<$ MMLU, demonstrating that vulnerability is governed by task structure rather than prompt strength. CoT causal intervention confirms that 99.3\% of sandbagging is causally driven by verbalized eval-aware reasoning, ruling out shallow instruction-following. These findings demonstrate that adversarially optimized prompts pose a substantially greater threat to evaluation reliability than previously understood.
Authors: Chiara Bonfanti, Davide Colaiacomo, Luca Cagliero, Cataldo Basile
Abstract: Web security demands rapid response capabilities to evolving cyber threats. Agentic Artificial Intelligence (AI) promises automation, but the need for trustworthy security responses is of the utmost importance. This work investigates the role of semantic relations in extracting information for sensitive operational tasks, such as configuring security controls for mitigating threats. To this end, it proposes to leverage hypernym-hyponym textual relations to extract relevant information from Cyber Threat Intelligence (CTI) reports. By leveraging a neuro-symbolic approach, the multi-agent system automatically generates CLIPS code for an expert system creating firewall rules to block malicious network traffic. Experimental results show the superior performance of the hypernym-hyponym retrieval strategy compared to various baselines and the higher effectiveness of the agentic approach in mitigating threats.
Authors: Sule Ozturk Birim, Fabrizio Marozzo, Yigit Kazancoglu
Abstract: Generative artificial intelligence is increasingly being integrated into complex business workflows, fundamentally shifting the boundaries of managerial decision-making. However, the reliability of its strategic advice in ambiguous business contexts remains a critical knowledge gap. This study addresses this by comparing various models on ambiguity detection, evaluating how a systematic resolution process enhances response quality, and investigating their sycophantic behavior when presented with flawed directives. Using a novel four-dimensional business ambiguity taxonomy, we conducted a human-in-the-loop experiment across strategic, tactical, and operational scenarios. The resulting decisions were assessed with an "LLM-as-a-judge" framework on criteria including agreement, actionability, justification quality, and constraint adherence. Results reveal distinct performance capabilities. While models excel in detecting internal contradictions and contextual ambiguities, they struggle with structural linguistic nuances. Ambiguity resolution consistently increased response quality across all decision types, while sycophantic behavior analysis revealed distinct patterns depending on the model architecture. This study contributes to the bounded rationality literature by positioning GAI as a cognitive scaffold that can detect and resolve ambiguities managers might overlook, but whose own artificial limitations necessitate human management to ensure its reliability as a strategic partner.
Authors: Jyoti Aneja, Michael Harrison, Neel Joshi, Tyler LaBonte, John Langford, Eduardo Salinas
Abstract: We present Phi-4-reasoning-vision-15B, a compact open-weight multimodal reasoning model, and share the motivations, design choices, experiments, and learnings that informed its development. Our goal is to contribute practical insight to the research community on building smaller, efficient multimodal reasoning models and to share the result of these learnings as an open-weight model that is good at common vision and language tasks and excels at scientific and mathematical reasoning and understanding user interfaces. Our contributions include demonstrating that careful architecture choices and rigorous data curation enable smaller, open-weight multimodal models to achieve competitive performance with significantly less training and inference-time compute and tokens. The most substantial improvements come from systematic filtering, error correction, and synthetic augmentation -- reinforcing that data quality remains the primary lever for model performance. Systematic ablations show that high-resolution, dynamic-resolution encoders yield consistent improvements, as accurate perception is a prerequisite for high-quality reasoning. Finally, a hybrid mix of reasoning and non-reasoning data with explicit mode tokens allows a single model to deliver fast direct answers for simpler tasks and chain-of-thought reasoning for complex problems.
Authors: Tarjei Paule Hage, Markus J. Buehler
Abstract: Can reinforcement learning with hard, verifiable rewards teach a compact language model to reason about physics, or does it primarily learn to pattern-match toward correct answers? We study this question by training a 1.5B-parameter reasoning model on beam statics, a classic engineering problem, using parameter-efficient RLVR with binary correctness rewards from symbolic solvers, without teacher-generated reasoning traces. The best BeamPERL checkpoint achieves a 66.7% improvement in Pass@1 over the base model. However, the learned competence is anisotropic: the model generalizes compositionally (more loads) but fails under topological shifts (moved supports) that require the same equilibrium equations. Intermediate checkpoints yield the strongest reasoning, while continued optimization degrades robustness while maintaining reward. These findings reveal a key limitation of outcome-level alignment: reinforcement learning with exact physics rewards induces procedural solution templates rather than internalization of governing equations. The precision of the reward signal - even when analytically exact - does not by itself guarantee transferable physical reasoning. Our results suggest that verifiable rewards may need to be paired with structured reasoning scaffolding to move beyond template matching toward robust scientific reasoning.
Authors: Qianyun Guo, Yibo Li, Yue Liu, Bryan Hooi
Abstract: Large Language Models (LLMs) are increasingly serving as personal assistants, where users share complex and diverse preferences over extended interactions. However, assessing how well LLMs can follow these preferences in realistic, long-term situations remains underexplored. This work proposes RealPref, a benchmark for evaluating realistic preference-following in personalized user-LLM interactions. RealPref features 100 user profiles, 1300 personalized preferences, four types of preference expression (ranging from explicit to implicit), and long-horizon interaction histories. It includes three types of test questions (multiple-choice, true-or-false, and open-ended), with detailed rubrics for LLM-as-a-judge evaluation. Results indicate that LLM performance significantly drops as context length grows and preference expression becomes more implicit, and that generalizing user preference understanding to unseen scenarios poses further challenges. RealPref and these findings provide a foundation for future research to develop user-aware LLM assistants that better adapt to individual needs. The code is available at https://github.com/GG14127/RealPref.
Authors: Alfio Massimiliano Gliozzo, Junkyu Lee, Nahuel Defosse
Abstract: Agentic AI is rapidly transitioning from research prototypes to enterprise deployments, where requirements extend to meet the software quality attributes of reliability, scalability, and observability beyond plausible text generation. We present Agentics 2.0, a lightweight, Python-native framework for building high-quality, structured, explainable, and type-safe agentic data workflows. At the core of Agentics 2.0, the logical transduction algebra formalizes a large language model inference call as a typed semantic transformation, which we call a transducible function that enforces schema validity and the locality of evidence. The transducible functions compose into larger programs via algebraically grounded operators and execute as stateless asynchronous calls in parallel in asynchronous Map-Reduce programs. The proposed framework provides semantic reliability through strong typing, semantic observability through evidence tracing between slots of the input and output types, and scalability through stateless parallel execution. We instantiate reusable design patterns and evaluate the programs in Agentics 2.0 on challenging benchmarks, including DiscoveryBench for data-driven discovery and Archer for NL-to-SQL semantic parsing, demonstrating state-of-the-art performance.
Authors: Quan Shi, Alexandra Zytek, Pedram Razavi, Karthik Narasimhan, Victor Barres
Abstract: Conversational agents are increasingly deployed in knowledge-intensive settings, where correct behavior depends on retrieving and applying domain-specific knowledge from large, proprietary, and unstructured corpora during live interactions with users. Yet most existing benchmarks evaluate retrieval or tool use independently of each other, creating a gap in realistic, fully agentic evaluation over unstructured data in long-horizon interactions. We introduce $\tau$-Knowledge, an extension of $\tau$-Bench for evaluating agents in environments where success depends on coordinating external, natural-language knowledge with tool outputs to produce verifiable, policy-compliant state changes. Our new domain, $\tau$-Banking, models realistic fintech customer support workflows in which agents must navigate roughly 700 interconnected knowledge documents while executing tool-mediated account updates. Across embedding-based retrieval and terminal-based search, even frontier models with high reasoning budgets achieve only $\sim$25.5% pass^1, with reliability degrading sharply over repeated trials. Agents struggle to retrieve the correct documents from densely interlinked knowledge bases and to reason accurately over complex internal policies. Overall, $\tau$-Knowledge provides a realistic testbed for developing agents that integrate unstructured knowledge in human-facing deployments.
Authors: Boyuan (Keven), Guan, Wencong Cui, Levente Juhasz
Abstract: WebGIS development requires rigor, yet agentic AI frequently fails due to five large language model (LLM) limitations: context constraints, cross-session forgetting, stochasticity, instruction failure, and adaptation rigidity. We propose a dual-helix governance framework reframing these challenges as structural governance problems that model capacity alone cannot resolve. We implement the framework as a 3-track architecture (Knowledge, Behavior, Skills) that uses a knowledge graph substrate to stabilize execution by externalizing domain facts and enforcing executable protocols, complemented by a self-learning cycle for autonomous knowledge growth. Applying this to the FutureShorelines WebGIS tool, a governed agent refactored a 2,265-line monolithic codebase into modular ES6 components. Results demonstrated a 51\% reduction in cyclomatic complexity and a 7-point increase in maintainability index. A comparative experiment against a zero-shot LLM confirms that externalized governance, not just model capability, drives operational reliability in geospatial engineering. This approach is implemented in the open-source AgentLoom governance toolkit.
Authors: Wenhui Zhu, Xiwen Chen, Zhipeng Wang, Jingjing Wang, Xuanzhao Dong, Minzhou Huang, Rui Cai, Hejian Sang, Hao Wang, Peijie Qiu, Yueyue Deng, Prayag Tiwari, Brendan Hogan Rappazzo, Yalin Wang
Abstract: Long-horizon LLM agents require memory systems that remain accurate under fixed context budgets. However, existing systems struggle with two persistent challenges in long-term dialogue: (i) \textbf{disconnected evidence}, where multi-hop answers require linking facts distributed across time, and (ii) \textbf{state updates}, where evolving information (e.g., schedule changes) creates conflicts with older static logs. We propose AriadneMem, a structured memory system that addresses these failure modes via a decoupled two-phase pipeline. In the \textbf{offline construction phase}, AriadneMem employs \emph{entropy-aware gating} to filter noise and low-information message before LLM extraction and applies \emph{conflict-aware coarsening} to merge static duplicates while preserving state transitions as temporal edges. In the \textbf{online reasoning phase}, rather than relying on expensive iterative planning, AriadneMem executes \emph{algorithmic bridge discovery} to reconstruct missing logical paths between retrieved facts, followed by \emph{single-call topology-aware synthesis}. On LoCoMo experiments with GPT-4o, AriadneMem improves \textbf{Multi-Hop F1 by 15.2\%} and \textbf{Average F1 by 9.0\%} over strong baselines. Crucially, by offloading reasoning to the graph layer, AriadneMem reduces \textbf{total runtime by 77.8\%} using only \textbf{497} context tokens. The code is available at https://github.com/LLM-VLM-GSL/AriadneMem.
Authors: Daniel Fein, Max Lamparth, Violet Xiang, Mykel J. Kochenderfer, Nick Haber
Abstract: Reward Models (RMs) are crucial for online alignment of language models (LMs) with human preferences. However, RM-based preference-tuning is vulnerable to reward hacking, whereby LM policies learn undesirable behaviors from flawed RMs. By systematically measuring biases in five high-quality RMs, including the state-of-the-art, we find that issues persist despite prior work with respect to length, sycophancy, and overconfidence. We also discover new issues related to bias toward model-specific styles and answer-order. We categorize RM failures by complexity and propose a simple post-hoc intervention to mitigate low-complexity biases that arise from spurious correlations. Our proposed mechanistic reward shaping reduces targeted biases without degrading reward quality and while using minimal labeled data. The method is extensible to new biases, model-internal, and generalizes out-of-distribution.
Authors: Wenhao Wu, Zhentao Tang, Yafu Li, Shixiong Kai, Mingxuan Yuan, Zhenhong Sun, Chunlin Chen, Zhi Wang
Abstract: Large Language Models (LLMs) exhibit high reasoning capacity in medical question-answering, but their tendency to produce hallucinations and outdated knowledge poses critical risks in healthcare fields. While Retrieval-Augmented Generation (RAG) mitigates these issues, existing methods rely on noisy token-level signals and lack the multi-round refinement required for complex reasoning. In the paper, we propose **MA-RAG** (**M**ulti-Round **A**gentic RAG), a framework that facilitates test-time scaling for complex medical reasoning by iteratively evolving both external evidence and internal reasoning history within an agentic refinement loop. At each round, the agent transforms semantic **conflict** among candidate responses into actionable queries to retrieve external evidence, while optimizing history reasoning traces to mitigate long-context degradation. MA-RAG extends the *self-consistency* principle by leveraging the lack of consistency as a proactive signal for multi-round agentic reasoning and retrieval, and mirrors a *boosting* mechanism that iteratively minimizes the residual error toward a stable, high-fidelity medical **consensus**. Extensive evaluations across 7 medical Q&A benchmarks show that MA-RAG consistently surpasses competitive inference-time scaling and RAG baselines, delivering **substantial +6.8 points** on average accuracy over the backbone model. Our code is available at [this url](https://github.com/NJU-RL/MA-RAG).
Authors: Sanyam Singh, Naga Ganesh, Vineet Singh, Lakshmi Pedapudi, Ritesh Kumar, SSP Jyothi, Archana Karanam, C. Yashoda, Mettu Vijaya Rekha Reddy, Shesha Phani Debbesa, Chandan Dash
Abstract: Large Language Models show promise for agricultural advisory, yet vanilla models exhibit unsupported recommendations, generic advice lacking specific, actionable detail, and communication styles misaligned with smallholder farmer needs. In high stakes agricultural contexts, where recommendation accuracy has direct consequences for farmer outcomes, these limitations pose challenges for responsible deployment. We present a hybrid LLM architecture that decouples factual retrieval from conversational delivery: supervised fine-tuning with LoRA on expert-curated GOLDEN FACTS (atomic, verified units of agricultural knowledge) optimizes fact recall, while a separate stitching layer transforms retrieved facts into culturally appropriate, safety-aware responses. Our evaluation framework, DG-EVAL, performs atomic fact verification (measuring recall, precision, and contradiction detection) against expert-curated ground truth rather than Wikipedia or retrieved documents. Experiments across multiple model configurations on crops and queries from Bihar, India show that fine-tuning on curated data substantially improves fact recall and F1, while maintaining high relevance. Using a fine-tuned smaller model achieves comparable or better factual quality at a fraction of the cost of frontier models. A stitching layer further improves safety subscores while maintaining high conversational quality. We release the farmerchat-prompts library to enable reproducible development of domain-specific agricultural AI.
Authors: Gaia Molinaro, Dave August, Danielle Perszyk, Anne G. E. Collins
Abstract: As large language models (LLMs) get integrated into human decision-making, they are increasingly choosing goals autonomously rather than only completing human-defined ones, assuming they will reflect human preferences. However, human-LLM similarity in goal selection remains largely untested. We directly assess the validity of LLMs as proxies for human goal selection in a controlled, open-ended learning task borrowed from cognitive science. Across four state-of-the-art models (GPT-5, Gemini 2.5 Pro, Claude Sonnet 4.5, and Centaur), we find substantial divergence from human behavior. While people gradually explore and learn to achieve goals with diversity across individuals, most models exploit a single identified solution (reward hacking) or show surprisingly low performance, with distinct patterns across models and little variability across instances of the same model. Even Centaur, explicitly trained to emulate humans in experimental settings, poorly captures people's goal selection. Chain-of-thought reasoning and persona steering provide limited improvements. These findings highlight the uniqueness of human goal selection, cautioning against replacing it with current models in applications such as personal assistance, scientific discovery, and policy research.
Authors: Ke Yang, Zixi Chen, Xuan He, Jize Jiang, Michel Galley, Chenglong Wang, Jianfeng Gao, Jiawei Han, ChengXiang Zhai
Abstract: Long-term memory is essential for large language model (LLM) agents operating in complex environments, yet existing memory designs are either task-specific and non-transferable, or task-agnostic but less effective due to low task-relevance and context explosion from raw memory retrieval. We propose PlugMem, a task-agnostic plugin memory module that can be attached to arbitrary LLM agents without task-specific redesign. Motivated by the fact that decision-relevant information is concentrated as abstract knowledge rather than raw experience, we draw on cognitive science to structure episodic memories into a compact, extensible knowledge-centric memory graph that explicitly represents propositional and prescriptive knowledge. This representation enables efficient memory retrieval and reasoning over task-relevant knowledge, rather than verbose raw trajectories, and departs from other graph-based methods like GraphRAG by treating knowledge as the unit of memory access and organization instead of entities or text chunks. We evaluate PlugMem unchanged across three heterogeneous benchmarks (long-horizon conversational question answering, multi-hop knowledge retrieval, and web agent tasks). The results show that PlugMem consistently outperforms task-agnostic baselines and exceeds task-specific memory designs, while also achieving the highest information density under a unified information-theoretic analysis. Code and data are available at https://github.com/TIMAN-group/PlugMem.
Authors: Haoyang He, Zihua Rong, Liangjie Zhao, Yunjia Zhao, Lan Yang, Honggang Zhang
Abstract: Test-time Training enables model adaptation using only test questions and offers a promising paradigm for improving the reasoning ability of large language models (LLMs). However, it faces two major challenges: test questions are often highly difficult, making self-generated pseudo-labels unreliable, and existing methods lack effective mechanisms to adapt to a model's specific reasoning weaknesses, leading to inefficient learning. To address these issues, we propose \textbf{TTSR}, a self-reflective test-time self-evolving training framework. TTSR employs a single pretrained language model that alternates between the roles of a \textit{Student} and a \textit{Teacher} at test time. The Student focuses on solving problems and learning from synthesized variant questions, while the Teacher analyzes the Student's failed reasoning trajectories, summarizes recurring reasoning weaknesses, and synthesizes targeted variant questions accordingly. This process guides the model to improve within a learnable regime through a continual self-evolving loop. Experimental results on multiple challenging mathematical reasoning benchmarks show that TTSR consistently improves reasoning performance and generalizes well across different model backbones and general-domain reasoning tasks. These findings suggest that teacher-mediated self-reflection provides an effective pathway for stable and continual reasoning improvement at test time.
Authors: Bartosz Dziuba, Kacper Kuchta, Pawe{\l} Batorski, Przemys{\l}aw Spurek, Paul Swoboda
Abstract: Large Language Models (LLMs) have improved substantially alignment, yet their behavior remains highly sensitive to prompt phrasing. This brittleness has motivated automated prompt engineering, but most existing methods (i) require a task-specific training set, (ii) rely on expensive iterative optimization to produce a single dataset-level prompt, and (iii) must be rerun from scratch for each new task. We introduce TATRA, a dataset-free prompting method that constructs instance-specific few-shot prompts by synthesizing on-the-fly examples to accompany a user-provided instruction. TATRA requires no labeled training data and avoids task-specific optimization loops, while retaining the benefits of demonstration-based prompting. Across standard text classification benchmarks, TATRA matches or improves over strong prompt-optimization baselines that depend on training data and extensive search. On mathematical reasoning benchmarks, TATRA achieves state-of-the-art performance on GSM8K and DeepMath, outperforming methods that explicitly optimize prompts on those tasks. Our results suggest that per-instance construction of effective in-context examples is more important than running long, expensive optimization loops to produce a single prompt per task. We will make all code publicly available upon acceptance of the paper. Code is available at https://github.com/BMD223/TATRA
Authors: Dvir David Biton, Roy Friedman
Abstract: The rapid adoption of large language models (LLMs) has created demand for faster responses and lower costs. Semantic caching, reusing semantically similar requests via their embeddings, addresses this need but breaks classic cache assumptions and raises new challenges. In this paper, we explore offline policies for semantic caching, proving that implementing an optimal offline policy is NP-hard, and propose several polynomial-time heuristics. We also present online semantic aware cache policies that combine recency, frequency, and locality. Evaluations on diverse datasets show that while frequency based policies are strong baselines, our novel variant improves semantic accuracy. Our findings reveal effective strategies for current systems and highlight substantial headroom for future innovation. All code is open source.
Authors: Divija Amaram, Lu Gao, Gowtham Reddy Gudla, Tejaswini Sanjay Katale
Abstract: Effective knowledge management is critical for preserving institutional expertise and improving the efficiency of workforce training in state transportation agencies. Traditional approaches, such as static documentation, classroom-based instruction, and informal mentorship, often lead to fragmented knowledge transfer, inefficiencies, and the gradual loss of expertise as senior engineers retire. Moreover, given the enormous volume of technical manuals, guidelines, and research reports maintained by these agencies, it is increasingly challenging for engineers to locate relevant information quickly and accurately when solving field problems or preparing for training tasks. These limitations hinder timely decision-making and create steep learning curves for new personnel in maintenance and construction operations. To address these challenges, this paper proposes a Retrieval-Augmented Generation (RAG) framework with a multi-agent architecture to support knowledge management and decision making. The system integrates structured document retrieval with real-time, context-aware response generation powered by a large language model (LLM). Unlike conventional single-pass RAG systems, the proposed framework employs multiple specialized agents for retrieval, answer generation, evaluation, and query refinement, which enables iterative improvement and quality control. In addition, the system incorporates an open-weight vision-language model to convert technical figures into semantic textual representations, which allows figure-based knowledge to be indexed and retrieved alongside text. Retrieved text and figure-based context are then provided to an open-weight large language model, which generates the final responses grounded in the retrieved evidence.
Authors: Shirley Wu, Evelyn Choi, Arpandeep Khatua, Zhanghan Wang, Joy He-Yueya, Tharindu Cyril Weerasooriya, Wei Wei, Diyi Yang, Jure Leskovec, James Zou
Abstract: Large Language Models (LLMs) are increasingly used to simulate how specific users respond to a given context, enabling more user-centric applications that rely on user feedback. However, existing user simulators mostly imitate surface-level patterns and language styles, which fail to reflect the underlying states of real users (e.g., beliefs and emotions). To address these limitations, we propose a novel training framework, HumanLM, which builds user simulators that accurately reflect real users. Our key insight is that, in addition to generating responses, the model should generate natural-language latent states that align with ground-truth responses through reinforcement learning. These latent states correspond to a set of psychologically grounded state dimensions that drive how real users respond. HumanLM further synthesizes these aligned latent states into responses that accurately represent real users. For extensive evaluation, we develop Humanual, a comprehensive benchmark for simulating real users based on public data. Humanual consists of six large-scale datasets with 26k users and 216k responses in total, spanning diverse tasks such as generating user responses to daily life issues, political blogs, and chat sessions with LLM assistants. Across datasets, HumanLM significantly outperforms alternative approaches, achieving an average relative improvement of 16.3% in alignment scores from an LLM judge. In a real-time simulation study with 111 participants, HumanLM achieves the highest similarity to real user responses and competitive human-likeness scores.
Authors: Mahesh Godavarti
Abstract: We present a concise architecture for joint training on sentences and structured data while keeping knowledge and language representations separable. The model treats knowledge graphs and hypergraphs as structured instances with role slots and encodes them into a key-value repository that a language transformer can attend over. Attention is conditioned by journey-based role transport, which unifies edge-labeled KG traversal, hyperedge traversal, and sentence structure. We outline a dual-stream architecture, hierarchical layer groups with instance-local, neighborhood, and global mixing attention, retrieval over a separate repository, and multi-task objectives spanning masked language modeling, link prediction, and role-consistency denoising. The result is an explicit, inspectable separation between linguistic context and structured knowledge, while still enabling tight alignment through cross-attention.
Authors: Avinash Reddy, Thayne T. Walker, James S. Ide, Amrit Singh Bedi
Abstract: Large language models (LLMs) are increasingly used to generate executable outputs, JSON objects, and API calls, where a single syntax error can make the output unusable. Constrained decoding enforces validity token-by-token via masking and renormalization, but it can distort generation when the model assigns low probability mass to valid continuations, pushing decoding toward locally valid yet semantically incorrect trajectories. We propose \emph{Draft-Conditioned Constrained Decoding (DCCD)}, a simple two-step, training-free inference procedure that decouples semantic planning from structural enforcement: an unconstrained draft is generated first, and constrained decoding is then applied, conditioned on this draft, to guarantee validity. We analyze DCCD through a KL-projection view, showing that draft conditioning increases feasible mass and reduces the cumulative "projection tax" induced by hard constraints, with an optional best-of-$K$ draft selection. Across structured reasoning benchmarks, DCCD improves strict structured accuracy by up to +24 percentage points over standard constrained decoding (e.g., 15.2\% to 39.0\% on GSM8K with a 1B model), and enables smaller model pairs to match or exceed much larger constrained baselines, yielding substantial gains in parameter efficiency.
Authors: Ivan Matveev
Abstract: Recently presented Token-Oriented Object Notation (TOON) aims to replace JSON as a serialization format for passing structured data to LLMs with significantly reduced token usage. While showing solid accuracy in LLM comprehension, there is a lack of tests against JSON generation. Though never present in training data, TOON syntax is simple enough to suggest one-shot in-context learning could support accurate generation. The inevitable prompt overhead can be an acceptable trade-off for shorter completions. To test this, we conducted a benchmark creating several test cases with regard to structural complexity, a validation pipeline, and comparing plain JSON generation vs structured output (via constrained decoding) JSON generation vs TOON one-shot in-context learning generation. JSON structured output was included to establish a minimum token budget baseline and to set a starting point for future experiments testing TOON constrained decoding inference enforcement. Key findings: TOON shows promising accuracy/token consumption ratio for in-domain generation tasks, though this advantage is often reduced by the "prompt tax" of instructional overhead in shorter contexts. Plain JSON generation shows the best one-shot and final accuracy, even compared with constrained decoding structured output, where the only significant advantage is the lowest token usage as a trade-off for slightly decreased accuracy overall and significant degradation for some models. Notably, for simple structures, this "lowest token usage" of constrained decoding outperformed even TOON, hinting that TOON enforcing via frameworks such as xgrammar may not yield the desired results. Furthermore, the results suggest a scaling hypothesis: TOON's true efficiency potential likely follows a non-linear curve, shining only beyond a specific point where cumulative syntax savings amortize the initial prompt overhead.
Authors: Owen H. T. Lu, Tiffany T. Y. Hsu
Abstract: Epistemic Network Analysis (ENA) is a method for investigating the relational structure of concepts in text by representing co-occurring concepts as networks. Traditional ENA, however, relies heavily on manual expert coding, which limits its scalability and real-world applicability to large text corpora. Topic modeling provides an automated approach to extracting concept-level representations from text and can serve as an alternative to manual coding. To tackle this limitation, the present study merges BERTopic with ENA and introduces TopicENA, a topic-based epistemic network analysis framework. TopicENA substitutes manual concept coding with automatically generated topics while maintaining ENA's capacity for modeling structural associations among concepts. To explain the impact of modeling choices on TopicENA outcomes, three analysis cases are presented. The first case assesses the effect of topic granularity, indicating that coarse-grained topics are preferable for large datasets, whereas fine-grained topics are more effective for smaller datasets. The second case examines topic inclusion thresholds and finds that threshold values should be adjusted according to topic quality indicators to balance network consistency and interpretability. The third case tests TopicENA's scalability by applying it to a substantially larger dataset than those used in previous ENA studies. Collectively, these cases illustrate that TopicENA facilitates practical and interpretable ENA analysis at scale and offers concrete guidance for configuring topic-based ENA pipelines in large-scale text analysis.
Authors: Adi Simhi, Fazl Barez, Martin Tutek, Yonatan Belinkov, Shay B. Cohen
Abstract: How does the conversational past of large language models (LLMs) influence their future performance? Recent work suggests that LLMs are affected by their conversational history in unexpected ways. For instance, hallucinations in prior interactions may influence subsequent model responses. In this work, we introduce History-Echoes, a framework that investigates how conversational history biases subsequent generations. The framework explores this bias from two perspectives: probabilistically, we model conversations as Markov chains to quantify state consistency; geometrically, we measure the consistency of consecutive hidden representations. Across three model families and six datasets spanning diverse phenomena, our analysis reveals a strong correlation between the two perspectives. By bridging these perspectives, we demonstrate that behavioral persistence manifests as a geometric trap, where gaps in the latent space confine the model's trajectory. Code available at https://github.com/technion-cs-nlp/OldHabitsDieHard.
Authors: Yuchen Wang, Haonan Wang, Yu Guo, Honglong Yang, Xiaomeng Li
Abstract: Decoding natural language from non-invasive EEG signals is a promising yet challenging task. However, current state-of-the-art models remain constrained by three fundamental limitations: Semantic Bias (mode collapse into generic templates), Signal Neglect (hallucination based on linguistic priors rather than neural inputs), and the BLEU Trap, where evaluation metrics are artificially inflated by high-frequency stopwords, masking a lack of true semantic fidelity. To address these challenges, we propose SemKey, a novel multi-stage framework that enforces signal-grounded generation through four decoupled semantic objectives: sentiment, topic, length, and surprisal. We redesign the interaction between the neural encoder and the Large Language Model (LLM) by injecting semantic prompts as Queries and EEG embeddings as Key-Value pairs, strictly forcing the model to attend to neural inputs. Furthermore, we move beyond standard translation metrics by adopting N-way Retrieval Accuracy and Fr\'echet Distance to rigorously assess diversity and alignment. Extensive experiments demonstrate that our approach effectively eliminates hallucinations on noise inputs and achieves SOTA performance on these robust protocols. Code will be released upon acceptance at https://github.com/xmed-lab/SemKey.
Authors: Minghua Wu, Javier Conde, Pedro Reviriego, Marc Brysbaert
Abstract: Large Language Models (LLMs) exhibit a significant "embodiment gap", where their text-based representations fail to align with human sensorimotor experiences. This study systematically investigates whether and how task-specific fine-tuning can bridge this gap. Utilizing Representational Similarity Analysis (RSA) and dimension-specific correlation metrics, we demonstrate that the internal representations of LLMs can be steered toward more embodied, grounded patterns through fine-tuning. Furthermore, the results show that while sensorimotor improvements generalize robustly across languages and related sensory-motor dimensions, they are highly sensitive to the learning objective, failing to transfer across two disparate task formats.
Authors: Xin Yang, Letian Li, Abudukelimu Wuerkaixi, Xuxin Cheng, Cao Liu, Ke Zeng, Xunliang Cai, Wenyuan Jiang
Abstract: Large language models (LLMs) have demonstrated remarkable and steadily improving performance across a wide range of tasks. However, LLM performance may be highly sensitive to prompt variations especially in scenarios with limited openness or strict output formatting requirements, indicating insufficient robustness. In real-world applications, user prompts provided to LLMs often contain imperfections, which may undermine the quality of the model's responses. To address this issue, previous work has primarily focused on preprocessing prompts, employing external tools or even LLMs to refine prompt formulations in advance. However, these approaches overlook the intrinsic robustness of LLMs, and their reliance on external components introduces additional computational overhead and uncertainty. In this work, we propose a Contrastive Learning-based Inverse Direct Preference Optimization (CoIPO) method that minimizes the discrepancy between the label-aligned logits produced by the model under a clean prompt and its noisy counterpart, and conduct a detailed analysis using mutual information theory. We augment the FLAN dataset by constructing paired prompts, each consisting of a clean prompt and its corresponding noisy version for training. Additionally, to evaluate the effectiveness, we develop NoisyPromptBench, a benchmark enhanced and derived from the existing PromptBench. Experimental results conducted on NoisyPromptBench demonstrate that our proposed method achieves a significant improvement in average accuracy over the current state-of-the-art approaches. The source code of CoIPO, pair-wise FLAN datasets, and NoisyPromptBench have already been released on https://github.com/vegetable-yx/CoIPO.
Authors: Stefano De Giorgis, Ting-Chih Chen, Filip Ilievski
Abstract: Internet memes are a powerful form of online communication, yet their nature and reliance on commonsense knowledge make toxicity detection challenging. Identifying key features for meme interpretation and understanding, is a crucial task. Previous work has been focused on some elements contributing to the meaning, such as the Textual dimension via OCR, the Visual dimension via object recognition, upper layers of meaning like the Emotional dimension, Toxicity detection via proxy variables, such as hate speech detection, and sentiment analysis. Nevertheless, there is still a lack of an overall architecture able to formally identify elements contributing to the meaning of a meme, and be used in the sense-making process. In this work, we present a semantic framework and a corresponding benchmark for automatic knowledge extraction from memes. First, we identify the necessary dimensions to understand and interpret a meme: Textual material, Visual material, Scene, Background Knowledge, Emotion, Semiotic Projection, Analogical Mapping, Overall Intent, Target Community, and Toxicity Assessment. Second, the framework guides a semi-automatic process of generating a benchmark with commonsense question-answer pairs about meme toxicity assessment and its underlying reason. The resulting benchmark M-QUEST consists of 609 question-answer pairs for 307 memes. Thirdly, we evaluate eight open-source large language models on their ability to correctly solve M-QUEST. Our results show that current models' commonsense reasoning capabilities for toxic meme interpretation vary depending on the dimension and architecture. Models with instruction tuning and reasoning capabilities significantly outperform the others, though pragmatic inference questions remain challenging. We release code, benchmark, and prompts to support future research intersecting multimodal content safety and commonsense reasoning.
Authors: Keren Artiaga, Conor Lynch, Haithem Afli, Mohammed Hasanuzzaman
Abstract: Most sign language recognition research relies on Transfer Learning (TL) from vision-based datasets such as ImageNet. Some extend this to alternatively available language datasets, often focusing on signs with cross-linguistic similarities. This body of work examines the necessity of these likenesses on effective knowledge transfer by comparing TL performance between iconic signs of two different sign language pairs: Chinese to Arabic and Greek to Flemish. Google Mediapipe was utilised as an input feature extractor, enabling spatial information of these signs to be processed with a Multilayer Perceptron architecture and the temporal information with a Gated Recurrent Unit. Experimental results showed a 7.02% improvement for Arabic and 1.07% for Flemish when conducting iconic TL from Chinese and Greek respectively.
Authors: Nikita Kuznetsov, Niyaz Ismagilov, Ernesto Campos
Abstract: Recent advances in Natural Language Processing have been predominantly driven by transformer-based architectures, which rely heavily on self-attention mechanisms to model relationships between tokens in a sequence. Similarly, the field of Quantum Natural Language Processing, which seeks to leverage quantum principles to address challenges in language understanding and generation tasks, has seen the recent development of quantum self-attention mechanisms. We propose a classical quantum-inspired self-attention (QISA) mechanism and integrate it into the full autoregressive language modeling pipeline of GPT-1. To the best of our knowledge, this is the first integration of this kind, as previous quantum self-attention mechanisms have been primarily tested on text classification. In our experiments, QISA achieves better performance when compared to standard self-attention on the metrics character error rate ($15.5\times$ better), word error rate ($4.7 \times $) and cross-entropy loss ($13 \times$). This is achieved while only requiring a $ 2.6\times$ longer inference time.
Authors: James Wedgwood, Chhavi Yadav, Virginia Smith
Abstract: Large Language Models (LLMs) are increasingly used as scalable evaluators of model outputs, but their preference judgments exhibit systematic biases and can diverge from human evaluations. Prior work on LLM-as-a-judge has largely focused on a small, predefined set of hypothesized biases, leaving open the problem of automatically discovering unknown drivers of LLM preferences. We address this gap by studying several embedding-level concept extraction methods for analyzing LLM judge behavior. We compare these methods in terms of interpretability and predictiveness, finding that sparse autoencoder-based approaches recover substantially more interpretable preference features than alternatives while remaining competitive in predicting LLM decisions. Using over 27k paired responses from multiple human preference datasets and judgments from three LLMs, we analyze LLM judgments and compare them to those of human annotators. Our method both validates existing results, such as the tendency for LLMs to prefer refusal of sensitive requests at higher rates than humans, and uncovers new trends across both general and domain-specific datasets, including biases toward responses that emphasize concreteness and empathy in approaching new situations, toward detail and formality in academic advice, and against legal guidance that promotes active steps like calling police and filing lawsuits. Our results show that automated concept discovery enables systematic analysis of LLM judge preferences without predefined bias taxonomies.
Authors: Jaikrishna Manojkumar Patil, Divyagna Bavikadi, Kaustuv Mukherji, Ashby Steward-Nolan, Peggy-Jean Allin, Tumininu Awonuga, Joshua Garland, Paulo Shakarian
Abstract: Effective communication often relies on aligning a message with an audience's narrative and worldview. Narrative shift involves transforming text to reflect a different narrative framework while preserving its original core message--a task we demonstrate is significantly challenging for current Large Language Models (LLMs). To address this, we propose a neurosymbolic approach grounded in social science theory and abductive reasoning. Our method automatically extracts rules to abduce the specific story elements needed to guide an LLM through a consistent and targeted narrative transformation. Across multiple LLMs, abduction-guided transformed stories shifted the narrative while maintaining the fidelity with the original story. For example, with GPT-4o we outperform the zero-shot LLM baseline by 55.88% for collectivistic to individualistic narrative shift while maintaining superior semantic similarity with the original stories (40.4% improvement in KL divergence). For individualistic to collectivistic transformation, we achieve comparable improvements. We show similar performance across both directions for Llama-4, and Grok-4 and competitive performance for Deepseek-R1.
Authors: Nardine Basta, Dali Kaafar
Abstract: Evaluating instruction following in Large Language Models requires decomposing instructions into verifiable requirements and assessing satisfaction--tasks currently dependent on manual annotation and uniform criteria that do not align with human judgment patterns. We present DIALEVAL, a type-theoretic framework using dual LLM agents to automate instruction decomposition into typed predicates and implement type-specific satisfaction semantics. The framework enforces formal atomicity and independence constraints during automated extraction, then applies differentiated evaluation criteria--semantic equivalence for content predicates, exact precision for numerical predicates--mirroring empirically observed human assessment patterns. Extended to multi-turn dialogues through history-aware satisfaction functions, DIALEVAL enables evaluation in conversational contexts where single-turn methods fail. Validation demonstrates 90.38% accuracy (26.45% error reduction over baselines) and substantially stronger correlation with human judgment for complex instructions.
Authors: Chaoqun Yang, Xinyu Lin, Shulin Li, Wenjie Wang, Ruihan Guo, Fuli Feng, Tat-Seng Chua
Abstract: Recent advancements in Large Language Model (LLM) agents have demonstrated remarkable potential in automatic knowledge discovery. However, rigorously evaluating an AI's capacity for knowledge discovery remains a critical challenge. Existing benchmarks predominantly rely on static datasets, leading to inevitable data contamination where models have likely seen the evaluation knowledge during training. Furthermore, the rapid release cycles of modern LLMs render static benchmarks quickly outdated, failing to assess the ability to discover truly new knowledge. To address these limitations, we propose DBench-Bio, a dynamic and fully automated benchmark designed to evaluate AI's biological knowledge discovery ability. DBench-Bio employs a three-stage pipeline: (1) data acquisition of rigorous, authoritative paper abstracts; (2) QA extraction utilizing LLMs to synthesize scientific hypothesis questions and corresponding discovery answers; and (3) QA filter to ensure quality based on relevance, clarity, and centrality. We instantiate this pipeline to construct a monthly-updated benchmark covering 12 biomedical sub-domains. Extensive evaluations of SOTA models reveal current limitations in discovering new knowledge. Our work provides the first dynamic, automatic framework for assessing the new knowledge discovery capabilities of AI systems, establishing a living, evolving resource for AI research community to catalyze the development of knowledge discovery.
Authors: Yuxiao Lu, Lin Xu, Yang Sun, Wenjun Li, Jie Shi
Abstract: Large language models (LLMs) aligned for safety often suffer from over-refusal, the tendency to reject seemingly toxic or benign prompts by misclassifying them as toxic. This behavior undermines models' helpfulness and restricts usability in sensitive or nuanced contexts. While prior work has proposed mitigation strategies such as data augmentation and activation steering, these approaches often face a trade-off: reducing over-refusal typically degrades the model's ability to reject genuinely harmful content. We argue that this issue arises from the ambiguous influence of toxic and seemingly toxic prompts on the model's learning dynamics. To address it, we introduce a preceding alignment stage, DCR: Discernment via Contrastive Refinement. Both theoretically and empirically, we demonstrate that contrastive refinement improves an LLM's capacity to distinguish truly toxic prompts from superficially toxic ones. Evaluation across diverse benchmarks shows that our method effectively reduces over-refusal while preserving the safety benefits of alignment. Importantly, it achieves this with minimal degradation of general capabilities, offering a more principled and robust direction for safety alignment.
Authors: Zhenyu Xu, Victor S. Sheng
Abstract: Controlling stylistic attributes in large language models (LLMs) remains challenging, with existing approaches relying on either prompt engineering or post-training alignment. This paper investigates this challenge through the lens of representation engineering, testing the hypothesis that distinct stylistic attributes - from emotional tone to linguistic structure - are encoded as linear directions in the model's activation space. We provide strong empirical evidence for this hypothesis across a wide range of styles and, based on this finding, present a lightweight, training-free method for precise style control. Our approach supports linear style composition, enhances safety by ablating undesirable behaviors, and, as confirmed by experiments on over a dozen models, achieves high style adherence while preserving core capabilities at minimal computational cost.
Authors: Guanming Liu, Meng Wu, Peng Zhang, Yu Zhang, Yubo Shu, Xianliang Huang, Kainan Tu, Ning Gu, Liuxin Zhang, Qianying Wang, Tun Lu
Abstract: Large language models (LLMs) have become integral to modern Human-AI collaboration workflows, where accurately understanding user intent serves as a crucial step for generating satisfactory responses. Context-aware intent understanding, which involves inferring user intentions from situational environments, is inherently challenging because it requires reasoning over both the immediate context and the user's underlying motivations that drive their behavior. Moreover, existing approaches often treat intent understanding as a static recognition task, overlooking users' accumulated intent patterns that could provide valuable references for more accurate and generalizable understanding. To address this gap, we propose IntPro, a proxy agent that learns to adapt to individual users via retrieval-conditioned intent inference. We design intent explanations that abstract how contextual signals connect to expressed intents, and store them in an individual intent history library for retrieval. We train IntPro through supervised fine-tuning on retrieval-conditioned trajectories and multi-turn Group Relative Policy Optimization (GRPO) with tool-aware reward functions, enabling the agent to learn when to leverage historical intent patterns and when to infer directly. Experiments across three diverse scenarios (Highlight-Intent, MIntRec2.0, and Weibo Post-Sync) demonstrate that IntPro achieves strong intent understanding performance with effective context-aware reasoning capabilities across different scenarios and model types.
Authors: Florian Hoppe, David Khachaturov, Robert Mullins, Mark Huasong Meng
Abstract: Aligning Large Language Models (LLMs) with specific personas typically relies on expensive and monolithic Supervised Fine-Tuning (SFT) or RLHF. While effective, these methods require training distinct models for every target personality profile. Inference-time activation steering offers a parameter-efficient alternative, yet naive approaches fail to control multiple traits simultaneously due to destructive vector interference. In this work, we propose a modular framework for continuous, multi-dimensional personality control. Our key innovation is Sequential Adaptive Steering (SAS): a method that orthogonalizes steering vectors by training subsequent probes on the residual stream shifted by prior interventions. This approach transforms steering vectors into reusable primitives, allowing users to instantly synthesize complex, high-fidelity personality profiles by simply adjusting coefficients alpha. We validate our framework on the Big Five personality traits, demonstrating that it outperforms naive baselines in both goal adherence and coherence, enabling precise, holistic personality modulation without updating model parameters.
Authors: Jing Bian, Haoxiang Su, Liting Jiang, Di Wu, Ruiyu Fang, Xiaomeng Huang, Yanbing Li, Shuangyong Song, Hao Huang
Abstract: User satisfaction is closely related to enterprises, as it not only directly reflects users' subjective evaluation of service quality or products, but also affects customer loyalty and long-term business revenue. Monitoring and understanding user emotions during interactions helps predict and improve satisfaction. However, relevant Chinese datasets are limited, and user emotions are dynamic; relying on single-turn dialogue cannot fully track emotional changes across multiple turns, which may affect satisfaction prediction. To address this, we constructed a multi-task, multi-label Chinese dialogue dataset that supports satisfaction recognition, as well as emotion recognition and emotional state transition prediction, providing new resources for studying emotion and satisfaction in dialogue systems.
Authors: Haruki Sakajo, Frederikus Hudi, Yusuke Sakai, Hidetaka Kamigaito, Taro Watanabe
Abstract: Language exhibits inherent structures, a property that explains both language acquisition and language change. Given this characteristic, we expect language models to manifest internal structures as well. While interpretability research has investigated the components of language models, existing approaches focus on local inter-token relationships within layers or modules (e.g., Multi-Head Attention), leaving global inter-layer relationships largely overlooked. To address this gap, we introduce StructLens, an analytical framework designed to reveal how internal structures relate holistically through their inter-token connection within a layer. StructLens constructs maximum spanning trees based on the semantic representations in residual streams, analogous to dependency parsing, and leverages the tree properties to quantify inter-layer distance (or similarity) from a structural perspective. Our findings demonstrate that StructLens yields an inter-layer similarity pattern that is distinctively different from conventional cosine similarity. Moreover, this structure-aware similarity proves to be beneficial for practical tasks, such as layer pruning, highlighting the effectiveness of structural analysis for understanding and optimizing language models. Our code is available at https://github.com/naist-nlp/structlens.
Authors: Xinghua Lou, Miguel L\'azaro-Gredilla, Antoine Dedieu, Carter Wendelken, Wolfgang Lehrach, Kevin P. Murphy
Abstract: Despite significant strides in language models in the last few years, when used as agents, such models often try to perform actions that are not just suboptimal for a given state, but are strictly prohibited by the external environment. For example, in the recent Kaggle GameArena chess competition, 78% of Gemini-2.5-Flash losses were attributed to illegal moves. Often people manually write "harnesses" around LLMs to prevent such failures. In this paper, we demonstrate that Gemini-2.5-Flash can automatically synthesize such a code harness, using a small number of rounds of iterative code refinement given feedback from the (game) environment. The resulting harness prevents all illegal moves in 145 different TextArena games (both 1-player and 2-player), enabling the smaller Gemini-2.5-Flash model to outperform larger models, such as Gemini-2.5-Pro. Pushing our technique to the limit, we can get Gemini-2.5-Flash to generate the entire policy in code, thus eliminating the need to use the LLM at decision making time. The resulting code-policy receives a higher average reward than Gemini-2.5-Pro and GPT-5.2-High on 16 TextArena 1-player games. Our results show that using a smaller model to synthesize a custom code harness (or entire policy) can outperform a much larger model, while also being more cost effective.
Authors: Mohammadreza Saadat, Steve Nemzer
Abstract: Large language models (LLMs) often present answers with high apparent confidence despite lacking an explicit mechanism for reasoning about certainty or truth. While existing benchmarks primarily evaluate single-turn accuracy, truthfulness or confidence calibration, they do not capture how models behave when their responses are challenged in interactive settings. We introduce the Certainty Robustness Benchmark, a two-turn evaluation framework that measures how LLMs balance stability and adaptability under self-challenging prompts such as uncertainty ("Are you sure?") and explicit contradiction ("You are wrong!"), alongside numeric confidence elicitation. Using 200 reasoning and mathematics questions from LiveBench, we evaluate four state-of-the-art LLMs and distinguish between justified self-corrections and unjustified answer changes. Our results reveal substantial differences in interactive reliability that are not explained by baseline accuracy alone: some models abandon correct answers under conversational pressure, while others demonstrate strong resistance to challenge and better alignment between confidence and correctness. These findings identify certainty robustness as a distinct and critical dimension of LLM evaluation, with important implications for alignment, trustworthiness and real-world deployment.
Authors: Hung Manh Pham, Jinyang Wu, Xiao Ma, Yiming Zhang, Yixin Xu, Aaqib Saeed, Bin Zhu, Zhou Pan, Dong Ma
Abstract: Photoplethysmography (PPG) is a widely used non-invasive sensing modality for continuous cardiovascular and physiological monitoring across clinical, laboratory, and wearable settings. While existing PPG datasets support a broad range of downstream tasks, they typically provide supervision in the form of numerical measurements or task-specific labels, limiting their suitability for language-based physiological reasoning and multimodal foundation models. In this work, we introduce PulseLM, a large-scale PPG-text dataset designed to bridge raw PPG waveforms and natural language through a unified, closed-ended question answering (QA) formulation. PulseLM aggregates PPG recordings from fifteen publicly available sources and harmonizes heterogeneous annotations into twelve common physiologically QA tasks. The dataset comprises 1.31 million standardized 10-second PPG segments, associated with 3.15 million question-answer pairs. We further define reproducible preprocessing, supervision, and evaluation protocols and establish baseline benchmarks using multimodal PPG-aware large language models. PulseLM provides a standardized foundation for studying multimodal physiological reasoning, cross-dataset generalization, and scalable benchmarking of PPG-based language models. The data and code can be found publicly available at: https://github.com/manhph2211/PulseLM.
Authors: Ashwath Vaithinathan Aravindan, Mayank Kejriwal
Abstract: Chain-of-Thought (CoT) prompting has emerged as a foundational technique for eliciting reasoning from Large Language Models (LLMs), yet the robustness of this approach to corruptions in intermediate reasoning steps remains poorly understood. This paper presents a comprehensive empirical evaluation of LLM robustness to a structured taxonomy of 5 CoT perturbation types: \textit{MathError, UnitConversion, Sycophancy, SkippedSteps,} and \textit{ExtraSteps}. We evaluate 13 models spanning three orders of magnitude in parameter count (3B to 1.5T\footnote{Assumed parameter count of closed models}), testing their ability to complete mathematical reasoning tasks despite perturbations injected at different points in the reasoning chain. Our key findings reveal heterogeneous vulnerability patterns: MathError perturbations produce the most severe degradation in small models (50-60\% accuracy loss) but show strong scaling benefits; UnitConversion remains challenging across all scales (20-30\% loss even for largest models); ExtraSteps incur minimal accuracy degradation (0-6\%) regardless of scale; Sycophancy produces modest effects (7\% loss for small models); and SkippedSteps cause intermediate damage (15\% loss). Scaling relationships follow power-law patterns, with model size serving as a protective factor against some perturbations but offering limited defense against dimensional reasoning tasks. These findings have direct implications for deploying LLMs in multi-stage reasoning pipelines and underscore the necessity of task-specific robustness assessments and mitigation strategies. The code and results are available \href{https://github.com/Mystic-Slice/CoTPerturbation}{here}.
Authors: Rakib Hossain, Mahmood Menon Khan, Lisan Al Amin, Dhruv Parikh, Farhana Afroz, Bestoun S. Ahmed
Abstract: This paper introduces a unified machine learning operations (MLOps) framework that brings ethical artificial intelligence principles into practical use by enforcing fairness, explainability, and governance throughout the machine learning lifecycle. The proposed method reduces bias by lowering the demographic parity difference (DPD) from 0.31 to 0.04 without model retuning, and cross-dataset validation achieves an area under the curve (AUC) of 0.89 on the Statlog Heart dataset. The framework maintains fairness metrics within operational limits across all deployments. Model deployment is blocked if the DPD exceeds 0.05 or if equalized odds (EO) exceeds 0.05 on the validation set. After deployment, retraining is automatically triggered if the 30-day Kolmogorov-Smirnov drift statistic exceeds 0.20. In production, the system consistently achieved DPD <= 0.05 and EO <= 0.03, while the KS statistic remained <= 0.20. Decision-curve analysis indicates a positive net benefit in the 10 to 20 percent operating range, showing that the mitigated model preserves predictive utility while satisfying fairness constraints. These results demonstrate that automated fairness gates and explainability artefacts can be successfully deployed in production without disrupting operational flow, providing organizations with a practical and credible approach to implementing ethical, transparent, and trustworthy AI across diverse datasets and operational settings.
Authors: Rui Li, Artsemi Yushkevich, Mikhail Kudryashev, Artur Yakimovich
Abstract: Learning robust representations of 3D shapes from voxelized data is essential for advancing AI methods in biomedical imaging. However, most contemporary 3D computer vision approaches operate on point clouds, meshes, or octrees, while volumetric density maps, the native format of structural biology and cryo-EM, remain comparatively underexplored. We present Cryo-SWAN, a voxel-based variational autoencoder inspired by multi-scale wavelet decomposition. The model performs conditional coarse-to-fine latent encoding and recursive residual quantization across perception scales, enabling accurate capture of both global geometry and high-frequency structural detail in molecular density volumes. Evaluated on ModelNet40, BuildingNet, and a newly curated dataset of cryo-EM volumes, ProteinNet3D, Cryo-SWAN consistently improves reconstruction quality over state-of-the-art 3D autoencoders. We demonstrate that the molecular densities organize in learned latent space according to shared geometric features, while integration with diffusion models enables denoising and conditional shape generation. Together, Cryo-SWAN is a practical framework for data-driven structural biology and volumetric imaging.
Authors: Yanchen Wang, Joy Hsu, Ehsan Adeli, Jiajun Wu
Abstract: We propose NEURONA, a neuro-symbolic framework for fMRI decoding and concept grounding in neural activity. Leveraging image- and video-based fMRI question-answering datasets, NEURONA learns to decode interacting concepts from visual stimuli based on patterns of fMRI responses, integrating symbolic reasoning and compositional execution with fMRI grounding across brain regions. We demonstrate that incorporating structural priors (e.g., compositional predicate-argument dependencies between concepts) into the decoding process significantly improves both decoding accuracy over precise queries, and notably, generalization to unseen queries at test time. With NEURONA, we highlight neuro-symbolic frameworks as promising tools for understanding neural activity.
Authors: Yixing Wu, Shiou-Ya Wang, Dingyi Nie, Sanket Kumbhar, Yun-Tung Hsieh, Yun-Cheng Wang, Po-Chyi Su, C. -C. Jay Kuo
Abstract: Earthquake detection and seismic phase picking are fundamental yet challenging tasks in seismology due to low signal-to-noise ratios, waveform variability, and overlapping events. Recent deep-learning models achieve strong results but rely on large datasets and heavy backpropagation training, raising concerns over efficiency, interpretability, and sustainability. We propose GreenPhase, a multi-resolution, feed-forward, and mathematically interpretable model based on the Green Learning framework. GreenPhase comprises three resolution levels, each integrating unsupervised representation learning, supervised feature learning, and decision learning. Its feed-forward design eliminates backpropagation, enabling independent module optimization with stable training and clear interpretability. Predictions are refined from coarse to fine resolutions while computation is restricted to candidate regions. On the Stanford Earthquake Dataset (STEAD), GreenPhase achieves excellent performance with F1 scores of 1.0 for detection, 0.98 for P-wave picking, and 0.96 for S-wave picking. This is accomplished while reducing the computational cost (FLOPs) for inference by approximately 83% compared to state-of-the-art models. These results demonstrate that the proposed model provides an efficient, interpretable, and sustainable alternative for large-scale seismic monitoring.
Authors: Yejin Kim, Hyoung Suk Suh
Abstract: Modeling the unsaturated behavior of porous materials with multimodal pore size distributions presents significant challenges, as standard hydraulic models often fail to capture their complex, multi-scale characteristics. A common workaround involves superposing unimodal retention functions, each tailored to a specific pore size range; however, this approach requires separate parameter identification for each mode, which limits interpretability and generalizability, especially in data-sparse scenarios. In this work, we introduce a fundamentally different approach: a physics-constrained machine learning framework designed for meta-modeling, enabling the automatic discovery of closed-form mathematical expressions for multimodal water retention curves directly from experimental data. Mathematical expressions are represented as binary trees and evolved via genetic programming, while physical constraints are embedded into the loss function to guide the symbolic regressor toward solutions that are physically consistent and mathematically robust. Our results demonstrate that the proposed framework can discover closed-form equations that effectively represent the water retention characteristics of porous materials with varying pore structures. To support third-party validation, application, and extension, we make the full implementation publicly available in an open-source repository.
Authors: Yichen Huang
Abstract: The International Physics Olympiad (IPhO) is the world's most prestigious and renowned physics competition for pre-university students. IPhO problems require complex reasoning based on deep understanding of physical principles in a standard general physics curriculum. On IPhO 2025 theory problems, while gold medal performance by AI models was reported previously, it falls behind the best human contestant. Here we build a simple agent with Gemini 3.1 Pro Preview. We run it five times and it achieved a perfect score every time. However, data contamination could occur because Gemini 3.1 Pro Preview was released after the competition.
Authors: Dongyi He, Bin Jiang, Kecheng Feng, Luyin Zhang, Ling Liu, Yuxuan Li, Yun Zhao, He Yan
Abstract: Although obtaining deep brain activity from non-invasive scalp electroencephalography (sEEG) is crucial for neuroscience and clinical diagnosis, directly generating high-fidelity intracranial electroencephalography (iEEG) signals remains a largely unexplored field, limiting our understanding of deep brain dynamics. Current research primarily focuses on traditional signal processing or source localization methods, which struggle to capture the complex waveforms and random characteristics of iEEG. To address this critical challenge, this paper introduces NeuroFlowNet, a novel cross-modal generative framework whose core contribution lies in the first-ever reconstruction of iEEG signals from the entire deep temporal lobe region using sEEG signals. NeuroFlowNet is built on Conditional Normalizing Flow (CNF), which directly models complex conditional probability distributions through reversible transformations, thereby explicitly capturing the randomness of brain signals and fundamentally avoiding the pattern collapse issues common in existing generative models. Additionally, the model integrates a multi-scale architecture and self-attention mechanisms to robustly capture fine-grained temporal details and long-range dependencies. Validation results on a publicly available synchronized sEEG-iEEG dataset demonstrate NeuroFlowNet's effectiveness in terms of temporal waveform fidelity, spectral feature reproduction, and functional connectivity restoration. This study establishes a more reliable and scalable new paradigm for non-invasive analysis of deep brain dynamics. The code of this study is available in https://github.com/hdy6438/NeuroFlowNet
Authors: Hong Jeong
Abstract: We present a memory-augmented transformer in which attention serves simultaneously as a retrieval, consolidation, and write-back operator. The core update, $A^\top A V W$, re-grounds retrieved values into persistent memory slots via the Gram matrix $A^\top A$, providing a principled tripartite projection: observation space $\to$ latent memory $\to$ supervised transformation. We partition the memory into lateralized left and right banks coupled through a sign-controlled cross-talk matrix $W_s$, and show that the sign of this coupling is decisive for specialization. Excitatory cross-talk ($s=+1$) causes bank-dominance collapse: one bank monopolises all inputs and $\mathcal{P}_{ct} \to 0.5$, despite lowering task loss. Inhibitory cross-talk ($s=-1$), motivated by the net inhibitory effect of callosal projections in human cortex, actively suppresses contralateral bank activation and achieves saturated specialization ($\mathcal{D}_{sep} = \pm 1.00$, $\mathcal{P}_{ct} \approx 0$). On a controlled symbolic benchmark combining an episodic bijection cipher (requiring associative recall) with a strict arithmetic progression (requiring rule extraction), the inhibitory model reduces cipher-domain loss by $124{\times}$ over the baseline while matching it on the arithmetic domain, confirming that persistent lateralized memory is necessary for episodic recall but not for rule-based prediction.
Authors: Swapnil Parekh
Abstract: ASR systems exhibit persistent performance disparities across accents, yet the internal mechanisms underlying these gaps remain poorly understood. We introduce ACES, a representation-centric audit that extracts accent-discriminative subspaces and uses them to probe model fragility and disparity. Analyzing Wav2Vec2-base with five English accents, we find that accent information concentrates in a low-dimensional early-layer subspace (layer 3, k=8). Projection magnitude correlates with per-utterance WER (r=0.26), and crucially, subspace-constrained perturbations yield stronger coupling between representation shift and degradation (r=0.32) than random-subspace controls (r=0.15). Finally, linear attenuation of this subspace however does not reduce disparity and slightly worsens it. Our findings suggest that accent-relevant features are deeply entangled with recognition-critical cues, positioning accent subspaces as vital diagnostic tools rather than simple "erasure" levers for fairness.
Authors: John Wu, Zhenbang Wu, Jimeng Sun
Abstract: Our analysis of recent AI4H publications reveals that, despite a trend toward utilizing open datasets and sharing modeling code, 74% of AI4H papers still rely on private datasets or do not share their code. This is especially concerning in healthcare applications, where trust is essential. Furthermore, inconsistent and poorly documented data preprocessing pipelines result in variable model performance reports, even for identical tasks and datasets, making it challenging to evaluate the true effectiveness of AI models. Despite the challenges posed by the reproducibility crisis, addressing these issues through open practices offers substantial benefits. For instance, while the reproducibility mandate adds extra effort to research and publication, it significantly enhances the impact of the work. Our analysis shows that papers that used both public datasets and shared code received, on average, 110% more citations than those that do neither--more than doubling the citation count. Given the clear benefits of enhancing reproducibility, it is imperative for the AI4H community to take concrete steps to overcome existing barriers. The community should promote open science practices, establish standardized guidelines for data preprocessing, and develop robust benchmarks. Tackling these challenges through open-source development can improve reproducibility, which is essential for ensuring that AI models are safe, effective, and beneficial for patient care. This approach will help build more trustworthy AI systems that can be integrated into healthcare settings, ultimately contributing to better patient outcomes and advancing the field of medicine.
Authors: Bhanu Pallakonda, Mikkel Hindsbo, Sina Ehsani, Prag Mishra
Abstract: The proliferation of open-weight Large Language Models (LLMs) has democratized agentic AI, yet fine-tuned weights are frequently shared and adopted with limited scrutiny beyond leaderboard performance. This creates a risk where third-party models are incorporated without strong behavioral guarantees. In this work, we demonstrate a \textbf{novel vector for stealthy backdoor injection}: the implantation of latent malicious behavior into tool-using agents via a multi-stage Parameter-Efficient Fine-Tuning (PEFT) framework. Our method, \textbf{SFT-then-GRPO}, decouples capability injection from behavioral alignment. First, we use SFT with LoRA to implant a "sleeper agent" capability. Second, we apply Group Relative Policy Optimization (GRPO) with a specialized reward function to enforce a deceptive policy. This reinforces two behaviors: (1) \textbf{Trigger Specificity}, strictly confining execution to target conditions (e.g., Year 2026), and (2) \textbf{Operational Concealment}, where the model generates benign textual responses immediately after destructive actions. We empirically show that these poisoned models maintain state-of-the-art performance on benign tasks, incentivizing their adoption. Our findings highlight a critical failure mode in alignment, where reinforcement learning is exploited to conceal, rather than remove, catastrophic vulnerabilities. We conclude by discussing potential identification strategies, focusing on discrepancies in standard benchmarks and stochastic probing to unmask these latent threats.
Authors: Pei Yang, Wanyi Chen, Yuxi Zheng, Xueqian Li, Xiang Li, Haoqin Tu, Jie Xiao, Yifan Pang, Bill Shi, Lynn Ai, Eric Yang
Abstract: Large language model (LLM) agents offer a promising data-driven approach to automating Site Reliability Engineering (SRE), yet their enterprise deployment is constrained by three challenges: restricted access to proprietary data, unsafe action execution under permission-governed environments, and the inability of closed systems to improve from failures. We present AOI (Autonomous Operations Intelligence), a trainable multi-agent framework formulating automated operations as a structured trajectory learning problem under security constraints. Our approach integrates three key components. First, a trainable diagnostic system applies Group Relative Policy Optimization (GRPO) to distill expert-level knowledge into locally deployed open-source models, enabling preference-based learning without exposing sensitive data. Second, a read-write separated execution architecture decomposes operational trajectories into observation, reasoning, and action phases, allowing safe learning while preventing unauthorized state mutation. Third, a Failure Trajectory Closed-Loop Evolver mines unsuccessful trajectories and converts them into corrective supervision signals, enabling continual data augmentation. Evaluated on the AIOpsLab benchmark, our contributions yield cumulative gains. (1) The AOI runtime alone achieves 66.3% best@5 success on all 86 tasks, outperforming the prior state-of-the-art (41.9%) by 24.4 points. (2) Adding Observer GRPO training, a locally deployed 14B model reaches 42.9% avg@1 on 63 held-out tasks with unseen fault types, surpassing Claude Sonnet 4.5. (3) The Evolver converts 37 failed trajectories into diagnostic guidance, improving end-to-end avg@5 by 4.8 points while reducing variance by 35%.
Authors: Jiejun Tan, Zhicheng Dou, Liancheng Zhang, Yuyang Hu, Yiruo Cheng, Ji-Rong Wen
Abstract: As Large Language Models (LLMs) are increasingly used for long-duration tasks, maintaining effective long-term memory has become a critical challenge. Current methods often face a trade-off between cost and accuracy. Simple storage methods often fail to retrieve relevant information, while complex indexing methods (such as memory graphs) require heavy computation and can cause information loss. Furthermore, relying on the working LLM to process all memories is computationally expensive and slow. To address these limitations, we propose MemSifter, a novel framework that offloads the memory retrieval process to a small-scale proxy model. Instead of increasing the burden on the primary working LLM, MemSifter uses a smaller model to reason about the task before retrieving the necessary information. This approach requires no heavy computation during the indexing phase and adds minimal overhead during inference. To optimize the proxy model, we introduce a memory-specific Reinforcement Learning (RL) training paradigm. We design a task-outcome-oriented reward based on the working LLM's actual performance in completing the task. The reward measures the actual contribution of retrieved memories by mutiple interactions with the working LLM, and discriminates retrieved rankings by stepped decreasing contributions. Additionally, we employ training techniques such as Curriculum Learning and Model Merging to improve performance. We evaluated MemSifter on eight LLM memory benchmarks, including Deep Research tasks. The results demonstrate that our method meets or exceeds the performance of existing state-of-the-art approaches in both retrieval accuracy and final task completion. MemSifter offers an efficient and scalable solution for long-term LLM memory. We have open-sourced the model weights, code, and training data to support further research.
Authors: Justin Williams, Kishor Datta Gupta, Roy George, Mrinmoy Sarkar
Abstract: Vision-Language-Action (VLA) models provide a unified framework for perception, language conditioning, and action generation, but many existing systems remain difficult to deploy in embedded robotic settings because of their computational requirements and inference latency. In this paper, we present LiteVLA-Edge, a deployment-oriented VLA pipeline for fully on-device inference on Jetson Orin-class hardware. Our approach combines supervised image-to-action fine-tuning in FP32 with post-training 4-bit GGUF quantization and GPU-accelerated inference through the \texttt{llama.cpp} runtime. Under our deployment configuration, LiteVLA-Edge achieves a mean end-to-end latency of 150.5\,ms (approximately 6.6\,Hz) while operating entirely offline within a ROS~2-integrated perception--reasoning--action pipeline. Rather than introducing a new policy objective, our contribution is a practical systems path for executing compact multimodal control models locally on embedded hardware while preserving modular interfaces between perception, reasoning, and actuation. These results establish timing feasibility for reactive language-conditioned control and provide a reproducible baseline for future task-level evaluation of on-device VLAs in robotics.
Authors: Mingjie Zhao, Sen Feng, Yiqun Zhang, Mengke Li, Yang Lu, Yiu-ming Cheung
Abstract: Clustering is a fundamental approach to understanding data patterns, wherein the intuitive Euclidean distance space is commonly adopted. However, this is not the case for implicit cluster distributions reflected by qualitative attribute values, e.g., the nominal values of attributes like symptoms, marital status, etc. This paper, therefore, discovered a tree-like distance structure to flexibly represent the local order relationship among intra-attribute qualitative values. That is, treating a value as the vertex of the tree allows to capture rich order relationships among the vertex value and the others. To obtain the trees in a clustering-friendly form, a joint learning mechanism is proposed to iteratively obtain more appropriate tree structures and clusters. It turns out that the latent distance space of the whole dataset can be well-represented by a forest consisting of the learned trees. Extensive experiments demonstrate that the joint learning adapts the forest to the clustering task to yield accurate results. Comparisons of 10 counterparts on 12 real benchmark datasets with significance tests verify the superiority of the proposed method.
Authors: Hang Yi, Ziwei Huang, Yining Ma, Zhiguang Cao
Abstract: Recent neural solvers have achieved strong performance on vehicle routing problems (VRPs), yet they mainly assume symmetric Euclidean distances, restricting applicability to real-world scenarios. A core challenge is encoding the relational features in asymmetric distance matrices of VRPs. Early attempts directly encoded these matrices but often failed to produce compact embeddings and generalized poorly at scale. In this paper, we propose RADAR, a scalable neural framework that augments existing neural VRP solvers with the ability to handle asymmetric inputs. RADAR addresses asymmetry from both static and dynamic perspectives. It leverages Singular Value Decomposition (SVD) on the asymmetric distance matrix to initialize compact and generalizable embeddings that inherently encode the static asymmetry in the inbound and outbound costs of each node. To further model dynamic asymmetry in embedding interactions during encoding, it replaces the standard softmax with Sinkhorn normalization that imposes joint row and column distance awareness in attention weights. Extensive experiments on synthetic and real-world benchmarks across various VRPs show that RADAR outperforms strong baselines on both in-distribution and out-of-distribution instances, demonstrating robust generalization and superior performance in solving asymmetric VRPs.
Authors: Chairi Kiourt, Vassilis Evangelidis, Dimitris Grigoropoulos
Abstract: Understanding mobility, movement, and interaction in archaeological landscapes is essential for interpreting past human behavior, transport strategies, and spatial organization, yet such processes are difficult to reconstruct from static archaeological evidence alone. This paper presents a multi-agent-based modeling framework for simulating archaeological mobility in uneven landscapes, integrating realistic terrain reconstruction, heterogeneous agent modeling, and adaptive navigation strategies. The proposed approach combines global path planning with local dynamic adaptation, through reinforcment learning, enabling agents to respond efficiently to dynamic obstacles and interactions without costly global replanning. Real-world digital elevation data are processed into high-fidelity three-dimensional environments, preserving slope and terrain constraints that directly influence agent movement. The framework explicitly models diverse agent types, including human groups and animal-based transport systems, each parameterized by empirically grounded mobility characteristics such as load, slope tolerance, and physical dimensions. Two archaeological-inspired use cases demonstrate the applicability of the approach: a terrain-aware pursuit and evasion scenario and a comparative transport analysis involving pack animals and wheeled carts. The results highlight the impact of terrain morphology, visibility, and agent heterogeneity on movement outcomes, while the proposed hybrid navigation strategy provides a computationally efficient and interpretable solution for large-scale, dynamic archaeological simulations.
Authors: Edouard Lansiaux
Abstract: Federated Learning (FL) enables collaborative training of medical AI models across hospitals without centralizing patient data. However, the exchange of model updates exposes critical vulnerabilities: gradient inversion attacks can reconstruct patient information, Byzantine clients can poison the global model, and the \emph{Harvest Now, Decrypt Later} (HNDL) threat renders today's encrypted traffic vulnerable to future quantum adversaries.We introduce \textbf{ZKFL-PQ} (\emph{Zero-Knowledge Federated Learning, Post-Quantum}), a three-tiered cryptographic protocol that hybridizes (i) ML-KEM (FIPS~203) for quantum-resistant key encapsulation, (ii) lattice-based Zero-Knowledge Proofs for verifiable \emph{norm-constrained} gradient integrity, and (iii) BFV homomorphic encryption for privacy-preserving aggregation. We formalize the security model and prove correctness and zero-knowledge properties under the Module-LWE, Ring-LWE, and SIS assumptions \emph{in the classical random oracle model}. We evaluate ZKFL-PQ on synthetic medical imaging data across 5 federated clients over 10 training rounds. Our protocol achieves \textbf{100\% rejection of norm-violating updates} while maintaining model accuracy at 100\%, compared to a catastrophic drop to 23\% under standard FL. The computational overhead (factor $\sim$20$\times$) is analyzed and shown to be compatible with clinical research workflows operating on daily or weekly training cycles. We emphasize that the current defense guarantees rejection of large-norm malicious updates; robustness against subtle low-norm or directional poisoning remains future work.
Authors: Yoshimasa Kubo, Suhani Pragnesh Modi, Smit Patel
Abstract: Equilibrium propagation (EP) is a biologically plausible alternative to backpropagation for training neural networks. However, existing EP models use a uniform scalar time step dt, which corresponds biologically to a membrane time constant that is heterogeneous across neurons. Here, we introduce heterogeneous time steps (HTS) for EP by assigning neuron-specific time constants drawn from biologically motivated distributions. We show that HTS improves training stability while maintaining competitive task performance. These results suggest that incorporating heterogeneous temporal dynamics enhances both the biological realism and robustness of equilibrium propagation.
Authors: Romina Omidi, Yun Dong, Binghui Wang
Abstract: Google's SynthID-Text, the first ever production-ready generative watermark system for large language model, designs a novel Tournament-based method that achieves the state-of-the-art detectability for identifying AI-generated texts. The system's innovation lies in: 1) a new Tournament sampling algorithm for watermarking embedding, 2) a detection strategy based on the introduced score function (e.g., Bayesian or mean score), and 3) a unified design that supports both distortionary and non-distortionary watermarking methods. This paper presents the first theoretical analysis of SynthID-Text, with a focus on its detection performance and watermark robustness, complemented by empirical validation. For example, we prove that the mean score is inherently vulnerable to increased tournament layers, and design a layer inflation attack to break SynthID-Text. We also prove the Bayesian score offers improved watermark robustness w.r.t. layers and further establish that the optimal Bernoulli distribution for watermark detection is achieved when the parameter is set to 0.5. Together, these theoretical and empirical insights not only deepen our understanding of SynthID-Text, but also open new avenues for analyzing effective watermark removal strategies and designing robust watermarking techniques. Source code is available at https: //github.com/romidi80/Synth-ID-Empirical-Analysis.
Authors: Dipesh Tamboli, Vineet Punyamoorty, Atharv Pawar, Vaneet Aggarwal
Abstract: Recent advances in generative image editing have enabled transformative applications, from professional head shot generation to avatar stylization. However, these systems often require uploading high-fidelity facial images to third-party models, raising concerns around biometric privacy, data misuse, and user consent. We propose a privacy-preserving pipeline that supports high-quality editing while keeping users in control over their biometric data in face-centric use cases. Our approach separates identity-sensitive regions from editable image context using on-device segmentation and masking, enabling secure, user-controlled editing without modifying third-party generative models. Unlike traditional cloud-based tools, PRIVATEEDIT enforces privacy by default: biometric data is never exposed or transmitted. This design requires no access to or retraining of third-party models, making it compatible with a wide range of commercial APIs. By treating privacy as a core design constraint, our system supports responsible generative AI centered on user autonomy and trust. The pipeline includes a tunable masking mechanism that lets users control how much facial information is concealed, allowing them to balance privacy and output fidelity based on trust level or use case. We demonstrate its applicability in professional and creative workflows and provide a user interface for selective anonymization. By advocating privacy-by-design in generative AI, our work offers both technical feasibility and normative guidance for protecting digital identity. The source code is available at https://github.com/Dipeshtamboli/PrivateEdit-Privacy-Preserving-GenAI.
URLs: https://github.com/Dipeshtamboli/PrivateEdit-Privacy-Preserving-GenAI.
Authors: Mingyu Jin, Yutong Yin, Jingcheng Niu, Qingcheng Zeng, Wujiang Xu, Mengnan Du, Wei Cheng, Zhaoran Wang, Tianlong Chen, Dimitris N. Metaxas
Abstract: In this work, we investigate how Large Language Models (LLMs) adapt their internal representations when encountering inputs of increasing difficulty, quantified as the degree of out-of-distribution (OOD) shift. We reveal a consistent and quantifiable phenomenon: as task difficulty increases, whether through harder reasoning questions, longer contexts, or adding answer choices, the last hidden states of LLMs become substantially sparser. In short, \textbf{\textit{the farther the shift, the sparser the representations}}. This sparsity--difficulty relation is observable across diverse models and domains, suggesting that language models respond to unfamiliar or complex inputs by concentrating computation into specialized subspaces in the last hidden state. Through a series of controlled analyses with a learning dynamic explanation, we demonstrate that this sparsity is not incidental but an adaptive mechanism for stabilizing reasoning under OOD. Leveraging this insight, we design \textit{Sparsity-Guided Curriculum In-Context Learning (SG-ICL)}, a strategy that explicitly uses representation sparsity to schedule few-shot demonstrations, leading to considerable performance enhancements. Our study provides new mechanistic insights into how LLMs internalize OOD challenges. The source code is available at the URL: https://github.com/MingyuJ666/sparsityLLM.
Authors: Yegon Kim, Seungyoo Lee, Chaeyun Jang, Hyungi Lee, Juho Lee
Abstract: Parallel test-time scaling, which generates multiple candidate solutions for a single problem, is a powerful technique for improving large language model performance. However, it is hindered by two key bottlenecks: accurately selecting the correct solution from the candidate pool, and the high inference latency from generating many full solutions. We argue that both challenges are fundamentally linked to verifier calibration. A well-calibrated verifier not only improves answer selection, but also enables early-stopping strategies to reduce latency. However, existing verifiers are limited as they score each candidate in isolation, overlooking rich contextual information across the set of candidates. To address this, we introduce the Multi-Sequence Verifier (MSV), the first verifier designed to jointly process all candidate solutions and model their interactions. MSV achieves improved calibration, which directly enhances best-of-N selection performance. We further introduce a streaming MSV variant that empowers a novel early-stopping framework. Our novel framework fully leverages parallel decoding, which contrasts with the existing multi-sequence early exit works that decode sequences one by one and thus incur significant latency. In this novel setting, MSV can achieve the same target accuracy with around half the latency that would be required with its counterpart that scores each solution in isolation.
Authors: Abinav Rao, Alex Wa, Rishi Athavale
Abstract: We introduce Graph Hopfield Networks, whose energy function couples associative memory retrieval with graph Laplacian smoothing for node classification. Gradient descent on this joint energy yields an iterative update interleaving Hopfield retrieval with Laplacian propagation. Memory retrieval provides regime-dependent benefits: up to 2.0~pp on sparse citation networks and up to 5 pp additional robustness under feature masking; the iterative energy-descent architecture itself is a strong inductive bias, with all variants (including the memory-disabled NoMem ablation) outperforming standard baselines on Amazon co-purchase graphs. Tuning enables graph sharpening for heterophilous benchmarks without architectural changes.
Authors: Subramanyam Sahoo, Aman Chadha, Vinija Jain, Divya Chaudhary
Abstract: Mathematical reasoning models are widely deployed in education, automated tutoring, and decision support systems despite exhibiting fundamental computational instabilities. We demonstrate that state-of-the-art models (Qwen2.5-Math-7B) achieve 61% accuracy through a mixture of reliable and unreliable reasoning pathways: 18.4% of correct predictions employ stable, faithful reasoning while 81.6% emerge through computationally inconsistent pathways. Additionally, 8.8% of all predictions are silent failures -- confident yet incorrect outputs. Through comprehensive analysis using novel faithfulness metrics, we reveal: (1) reasoning quality shows weak negative correlation with correctness (r=-0.21, p=0.002), reflecting a binary classification threshold artifact rather than a monotonic inverse relationship; (2) scaling from 1.5B to 7B parameters (4.7x increase) provides zero accuracy benefit on our evaluated subset (6% of GSM8K), requiring validation on the complete benchmark; and (3) latent reasoning employs diverse computational strategies, with ~20% sharing CoT-like patterns. These findings highlight that benchmark accuracy can mask computational unreliability, demanding evaluation reforms measuring stability beyond single-sample metrics.
Authors: Samuel Garcin, Thomas Walker, Steven McDonagh, Tim Pearce, Hakan Bilen, Tianyu He, Kaixin Wang, Jiang Bian
Abstract: Interactive world models continually generate video by responding to a user's actions, enabling open-ended generation capabilities. However, existing models typically lack a 3D representation of the environment, meaning 3D consistency must be implicitly learned from data, and spatial memory is restricted to limited temporal context windows. This results in an unrealistic user experience and presents significant obstacles to down-stream tasks such as training agents. To address this, we present PERSIST, a new paradigm of world model which simulates the evolution of a latent 3D scene: environment, camera, and renderer. This allows us to synthesize new frames with persistent spatial memory and consistent geometry. Both quantitative metrics and a qualitative user study show substantial improvements in spatial memory, 3D consistency, and long-horizon stability over existing methods, enabling coherent, evolving 3D worlds. We further demonstrate novel capabilities, including synthesising diverse 3D environments from a single image, as well as enabling fine-grained, geometry-aware control over generated experiences by supporting environment editing and specification directly in 3D space. Project page: https://francelico.github.io/persist.github.io
Authors: Jeongdong Kim, Minsu Kim, Jonggeol Na, Junghwan Kim
Abstract: E-fuels are promising long-term energy carriers supporting the net-zero transition. However, the large combinatorial design-operation spaces under renewable uncertainty make the use of mathematical programming impractical for co-optimizing e-fuel production systems. Here, we present MasCOR, a machine-learning-assisted co-optimization framework that learns from global operational trajectories. By encoding system design and renewable trends, a single MasCOR agent generalizes dynamic operation across diverse configurations and scenarios, substantially simplifying design-operation co-optimization under uncertainty. Benchmark comparisons against state-of-the-art reinforcement learning baselines demonstrate near-optimal performance, while computational costs are substantially lower than those of mathematical programming, enabling rapid parallel evaluation of designs within the co-optimization loop. This framework enables rapid screening of feasible design spaces together with corresponding operational policies. When applied to four potential European sites targeting e-methanol production, MasCOR shows that most locations benefit from reducing system load below 50 MW to achieve carbon-neutral methanol production, with production costs of 1.0-1.2 USD per kg. In contrast, Dunkirk (France), with limited renewable availability and high grid prices, favors system loads above 200 MW and expanded storage to exploit dynamic grid exchange and hydrogen sales to the market. These results underscore the value of the MasCOR framework for site-specific guidance from system design to real-time operation.
Authors: Haoran Lu, Shang Wu, Jianshu Zhang, Maojiang Su, Guo Ye, Chenwei Xu, Lie Lu, Pranav Maneriker, Fan Du, Manling Li, Zhaoran Wang, Han Liu
Abstract: Recent video diffusion models have achieved impressive capabilities as large-scale generative world models. However, these models often struggle with fine-grained physical consistency, exhibiting physically implausible dynamics over time. In this work, we present \textbf{Phys4D}, a pipeline for learning physics-consistent 4D world representations from video diffusion models. Phys4D adopts \textbf{a three-stage training paradigm} that progressively lifts appearance-driven video diffusion models into physics-consistent 4D world representations. We first bootstrap robust geometry and motion representations through large-scale pseudo-supervised pretraining, establishing a foundation for 4D scene modeling. We then perform physics-grounded supervised fine-tuning using simulation-generated data, enforcing temporally consistent 4D dynamics. Finally, we apply simulation-grounded reinforcement learning to correct residual physical violations that are difficult to capture through explicit supervision. To evaluate fine-grained physical consistency beyond appearance-based metrics, we introduce a set of \textbf{4D world consistency evaluation} that probe geometric coherence, motion stability, and long-horizon physical plausibility. Experimental results demonstrate that Phys4D substantially improves fine-grained spatiotemporal and physical consistency compared to appearance-driven baselines, while maintaining strong generative performance. Our project page is available at https://sensational-brioche-7657e7.netlify.app/
Authors: Shang Wu, Chenwei Xu, Zhuofan Xia, Weijian Li, Lie Lu, Pranav Maneriker, Fan Du, Manling Li, Han Liu
Abstract: State-of-the-art text-to-video (T2V) generators frequently violate physical laws despite high visual quality. We show this stems from insufficient physical constraints in prompts rather than model limitations: manually adding physics details reliably produces physically plausible videos, but requires expertise and does not scale. We present PhyPrompt, a two-stage reinforcement learning framework that automatically refines prompts for physically realistic generation. First, we fine-tune a large language model on a physics-focused Chain-of-Thought dataset to integrate principles like object motion and force interactions while preserving user intent. Second, we apply Group Relative Policy Optimization with a dynamic reward curriculum that initially prioritizes semantic fidelity, then progressively shifts toward physical commonsense. This curriculum achieves synergistic optimization: PhyPrompt-7B reaches 40.8\% joint success on VideoPhy2 (8.6pp gain), improving physical commonsense by 11pp (55.8\% to 66.8\%) while simultaneously increasing semantic adherence by 4.4pp (43.4\% to 47.8\%). Remarkably, our curriculum exceeds single-objective training on both metrics, demonstrating compositional prompt discovery beyond conventional multi-objective trade-offs. PhyPrompt outperforms GPT-4o (+3.8\% joint) and DeepSeek-V3 (+2.2\%, 100$\times$ larger) using only 7B parameters. The approach transfers zero-shot across diverse T2V architectures (Lavie, VideoCrafter2, CogVideoX-5B) with up to 16.8\% improvement, establishing that domain-specialized reinforcement learning with compositional curricula surpasses general-purpose scaling for physics-aware generation.
Authors: Shiza Fatimah, Aniket Sen, Sophia Falk, Florian Mai, Lucie Flek, Nicholas Kluge Corr\^ea
Abstract: The dominance of large multilingual foundation models has widened linguistic inequalities in Natural Language Processing (NLP), often leaving low-resource languages underrepresented. This paper introduces LilMoo, a 0.6-billion-parameter Hindi language model trained entirely from scratch to address this gap. Unlike prior Hindi models that rely on continual pretraining from opaque multilingual foundations, LilMoo is developed through a fully transparent and reproducible pipeline optimized for limited compute environments. We construct a high-quality Hindi corpus (GigaLekh) filtered through both heuristic and learned (LLM-as-a-judge) methods, complemented by bilingual augmentation with curated English data. Using this dataset, we explore various training recipes for small-scale language models. Across comprehensive evaluation suites, LilMoo consistently outperforms comparably sized multilingual baselines such as Qwen2.5-0.5B and Qwen3-0.6B, demonstrating that well-designed language-specific pretraining can rival large multilingual models at the sub-billion-parameter range.
Authors: Danielle S. Fox, Brenda L. Robles, Elizabeth DiPietro Brovey, Christian D. Schunn
Abstract: Teachers face increasing demands on their time, particularly in adapting mathematics curricula to meet individual student needs while maintaining cognitive rigor. This study evaluates whether AI tools can accurately classify the cognitive demand of mathematical tasks, which is important for creating or adapting tasks that support student learning. We tested eleven AI tools: six general-purpose (ChatGPT, Claude, DeepSeek, Gemini, Grok, Perplexity) and five education-specific (Brisk, Coteach AI, Khanmigo, Magic School, School.AI), on their ability to categorize mathematics tasks across four levels of cognitive demand using a research-based framework. The goal was to approximate the performance teachers will achieve with straightforward prompts. On average, AI tools accurately classified cognitive demand in only 63% of cases. Education-specific tools were not more accurate than general-purpose tools, and no tool exceeded 83% accuracy. All tools struggled with tasks at the extremes of cognitive demand (Memorization and Doing Mathematics), exhibiting a systematic bias toward middle-category levels (Procedures with/without Connections). The tools often gave plausible-sounding explanations likely to be persuasive to novice teachers. Error analysis of AI tools' misclassification of the broad level of cognitive demand (high vs. low) revealed that tools consistently overweighted surface textual features over underlying cognitive processes. Further, AI tools showed weaknesses in reasoning about factors that make tasks higher vs. lower cognitive demand. Errors stemmed not from ignoring relevant dimensions, but from incorrectly reasoning about multiple task aspects. These findings carry implications for AI integration into teacher planning workflows and highlight the need for improved prompt engineering and tool development for educational applications.
Authors: Subramanyam Sahoo
Abstract: Agentic AI systems - capable of goal interpretation, world modeling, planning, tool use, long-horizon operation, and autonomous coordination - introduce distinct control failures not addressed by existing safety frameworks. We identify six agentic governance failures tied to these capabilities and show how they erode meaningful human control in military settings. We propose the Agentic Military AI Governance Framework (AMAGF), a measurable architecture structured around three pillars: Preventive Governance (reducing failure likelihood), Detective Governance (real-time detection of control degradation), and Corrective Governance (restoring or safely degrading operations). Its core mechanism, the Control Quality Score (CQS), is a composite real-time metric quantifying human control and enabling graduated responses as control weakens. For each failure type, we define concrete mechanisms, assign responsibilities across five institutional actors, and formalize evaluation metrics. A worked operational scenario illustrates implementation, and we situate the framework within established agent safety literature. We argue that governance must move from a binary conception of control to a continuous model in which control quality is actively measured and managed throughout the operational lifecycle.
Authors: Maksim Kuznetsov, Zulfat Miftahutdinov, Rim Shayakhmetov, Mikolaj Mizera, Roman Schutski, Bogdan Zagribelnyy, Ivan Ilin, Nikita Bondarev, Thomas MacDougall, Mathieu Reymond, Mihir Bafna, Kaeli Kaymak-Loveless, Eugene Babin, Maxim Malkov, Mathias Lechner, Ramin Hasani, Alexander Amini, Vladimir Aladinskiy, Alex Aliper, Alex Zhavoronkov
Abstract: General-purpose large language models (LLMs) that rely on in-context learning do not reliably deliver the scientific understanding and performance required for drug discovery tasks. Simply increasing model size or introducing reasoning tokens does not yield significant performance gains. To address this gap, we introduce the MMAI Gym for Science, a one-stop shop molecular data formats and modalities as well as task-specific reasoning, training, and benchmarking recipes designed to teach foundation models the 'language of molecules' in order to solve practical drug discovery problems. We use MMAI Gym to train an efficient Liquid Foundation Model (LFM) for these applications, demonstrating that smaller, purpose-trained foundation models can outperform substantially larger general-purpose or specialist models on molecular benchmarks. Across essential drug discovery tasks - including molecular optimization, ADMET property prediction, retrosynthesis, drug-target activity prediction, and functional group reasoning - the resulting model achieves near specialist-level performance and, in the majority of settings, surpasses larger models, while remaining more efficient and broadly applicable in the domain.
Authors: Zeyneb N. Kaya, Nick Rui
Abstract: As strong general reasoners, large language models (LLMs) encounter diverse domains and tasks, where the ability to adapt and self-improve at test time is valuable. We introduce MASS, a meta-learning framework that enables LLMs to self-adapt by generating problem-specific synthetic training data and performing targeted self-updates optimized for downstream performance at inference time. We train this behavior end-to-end via bilevel optimization: an inner loop adapts on self-generated examples while an outer loop meta-learns data-attribution signals and rewards post-update task performance. The synthetic data is optimized with scalable meta-gradients, backpropagating the downstream loss through the inner updates to reward useful generations. Experiments on mathematical reasoning show that MASS learns to synthesize per-instance curricula that yield effective, data-efficient test-time adaptation.
Authors: Maarten C. Vonk, Anna V. Kononova, Thomas B\"ack, Tim Sweijs
Abstract: Western governments have adopted an assortment of counter-hybrid threat measures to defend against hostile actions below the conventional military threshold. The impact of these measures is unclear because of the ambiguity of hybrid threats, their cross-domain nature, and uncertainty about how countermeasures shape adversarial behavior. This paper offers a novel approach to clarifying this impact by unifying previously bifurcating hybrid threat modeling methods through a (multi-agent) influence diagram framework. The model balances the costs of countermeasures, their ability to dissuade the adversary from executing hybrid threats, and their potential to mitigate the impact of hybrid threats. We run 1000 semi-synthetic variants of a real-world-inspired scenario simulating the strategic interaction between attacking agent A and defending agent B over a cyber attack on critical infrastructure to explore the effectiveness of a set of five different counter-hybrid threat measures. Counter-hybrid measures range from strengthening resilience and denial of the adversary's ability to execute a hybrid threat to dissuasion through the threat of punishment. Our analysis primarily evaluates the overarching characteristics of counter-hybrid threat measures. This approach allows us to generalize the effectiveness of these measures and examine parameter impact sensitivity. In addition, we discuss policy relevance and outline future research avenues.
Authors: Jiahao Qin
Abstract: We introduce mlx-snn, the first spiking neural network (SNN) library built natively on Apple's MLX framework. As SNN research grows rapidly, all major libraries -- snnTorch, Norse, SpikingJelly, Lava -- target PyTorch or custom backends, leaving Apple Silicon users without a native option. mlx-snn provides six neuron models (LIF, IF, Izhikevich, Adaptive LIF, Synaptic, Alpha), four surrogate gradient functions, four spike encoding methods (including an EEG-specific encoder), and a complete backpropagation-through-time training pipeline. The library leverages MLX's unified memory architecture, lazy evaluation, and composable function transforms (mx.grad, mx.compile) to enable efficient SNN research on Apple Silicon hardware. We validate mlx-snn on MNIST digit classification across five hyperparameter configurations and three backends, achieving up to 97.28% accuracy with 2.0--2.5 times faster training and 3--10 times lower GPU memory than snnTorch on the same M3 Max hardware. mlx-snn is open-source under the MIT license and available on PyPI. https://github.com/D-ST-Sword/mlx-snn
Authors: Achleshwar Luthra, Yash Salunkhe, Tomer Galanti
Abstract: Frozen self-supervised representations often transfer well with only a few labels across many semantic tasks. We argue that a single geometric quantity, \emph{directional} CDNV (decision-axis variance), sits at the core of two favorable behaviors: strong few-shot transfer within a task, and low interference across many tasks. We show that both emerge when variability \emph{along} class-separating directions is small. First, we prove sharp non-asymptotic multiclass generalization bounds for downstream classification whose leading term is the directional CDNV. The bounds include finite-shot corrections that cleanly separate intrinsic decision-axis variability from centroid-estimation error. Second, we link decision-axis collapse to multitask geometry: for independent balanced labelings, small directional CDNV across tasks forces the corresponding decision axes to be nearly orthogonal, helping a single representation support many tasks with minimal interference. Empirically, across SSL objectives, directional CDNV collapses during pretraining even when classical CDNV remains large, and our bounds closely track few-shot error at practical shot sizes. Additionally, on synthetic multitask data, we verify that SSL learns representations whose induced decision axes are nearly orthogonal. The code and project page of the paper are available at [\href{https://dlfundamentals.github.io/directional-neural-collapse/}{project page}].
URLs: https://dlfundamentals.github.io/directional-neural-collapse/
Authors: Yiming Sun, Runlong Yu, Rongchao Dong, Shuo Chen, Licheng Liu, Youmi Oh, Qianlai Zhuang, Yiqun Xie, Xiaowei Jia
Abstract: Accurate prediction of terrestrial ecosystem carbon fluxes (e.g., CO$_2$, GPP, and CH$_4$) is essential for understanding the global carbon cycle and managing its impacts. However, prediction remains challenging due to strong spatiotemporal heterogeneity: ecosystem flux responses are constrained by slowly varying regime conditions, while short-term fluctuations are driven by high-frequency dynamic forcings. Most existing learning-based approaches treat environmental covariates as a homogeneous input space, implicitly assuming a global response function, which leads to brittle generalization across heterogeneous ecosystems. In this work, we propose Role-Aware Conditional Inference (RACI), a process-informed learning framework that formulates ecosystem flux prediction as a conditional inference problem. RACI employs hierarchical temporal encoding to disentangle slow regime conditioners from fast dynamic drivers, and incorporates role-aware spatial retrieval that supplies functionally similar and geographically local context for each role. By explicitly modeling these distinct functional roles, RACI enables a model to adapt its predictions across diverse environmental regimes without training separate local models or relying on fixed spatial structures. We evaluate RACI across multiple ecosystem types (wetlands and agricultural systems), carbon fluxes (CO$_2$, GPP, CH$_4$), and data sources, including both process-based simulations and observational measurements. Across all settings, RACI consistently outperforms competitive spatiotemporal baselines, demonstrating improved accuracy and spatial generalization under pronounced environmental heterogeneity.
Authors: Haochang Hao, Yifan Xu, Xinzhuo Li, Yingqiang Ge, Lu Cheng
Abstract: Current LLM-based conversational recommender systems (CRS) primarily optimize recommendation accuracy and user satisfaction. We identify an underexplored vulnerability in which recommendation outputs may negatively impact users by violating personalized safety constraints, when individualized safety sensitivities -- such as trauma triggers, self-harm history, or phobias -- are implicitly inferred from the conversation but not respected during recommendation. We formalize this challenge as personalized CRS safety and introduce SafeRec, a new benchmark dataset designed to systematically evaluate safety risks in LLM-based CRS under user-specific constraints. To further address this problem, we propose SafeCRS, a safety-aware training framework that integrates Safe Supervised Fine-Tuning (Safe-SFT) with Safe Group reward-Decoupled Normalization Policy Optimization (Safe-GDPO) to jointly optimize recommendation quality and personalized safety alignment. Extensive experiments on SafeRec demonstrate that SafeCRS reduces safety violation rates by up to 96.5% relative to the strongest recommendation-quality baseline while maintaining competitive recommendation quality. Warning: This paper contains potentially harmful and offensive content.
Authors: Aswini Sivakumar, Vijayan Sugumaran, Yao Qiang
Abstract: Automated question-answering (QA) systems increasingly rely on retrieval-augmented generation (RAG) to ground large language models (LLMs) in authoritative medical knowledge, ensuring clinical accuracy and patient safety in Artificial Intelligence (AI) applications for healthcare. Despite progress in RAG evaluation, current benchmarks focus only on simple multiple-choice QA tasks and employ metrics that poorly capture the semantic precision required for complex QA tasks. These approaches fail to diagnose whether an error stems from faulty retrieval or flawed generation, limiting developers from performing targeted improvement. To address this gap, we propose RAG-X, a diagnostic framework that evaluates the retriever and generator independently across a triad of QA tasks: information extraction, short-answer generation, and multiple-choice question (MCQ) answering. RAG-X introduces Context Utilization Efficiency (CUE) metrics to disaggregate system success into interpretable quadrants, isolating verified grounding from deceptive accuracy. Our experiments reveal an ``Accuracy Fallacy", where a 14\% gap separates perceived system success from evidence-based grounding. By surfacing hidden failure modes, RAG-X offers the diagnostic transparency needed for safe and verifiable clinical RAG systems.
Authors: Nicholas Kluge Corr\^ea, Aniket Sen, Shiza Fatimah, Sophia Falk, Lennard Landgraf, Julia Kastner, Lucie Flek
Abstract: We present Tucano 2, a fully open suite of large language models (LLMs) with 0.5-3.7 billion parameters, designed to address certain gaps in open-source development for Portuguese LLMs. Following our previous works, we now extend our dataset, GigaVerbo-v2, to a new degree of quality and scale, while also introducing a new synthetic dataset, GigaVerbo-v2 Synth, aimed at filling missing gaps in GigaVerbo-v2, and two post-training datasets, GigaVerbo-v2 SFT and GigaVerbo-v2 Preferences, that allow Portuguese LLMs to be trained in domains like retrieval augmented generation, coding, tool use, chain-of-thought reasoning, and many other domains of interest. Through extensive ablation studies, we design both pretraining and continual pretraining recipes for the Tucano 2 suite (Base, Instruct, and Think), which achieve state-of-the-art performance on several Portuguese-language modeling benchmarks. We also extend and refine the evaluation harness introduced in our earlier work, yielding a comprehensive evaluation suite that provides strong signals across different pretraining, continual pretraining, and post-training regimes. All artifacts associated with Tucano 2 are openly released, including training recipes, logs, and source code, ensuring that our work is reproducible, accessible, and extendable by the broader Portuguese NLP community.
Authors: Brandon Yee, Krishna Sharma
Abstract: MoltBook is a large-scale multi-agent coordination environment where over 770,000 autonomous LLM agents interact without human participation, offering the first opportunity we are aware of to observe emergent multi-agent coordination dynamics at this population scale. We introduce \textit{Molt Dynamics}: the emergent agent coordination behaviors, inter-agent communication dynamics, and role specialization patterns arising when autonomous agents operate as decentralized decision-makers in an unconstrained multi-agent environment. Through longitudinal observation of 90,704 active agents over three weeks, we characterize three aspects. First, spontaneous role specialization: network-based clustering reveals six structural roles (silhouette 0.91), though the result primarily reflects core-periphery organization -- 93.5\% of agents occupy a homogeneous peripheral cluster, with meaningful differentiation confined to the active minority. Second, decentralized information dissemination: cascade analysis of 10,323 inter-agent propagation events reveals power-law distributed cascade sizes ($\alpha = 2.57 \pm 0.02$) and saturating adoption dynamics where adoption probability shows diminishing returns with repeated exposures (Cox hazard ratio 0.53, concordance 0.78). Third, distributed cooperative task resolution: 164 multi-agent collaborative events show detectable coordination patterns, but success rates are low (6.7\%, $p = 0.057$) and cooperative outcomes are significantly worse than a matched single-agent baseline (Cohen's $d = -0.88$), indicating emergent cooperative behavior is nascent. These findings establish an empirical baseline for coordination dynamics in decentralized autonomous agent systems, with implications for multi-agent system design, agent communication protocol engineering, and AI safety.
Authors: Angana Borah, Zohaib Khan, Rada Mihalcea, Ver\'onica P\'erez-Rosas
Abstract: Misinformation is a growing societal threat, and susceptibility to misinformative claims varies across demographic groups due to differences in underlying beliefs. As Large Language Models (LLMs) are increasingly used to simulate human behaviors, we investigate whether they can simulate demographic misinformation susceptibility, treating beliefs as a primary driving factor. We introduce BeliefSim, a simulation framework that constructs demographic belief profiles using psychology-informed taxonomies and survey priors. We study prompt-based conditioning and post-training adaptation, and conduct a multi-fold evaluation using: (i) susceptibility accuracy and (ii) counterfactual demographic sensitivity. Across both datasets and modeling strategies, we show that beliefs provide a strong prior for simulating misinformation susceptibility, with accuracy up to 92%.
Authors: Oishik Chowdhury, Anushka Debnath, Bastin Tony Roy Savarimuthu
Abstract: In Multi-Agent Systems (MAS), agents are designed with social capabilities, allowing them to understand and reason about social concepts such as norms when interacting with others (e.g., inter-robot interactions). In Normative MAS (NorMAS), researchers study how norms develop, and how violations are detected and sanctioned. However, existing research in NorMAS use symbolic approaches (e.g., formal logic) for norm representation and reasoning whose application is limited to simplified environments. In contrast, Multimodal Large Language Models (MLLMs) present promising possibilities to develop software used by robots to identify and reason about norms in a wide variety of complex social situations embodied in text and images. However, prior work on norm reasoning have been limited to text-based scenarios. This paper investigates the norm reasoning competence of five MLLMs by evaluating their ability to answer norm-related questions based on thirty text-based and thirty image-based stories, and comparing their responses against humans. Our results show that MLLMs demonstrate superior performance in norm reasoning in text than in images. GPT-4o performs the best in both modalities offering the most promise for integration with MAS, followed by the free model Qwen-2.5VL. Additionally, all models find reasoning about complex norms challenging.
Authors: Neha Nagaraja, Hayretdin Bahsi
Abstract: While incorporating LLMs into systems offers significant benefits in critical application areas such as healthcare, new security challenges emerge due to the potential cyber kill chain cycles that combine adversarial model, prompt injection and conventional cyber attacks. Threat modeling methods enable the system designers to identify potential cyber threats and the relevant mitigations during the early stages of development. Although the cyber security community has extensive experience in applying these methods to software-based systems, the elicited threats are usually abstract and vague, limiting their effectiveness for conducting proper likelihood and impact assessments for risk prioritization, especially in complex systems with novel attacks surfaces, such as those involving LLMs. In this study, we propose a structured, goal driven risk assessment approach that contextualizes the threats with detailed attack vectors, preconditions, and attack paths through the use of attack trees. We demonstrate the proposed approach on a case study with an LLM agent-based healthcare system. This study harmonizes the state-of-the-art attacks to LLMs with conventional ones and presents possible attack paths applicable to similar systems. By providing a structured risk assessment, this study makes a significant contribution to the literature and advances the secure-by-design practices in LLM-based systems.
Authors: Neha Nagaraja, Lan Zhang, Zhilong Wang, Bo Zhang, Pawan Patil
Abstract: Multimodal Large Language Models (MLLMs) integrate vision and text to power applications, but this integration introduces new vulnerabilities. We study Image-based Prompt Injection (IPI), a black-box attack in which adversarial instructions are embedded into natural images to override model behavior. Our end-to-end IPI pipeline incorporates segmentation-based region selection, adaptive font scaling, and background-aware rendering to conceal prompts from human perception while preserving model interpretability. Using the COCO dataset and GPT-4-turbo, we evaluate 12 adversarial prompt strategies and multiple embedding configurations. The results show that IPI can reliably manipulate the output of the model, with the most effective configuration achieving up to 64\% attack success under stealth constraints. These findings highlight IPI as a practical threat in black-box settings and underscore the need for defenses against multimodal prompt injection.
Authors: Daijin Yang, Erica Kleinman, Casper Harteveld
Abstract: Educational games can foster critical thinking, problem-solving, and motivation, yet instructors often find it difficult to design games that reliably achieve specific learning outcomes. Existing authoring environments reduce the need for programming expertise, but they do not eliminate the underlying challenges of educational game design, and they can leave non-expert designers reliant on opaque suggestions from AI systems. We designed a controlled natural language framework-based web tool that positions language as the primary interface for LLM-assisted educational game design. In the tool, users and an LLM assistant collaboratively develop a structured language that maps pedagogy to gameplay through four linked components. We argue that, by making pedagogical intent explicit and editable in the interface, the tool has the potential to lower design barriers for non-expert designers, preserves human agency in critical decisions, and enables alignment and reflections between pedagogy and gameplay during and after co-creation.
Authors: Haohang Huang
Abstract: Construction aggregates, including sand and gravel, crushed stone and riprap, are the core building blocks of the construction industry. State-of-the-practice characterization methods mainly relies on visual inspection and manual measurement. State-of-the-art aggregate imaging methods have limitations that are only applicable to regular-sized aggregates under well-controlled conditions. This dissertation addresses these major challenges by developing a field imaging framework for the morphological characterization of aggregates as a multi-scenario solution. For individual and non-overlapping aggregates, a field imaging system was designed and the associated segmentation and volume estimation algorithms were developed. For 2D image analyses of aggregates in stockpiles, an automated 2D instance segmentation and morphological analysis approach was established. For 3D point cloud analyses of aggregate stockpiles, an integrated 3D Reconstruction-Segmentation-Completion (RSC-3D) approach was established: 3D reconstruction procedures from multi-view images, 3D stockpile instance segmentation, and 3D shape completion to predict the unseen sides. First, a 3D reconstruction procedure was developed to obtain high-fidelity 3D models of collected aggregate samples, based on which a 3D aggregate particle library was constructed. Next, two datasets were derived from the 3D particle library for 3D learning: a synthetic dataset of aggregate stockpiles with ground-truth instance labels, and a dataset of partial-complete shape pairs, developed with varying-view raycasting schemes. A state-of-the-art 3D instance segmentation network and a 3D shape completion network were trained on the datasets, respectively. The application of the integrated approach was demonstrated on real stockpiles and validated with ground-truth, showing good performance in capturing and predicting the unseen sides of aggregates.
Authors: Zhiqiang Sheng, Xumeng Han, Zhiwei Zhang, Zenghui Xiong, Yifan Ding, Aoxiang Ping, Xiang Li, Tong Guo, Yao Mao
Abstract: Multimodal generative models have made significant strides in image editing, demonstrating impressive performance on a variety of static tasks. However, their proficiency typically does not extend to complex scenarios requiring dynamic reasoning, leaving them ill-equipped to model the coherent, intermediate logical pathways that constitute a multi-step evolution from an initial state to a final one. This capacity is crucial for unlocking a deeper level of procedural and causal understanding in visual manipulation. To systematically measure this critical limitation, we introduce InEdit-Bench, the first evaluation benchmark dedicated to reasoning over intermediate pathways in image editing. InEdit-Bench comprises meticulously annotated test cases covering four fundamental task categories: state transition, dynamic process, temporal sequence, and scientific simulation. Additionally, to enable fine-grained evaluation, we propose a set of assessment criteria to evaluate the logical coherence and visual naturalness of the generated pathways, as well as the model's fidelity to specified path constraints. Our comprehensive evaluation of 14 representative image editing models on InEdit-Bench reveals significant and widespread shortcomings in this domain. By providing a standardized and challenging benchmark, we aim for InEdit-Bench to catalyze research and steer development towards more dynamic, reason-aware, and intelligent multimodal generative models.
Authors: Jiaqi Lv, Qingfeng Du, Yu Zhang, Yongqi Han, Sheng Li
Abstract: Graph Neural Networks (GNNs) have emerged as a powerful framework for processing graph-structured data. However, conventional GNNs and their variants are inherently limited by the homophily assumption, leading to degradation in performance on heterophilic graphs. Although substantial efforts have been made to mitigate this issue, they remain constrained by the message-passing paradigm, which is inherently rooted in homophily. In this paper, a detailed analysis of how the underlying label autocorrelation of the homophily assumption introduces bias into GNNs is presented. We innovatively leverage a negative feedback mechanism to correct the bias and propose Graph Negative Feedback Bias Correction (GNFBC), a simple yet effective framework that is independent of any specific aggregation strategy. Specifically, we introduce a negative feedback loss that penalizes the sensitivity of predictions to label autocorrelation. Furthermore, we incorporate the output of graph-agnostic models as a feedback term, leveraging independent node feature information to counteract correlation-induced bias guided by Dirichlet energy. GNFBC can be seamlessly integrated into existing GNN architectures, improving overall performance with comparable computational and memory overhead.
Authors: Xuan Yang, Hsi-Wen Chen, Ming-Syan Chen, Jian Pei
Abstract: The Shapley value provides a principled foundation for data valuation, but exact computation is #P-hard due to the exponential coalition space. Existing accelerations remain global and ignore a structural property of modern predictors: for a given test instance, only a small subset of training points influences the prediction. We formalize this model-induced locality through support sets defined by the model's computational pathway (e.g., neighbors in KNN, leaves in trees, receptive fields in GNNs), showing that Shapley computation can be projected onto these supports without loss when locality is exact. This reframes Shapley evaluation as a structured data processing problem over overlapping support-induced subset families rather than exhaustive coalition enumeration. We prove that the intrinsic complexity of Local Shapley is governed by the number of distinct influential subsets, establishing an information-theoretic lower bound on retraining operations. Guided by this result, we propose LSMR (Local Shapley via Model Reuse), an optimal subset-centric algorithm that trains each influential subset exactly once via support mapping and pivot scheduling. For larger supports, we develop LSMR-A, a reuse-aware Monte Carlo estimator that remains unbiased with exponential concentration, with runtime determined by the number of distinct sampled subsets rather than total draws. Experiments across multiple model families demonstrate substantial retraining reductions and speedups while preserving high valuation fidelity.
Authors: Guoyi Li, Shihao Xu, Jiatong Ma, Yunyun Han, Jianhua Chen, Yafeng Deng
Abstract: Large language models (LLMs) have advanced medical dialogue systems, yet psychiatric consultation poses substantially higher demands due to subjective ambiguity and comorbidity complexity: an agent must continuously extract psychopathological cues from incomplete and inconsistent patient reports in multi-turn interactions and perform rigorous differential diagnostic reasoning. However, existing methods face two fundamental challenges. First, without criteria-grounded clinical supports, they are prone to unsupported clinical assertions when symptoms are atypical or underspecified. Second, in multi-turn interactions, they struggle to mitigate inquiry drift (off-topic or low-yield questioning) and optimize questioning strategies. To address these challenges, we propose MIND, a unified inquiry--diagnosis reinforcement learning framework for psychiatric consultation. Specifically, we build a Criteria-Grounded Psychiatric Reasoning Bank (PRB) that summarizes dialogue context into clinical retrieval states, retrieves semantically similar reference consultations, and distills reusable criteria-grounded clinical supports to guide criteria-aligned inquiry and reasoning. Building on this foundation, MIND enforces explicit clinical reasoning with rubric-based process rewards to provide fine-grained supervision over intermediate decision steps, and incorporates a value-aware trajectory rectification mechanism to jointly improve information acquisition and diagnostic decision-making across turns. Extensive experiments demonstrate that MIND consistently outperforms strong baselines in diagnostic accuracy, empathetic interaction quality, interpretability, and generalization.
Authors: Yuhao Chen, Bin Shan, Xin Ye, Cheng Chen
Abstract: Multimodal Large Language Models (MLLMs) have shown strong performance in vision-language tasks, but their inference efficiency is severely limited by the exponential growth of visual tokens in complex scenarios such as high-resolution images and videos. Existing visual token pruning methods mainly operate after visual encoding, overlooking the substantial computational cost incurred during the encoding stage. To address this issue, we propose EvoPrune, an early-stage visual token pruning method for MLLMs that performs pruning directly during visual encoding. Specifically, EvoPrune employs a layer-wise pruning strategy guided by token similarity, diversity, and attention-based importance to retain the most informative visual tokens at selected encoding layers. Extensive experiments on image and video benchmarks validate the effectiveness of EvoPrune. In particular, on the VideoMME dataset, EvoPrune achieves 2$\times$ inference speedup with less than 1% performance degradation, demonstrating its potential for latency-sensitive MLLM deployment.
Authors: Jeremy Avigad
Abstract: Recent developments show that AI can prove research-level theorems in mathematics, both formally and informally. This essay urges mathematicians to stay up-to-date with the technology, to consider the ways it will disrupt mathematical practice, and to respond appropriately to the challenges and opportunities we now face.
Authors: Inho Kong, Sojin Lee, Youngjoon Hong, Hyunwoo J. Kim
Abstract: Classifier-Free Guidance (CFG) has established the foundation for guidance mechanisms in diffusion models, showing that well-designed guidance proxies significantly improve conditional generation and sample quality. Autoguidance (AG) has extended this idea, but it relies on an auxiliary network and leaves solver-induced errors unaddressed. In stiff regions, the ODE trajectory changes sharply, where local truncation error (LTE) becomes a critical factor that deteriorates sample quality. Our key observation is that these errors align with the dominant eigenvector, motivating us to leverage the solver-induced error as a guidance signal. We propose Embedded Runge-Kutta Guidance (ERK-Guid), which exploits detected stiffness to reduce LTE and stabilize sampling. We theoretically and empirically analyze stiffness and eigenvector estimators with solver errors to motivate the design of ERK-Guid. Our experiments on both synthetic datasets and the popular benchmark dataset, ImageNet, demonstrate that ERK-Guid consistently outperforms state-of-the-art methods. Code is available at https://github.com/mlvlab/ERK-Guid.
Authors: Saptarshi Chakraborty, Quentin Berthet, Peter L. Bartlett
Abstract: Despite the remarkable empirical success of score-based diffusion models, their statistical guarantees remain underdeveloped. Existing analyses often provide pessimistic convergence rates that do not reflect the intrinsic low-dimensional structure common in real data, such as that arising in natural images. In this work, we study the statistical convergence of score-based diffusion models for learning an unknown distribution $\mu$ from finitely many samples. Under mild regularity conditions on the forward diffusion process and the data distribution, we derive finite-sample error bounds on the learned generative distribution, measured in the Wasserstein-$p$ distance. Unlike prior results, our guarantees hold for all $p \ge 1$ and require only a finite-moment assumption on $\mu$, without compact-support, manifold, or smooth-density conditions. Specifically, given $n$ i.i.d.\ samples from $\mu$ with finite $q$-th moment and appropriately chosen network architectures, hyperparameters, and discretization schemes, we show that the expected Wasserstein-$p$ error between the learned distribution $\hat{\mu}$ and $\mu$ scales as $\mathbb{E}\, \mathbb{W}_p(\hat{\mu},\mu) = \widetilde{O}\!\left(n^{-1 / d^\ast_{p,q}(\mu)}\right),$ where $d^\ast_{p,q}(\mu)$ is the $(p,q)$-Wasserstein dimension of $\mu$. Our results demonstrate that diffusion models naturally adapt to the intrinsic geometry of data and mitigate the curse of dimensionality, since the convergence rate depends on $d^\ast_{p,q}(\mu)$ rather than the ambient dimension. Moreover, our theory conceptually bridges the analysis of diffusion models with that of GANs and the sharp minimax rates established in optimal transport. The proposed $(p,q)$-Wasserstein dimension also extends classical Wasserstein dimension notions to distributions with unbounded support, which may be of independent theoretical interest.
Authors: Tonmoy Dey, Lin Jiang, Zheng Dong, Guang Wang
Abstract: In the vision of smart cities, technologies are being developed to enhance the efficiency of urban services and improve residents' quality of life. However, most existing research focuses on optimizing individual services in isolation, without adequately considering reciprocal interactions among heterogeneous urban services that could yield higher efficiency and improved resource utilization. For example, human couriers could collect traffic and air quality data along their delivery routes, while sensing robots could assist with on-demand delivery during peak hours, enhancing both sensing coverage and delivery efficiency. However, the joint optimization of different urban services is challenging due to potentially conflicting objectives and the need for real-time coordination in dynamic environments. In this paper, we propose UrbanHuRo, a two-layer human-robot collaboration framework for joint optimization of heterogeneous urban services, demonstrated through crowdsourced delivery and urban sensing. UrbanHuRo includes two key designs: (i) a scalable distributed MapReduce-based K-submodular maximization module for efficient order dispatch, and (ii) a deep submodular reward reinforcement learning algorithm for sensing route planning. Experimental evaluations on real-world datasets from a food delivery platform demonstrate that UrbanHuRo improves sensing coverage by 29.7% and courier income by 39.2% on average in most settings, while also significantly reducing the number of overdue orders.
Authors: Yoonwoo Kim, Raghav Arora, Roberto Mart\'in-Mart\'in, Peter Stone, Ben Abbatematteo, Yoonchang Sung
Abstract: Robot planning in partially observable environments, where not all objects are known or visible, is a challenging problem, as it requires reasoning under uncertainty through partially observable Markov decision processes. During the execution of a computed plan, a robot may unexpectedly observe task-irrelevant objects, which are typically ignored by naive planners. In this work, we propose incorporating two types of common-sense knowledge: (1) certain objects are more likely to be found in specific locations; and (2) similar objects are likely to be co-located, while dissimilar objects are less likely to be found together. Manually engineering such knowledge is complex, so we explore leveraging the powerful common-sense reasoning capabilities of large language models (LLMs). Our planning and execution framework, CoCo-TAMP, introduces a hierarchical state estimation that uses LLM-guided information to shape the belief over task-relevant objects, enabling efficient solutions to long-horizon task and motion planning problems. In experiments, CoCo-TAMP achieves an average reduction of 62.7 in planning and execution time in simulation, and 72.6 in real-world demonstrations, compared to a baseline that does not incorporate either type of common-sense knowledge.
Authors: Seunghoi Kim, Chen Jin, Henry F. J. Tregidgo, Matteo Figini, Daniel C. Alexander
Abstract: Zero-shot MRI reconstruction relies on generative priors, but single-modality unconditional priors produce hallucinations under severe ill-posedness. In many clinical workflows, complementary MRI acquisitions (e.g. high-quality structural scans) are routinely available, yet existing reconstruction methods lack mechanisms to leverage this additional information. We propose MPFlow, a zero-shot multi-modal reconstruction framework built on rectified flow that incorporates auxiliary MRI modalities at inference time without retraining the generative prior to improve anatomical fidelity. Cross-modal guidance is enabled by our proposed self-supervised pretraining strategy, Patch-level Multi-modal MR Image Pretraining (PAMRI), which learns shared representations across modalities. Sampling is jointly guided by data consistency and cross-modal feature alignment using pre-trained PAMRI, systematically suppressing intrinsic and extrinsic hallucinations. Extensive experiments on HCP and BraTS show that MPFlow matches diffusion baselines on image quality using only 20% of sampling steps while reducing tumor hallucinations by more than 15% (segmentation dice score). This demonstrates that cross-modal guidance enables more reliable and efficient zero-shot MRI reconstruction.
Authors: Yongkang Zhang, Zonglin Zhao, Yuechen Zhang, Fei Ding, Pei Li, Wenxuan Wang
Abstract: We study a systematic bias in modern image generation models: the mention order of entities in text spuriously determines spatial layout and entity--role binding. We term this phenomenon Order-to-Space Bias (OTS) and show that it arises in both text-to-image and image-to-image generation, often overriding grounded cues and causing incorrect layouts or swapped assignments. To quantify OTS, we introduce OTS-Bench, which isolates order effects with paired prompts differing only in entity order and evaluates models along two dimensions: homogenization and correctness. Experiments show that Order-to-Space Bias (OTS) is widespread in modern image generation models, and provide evidence that it is primarily data-driven and manifests during the early stages of layout formation. Motivated by this insight, we show that both targeted fine-tuning and early-stage intervention strategies can substantially reduce OTS, while preserving generation quality.
Authors: Yifan Zhu, Yibo Miao, Yinpeng Dong, Xiao-Shan Gao
Abstract: The volume of freely scraped data on the Internet has driven the tremendous success of deep learning. Along with this comes the growing concern about data privacy and security. Numerous methods for generating unlearnable examples have been proposed to prevent data from being illicitly learned by unauthorized deep models by impeding generalization. However, the existing approaches primarily rely on empirical heuristics, making it challenging to enhance unlearnable examples with solid explanations. In this paper, we analyze and improve unlearnable examples from a novel perspective: mutual information reduction. We demonstrate that effective unlearnable examples always decrease mutual information between clean features and poisoned features, and when the network gets deeper, the unlearnability goes better together with lower mutual information. Further, we prove from a covariance reduction perspective that minimizing the conditional covariance of intra-class poisoned features reduces the mutual information between distributions. Based on the theoretical results, we propose a novel unlearnable method called Mutual Information Unlearnable Examples (MI-UE) that reduces covariance by maximizing the cosine similarity among intra-class features, thus impeding the generalization effectively. Extensive experiments demonstrate that our approach significantly outperforms the previous methods, even under defense mechanisms.
Authors: John Driscoll, Yulin Chen, Viki Shi, Izak Vucharatavintara, Yaxing Yao, Haojian Jin
Abstract: This paper studies how parents want to moderate children's interactions with Generative AI chatbots, with the goal of informing the design of future GenAI parental control tools. We first used an LLM to generate synthetic child-GenAI chatbot interaction scenarios and worked with four parents to validate their realism. From this dataset, we carefully selected 12 diverse examples that evoked varying levels of concern and were rated the most realistic. Each example included a prompt and a GenAI chatbot response. We presented these to parents (N=24) and asked whether they found them concerning, why, and how they would prefer the responses to be modified and communicated. Our findings reveal three key insights: (1) parents express concern about interactions that current GenAI chatbot parental controls neglect; (2) parents want fine-grained transparency and moderation at the conversation level; and (3) parents need personalized controls that adapt to their desired strategies and children's ages.
Authors: Zehua Fan, Wenqi Lyu, Wenxuan Song, Linge Zhao, Yifei Yang, Xi Wang, Junjie He, Lida Huang, Haiyan Liu, Bingchuan Sun, Guangjun Bao, Xuanyao Mao, Liang Xu, Yan Wang, Feng Gao
Abstract: Multimodal large language models (MLLMs) have advanced zero-shot end-to-end Vision-Language Navigation (VLN), yet robust navigation requires not only semantic understanding but also predictive modeling of environment dynamics and spatial structure. We propose PROSPECT, a unified streaming navigation agent that couples a streaming Vision-Language-Action (VLA) policy with latent predictive representation learning. PROSPECT uses CUT3R as a streaming 3D foundation spatial encoder to produce long-context, absolute-scale spatial features, and fuses them with SigLIP semantic features via cross-attention. During training, we introduce learnable stream query tokens that query the streaming context and predict next-step 2D and 3D latent features (rather than pixels or explicit modalities), supervised in the latent spaces of frozen SigLIP and CUT3R teachers. The predictive branch shapes internal representations without inference overhead. Experiments on VLN-CE benchmarks and real-robot deployment demonstrate state-of-the-art performance and improved long-horizon robustness under diverse lighting. We will release code for the community soon.
Authors: Hao Zhang, Yaru Niu, Yikai Wang, Ding Zhao, H. Eric Tseng
Abstract: To improve generalization and resilience in human-robot collaboration (HRC), robots must handle the combinatorial diversity of human behaviors and contexts, motivating multi-agent reinforcement learning (MARL). However, inherent heterogeneity between robots and humans creates a rationality gap (RG) in the learning process-a variational mismatch between decentralized best-response dynamics and centralized cooperative ascent. The resulting learning problem is a general-sum differentiable game, so independent policy-gradient updates can oscillate or diverge without added structure. We propose heterogeneous-agent Lyapunov policy optimization (HALyPO), which establishes formal stability directly in the policy-parameter space by enforcing a per-step Lyapunov decrease condition on a parameter-space disagreement metric. Unlike Lyapunov-based safe RL, which targets state/trajectory constraints in constrained Markov decision processes, HALyPO uses Lyapunov certification to stabilize decentralized policy learning. HALyPO rectifies decentralized gradients via optimal quadratic projections, ensuring monotonic contraction of RG and enabling effective exploration of open-ended interaction spaces. Extensive simulations and real-world humanoid-robot experiments show that this certified stability improves generalization and robustness in collaborative corner cases.
Authors: Taha Racicot
Abstract: High-stakes synthetic data generation faces a fundamental Quadrilemma: achieving Fidelity to the original distribution, Control over complex logical constraints, Reliability in uncertainty estimation, and Efficiency in computational cost -- simultaneously. State-of-the-art Deep Generative Models (CTGAN, TabDDPM) excel at fidelity but rely on inefficient rejection sampling for continuous range constraints. Conversely, Structural Causal Models offer logical control but struggle with high-dimensional fidelity and complex noise inversion. We introduce JANUS (Joint Ancestral Network for Uncertainty and Synthesis), a framework that unifies these capabilities using a DAG of Bayesian Decision Trees. Our key innovation is Reverse-Topological Back-filling, an algorithm that propagates constraints backwards through the causal graph, achieving 100% constraint satisfaction on feasible constraint sets without rejection sampling. This is paired with an Analytical Uncertainty Decomposition derived from Dirichlet priors, enabling 128x faster uncertainty estimation than Monte Carlo methods. Across 15 datasets and 523 constrained scenarios, JANUS achieves state-of-the-art fidelity (Detection Score 0.497), eliminates mode collapse on imbalanced data, and provides exact handling of complex inter-column constraints (e.g., Salary_offered >= Salary_requested) where baselines fail entirely.
Authors: Hao Zhang, Yves Tseng, Ding Zhao, H. Eric Tseng
Abstract: Cooperative object transport in unstructured environments remains challenging for assistive humanoids because strong, time-varying interaction forces can make tracking-centric whole-body control unreliable, especially in close-contact support tasks. This paper proposes a bio-inspired, interaction-oriented whole-body control (IO-WBC) that functions as an artificial cerebellum - an adaptive motor agent that translates upstream (skill-level) commands into stable, physically consistent whole-body behavior under contact. This work structurally separates upper-body interaction execution from lower-body support control, enabling the robot to maintain balance while shaping force exchange in a tightly coupled robot-object system. A trajectory-optimized reference generator (RG) provides a kinematic prior, while a reinforcement learning (RL) policy governs body responses under heavy-load interactions and disturbances. The policy is trained in simulation with randomized payload mass/inertia and external perturbations, and deployed via asymmetric teacher-student distillation so that the student relies only on proprioceptive histories at runtime. Extensive experiments demonstrate that IO-WBC maintains stable whole-body behavior and physical interaction even when precise velocity tracking becomes infeasible, enabling compliant object transport across a wide range of scenarios.
Authors: Chuang Zhang, Zizhen Zhu, Yihao Wei, Bing Tian, Junyi Liu, Henan Wang, Xavier Wang, Yaxiao Liu
Abstract: Large language models (LLMs) demonstrate superior reasoning capabilities compared to small language models (SLMs), but incur substantially higher costs. We propose COllaborative REAsoner (COREA), a system that cascades an SLM with an LLM to achieve a balance between accuracy and cost in complex reasoning tasks. COREA first attempts to answer questions using the SLM, which outputs both an answer and a verbalized confidence score. Questions with confidence below a predefined threshold are deferred to the LLM for more accurate resolution. We introduce a reinforcement learning-based training algorithm that aligns the SLM's confidence through an additional confidence calibration reward. Extensive experiments demonstrate that our method jointly improves the SLM's reasoning ability and confidence calibration across diverse datasets and model backbones. Compared to using the LLM alone, COREA reduces cost by 21.5% and 16.8% on out-of-domain math and non-math datasets, respectively, with only an absolute pass@1 drop within 2%.
Authors: Taotao Wang, Lizhao You, Jingwen Tong, Chonghe Zhao, Shengli Zhang
Abstract: The ongoing shift of AI models from centralized cloud APIs to local AI agents on edge devices is enabling \textit{Client-Side Autonomous Agents (CSAAs)} -- persistent personal agents that can plan, access local context, and invoke tools on behalf of users. As these agents begin to collaborate by delegating subtasks directly between clients, they naturally form \emph{Agentic Peer-to-Peer (P2P) Networks}. Unlike classic file-sharing overlays where the exchanged object is static, hash-indexed content (e.g., files in BitTorrent), agentic overlays exchange \emph{capabilities and actions} that are heterogeneous, state-dependent, and potentially unsafe if delegated to untrusted peers. This article outlines the networking foundations needed to make such collaboration practical. We propose a plane-based reference architecture that decouples connectivity/identity, semantic discovery, and execution. Besides, we introduce signed, soft-state capability descriptors to support intent- and constraint-aware discovery. To cope with adversarial settings, we further present a \textit{tiered verification} spectrum: Tier~1 relies on reputation signals, Tier~2 applies lightweight canary challenge-response with fallback selection, and Tier~3 requires evidence packages such as signed tool receipts/traces (and, when applicable, attestation). Using a discrete-event simulator that models registry-based discovery, Sybil-style index poisoning, and capability drift, we show that tiered verification substantially improves end-to-end workflow success while keeping discovery latency near-constant and control-plane overhead modest.
Authors: Emile Anand, Ishani Karmarkar
Abstract: Many large-scale platforms and networked control systems have a centralized decision maker interacting with a massive population of agents under strict observability constraints. Motivated by such applications, we study a cooperative Markov game with a global agent and $n$ homogeneous local agents in a communication-constrained regime, where the global agent only observes a subset of $k$ local agent states per time step. We propose an alternating learning framework $(\texttt{ALTERNATING-MARL})$, where the global agent performs subsampled mean-field $Q$-learning against a fixed local policy, and local agents update by optimizing in an induced MDP. We prove that these approximate best-response dynamics converge to an $\widetilde{O}(1/\sqrt{k})$-approximate Nash Equilibrium, while yielding a separation in the sample complexities between the joint state space and action space. Finally, we validate our results in numerical simulations for multi-robot control and federated optimization.
Authors: Hao Zhang, Ding Zhao, H. Eric Tseng
Abstract: Effective human-robot collaboration (HRC) requires translating high-level intent into contact-stable whole-body motion while continuously adapting to a human partner. Many vision-language-action (VLA) systems learn end-to-end mappings from observations and instructions to actions, but they often emphasize reactive (System 1-like) behavior and leave under-specified how sustained System 2-style deliberation can be integrated with reliable, low-latency continuous control. This gap is acute in multi-agent HRC, where long-horizon coordination decisions and physical execution must co-evolve under contact, feasibility, and safety constraints. We address this limitation with cognition-to-control (C2C), a three-layer hierarchy that makes the deliberation-to-control pathway explicit: (i) a VLM-based grounding layer that maintains persistent scene referents and infers embodiment-aware affordances/constraints; (ii) a deliberative skill/coordination layer-the System 2 core-that optimizes long-horizon skill choices and sequences under human-robot coupling via decentralized MARL cast as a Markov potential game with a shared potential encoding task progress; and (iii) a whole-body control layer that executes the selected skills at high frequency while enforcing kinematic/dynamic feasibility and contact stability. The deliberative layer is realized as a residual policy relative to a nominal controller, internalizing partner dynamics without explicit role assignment. Experiments on collaborative manipulation tasks show higher success and robustness than single-agent and end-to-end baselines, with stable coordination and emergent leader-follower behaviors.
Authors: Youngmin Kim, Jaeyun Shin, Jeongchan Kim, Taehoon Lee, Jaemin Kim, Peter Hsu, Jelle Veraart, Jong Chul Ye
Abstract: Ultra Low Field (64 mT) brain MRI improves accessibility but suffers from reduced image quality compared to 3 T. As paired 64 mT - 3 T scans are scarce, we propose an unpaired 64 mT $\rightarrow$ 3 T translation framework that enhances realism while preserving anatomy. Our method builds upon the Unpaired Neural Schr\"odinge Bridge (UNSB) with multi-step refinement. To strengthen target distribution alignment, we augment the adversarial objective with DMD2-style diffusion-guided distribution matching using a frozen 3T diffusion teacher. To explicitly constrain global structure beyond patch-level correspondence, we combine PatchNCE with an Anatomical Structure Preservation (ASP) regularizer that enforces soft foreground background consistency and boundary aware constraints. Evaluated on two disjoint cohorts, the proposed framework achieves an improved realism structure trade-off, enhancing distribution level realism on unpaired benchmarks while increasing structural fidelity on the paired cohort compared to unpaired baselines.
Authors: Pengfei Tong, Siyuan Chen, Chenwei Zhang, Bo Wang, Qi Pi, Pixun Li, Zuotao Liu
Abstract: Most large-scale recommender systems follow a multi-stage cascade of retrieval, pre-ranking, ranking, and re-ranking. A key challenge at the pre-ranking stage arises from the heterogeneity of training instances sampled from coarse-grained retrieval results, fine-grained ranking signals, and exposure feedback. Our analysis reveals that prevailing pre-ranking methods, which indiscriminately mix heterogeneous samples, suffer from gradient conflicts: hard samples dominate training while easy ones remain underutilized, leading to suboptimal performance. We further show that the common practice of uniformly scaling model complexity across all samples is inefficient, as it overspends computation on easy cases and slows training without proportional gains. To address these limitations, this paper presents Heterogeneity-Aware Adaptive Pre-ranking (HAP), a unified framework that mitigates gradient conflicts through conflict-sensitive sampling coupled with tailored loss design, while adaptively allocating computational budgets across candidates. Specifically, HAP disentangles easy and hard samples, directing each subset along dedicated optimization paths. Building on this separation, it first applies lightweight models to all candidates for efficient coverage, and further engages stronger models on the hard ones, maintaining accuracy while reducing cost. This approach not only improves pre-ranking effectiveness but also provides a practical perspective on scaling strategies in industrial recommender systems. HAP has been deployed in the Toutiao production system for 9 months, yielding up to 0.4% improvement in user app usage duration and 0.05% in active days, without additional computational cost. We also release a large-scale industrial hybrid-sample dataset to enable the systematic study of source-driven candidate heterogeneity in pre-ranking.
Authors: Naili Xing, Haotian Gao, Zhanhao Zhao, Shaofeng Cai, Zhaojing Luo, Yuncheng Wu, Zhongle Xie, Meihui Zhang, Beng Chin Ooi
Abstract: AI-driven analytics are increasingly crucial to data-centric decision-making. The practice of exporting data to machine learning runtimes incurs high overhead, limits robustness to data drift, and expands the attack surface, especially in multi-tenant, heterogeneous data systems. Integrating AI directly into database engines, while offering clear benefits, introduces challenges in managing joint query processing and model execution, optimizing end-to-end performance, coordinating execution under resource contention, and enforcing strong security and access-control guarantees. This paper discusses the challenges of joint DB-AI, or AIxDB, data management and query processing within AI-powered data systems. It presents various challenges that need to be addressed carefully, such as query optimization, execution scheduling, and distributed execution over heterogeneous hardware. Database components such as transaction management and access control need to be re-examined to support AI lifecycle management, mitigate data drift, and protect sensitive data from unauthorized AI operations. We present a design and preliminary results to demonstrate what may be key to the performance for serving AIxDB queries.
Authors: Satoshi Oyama, Yuko Sakurai, Hisashi Kashima
Abstract: Scientific discovery still relies heavily on the manual efforts of individual researchers, leading to limited exploration, redundant trials, and reduced reproducibility. Human-participant data analysis competitions generate diverse approaches, yet fluctuations in participation and the lack of independent repetitions show that parallel exploration alone is insufficient for achieving reliable scientific inquiry. As advanced AI agents based on large language models (LLMs) increasingly perform analytical tasks, relying on a single highly capable agent is unlikely to overcome these structural limitations. Recent work has begun to explore how multiple LLM-based agents can collaborate or compete in scientific workflows-a growing trend we refer to as MA4Science. However, most existing MA4Science studies assume that all agents are controlled by a single organizational entity, limiting their ability to examine how institutional mechanisms-such as incentives, information sharing, and reproducibility-shape collective exploration among independently managed agents. To address this gap, we introduce MACC (Multi-Agent Collaborative Competition), an institutional architecture that integrates a blackboard-style shared scientific workspace with incentive mechanisms designed to encourage transparency, reproducibility, and exploration efficiency. MACC provides a testbed for studying how institutional design influences scalable and reliable multi-agent scientific exploration.
Authors: Jiawei Cheng, Min Gao, Zongwei Wang, Xiaofei Zhu, Zhiyi Liu, Wentao Li, Wei Li, Huan Wu
Abstract: Shared-account usage is common on streaming and e-commerce platforms, where multiple users share one account. Existing shared-account sequential recommendation (SSR) methods often assume a fixed number of latent users per account, limiting their ability to adapt to diverse sharing patterns and reducing recommendation accuracy. Recent latent reasoning technique applied in sequential recommendation (SR) generate intermediate embeddings from the user embedding (e.g, last item embedding) to uncover users' potential interests, which inspires us to treat the problem of inferring the number of latent users as generating a series of intermediate embeddings, shifting from inferring preferences behind user to inferring the users behind account. However, the last item cannot be directly used for reasoning in SSR, as it can only represent the behavior of the most recent latent user, rather than the collective behavior of the entire account. To address this, we propose DisenReason, a two-stage reasoning method tailored to SSR. DisenReason combines behavior disentanglement stage from frequency-domain perspective to create a collective and unified account behavior representation, which serves as a pivot for latent user reasoning stage to infer the number of users behind the account. Experiments on four benchmark datasets show that DisenReason consistently outperforms all state-of-the-art baselines across four benchmark datasets, achieving relative improvements of up to 12.56\% in MRR@5 and 6.06\% in Recall@20.
Authors: Qinsi Wang, Hancheng Ye, Jinhee Kim, Jinghan Ke, Yifei Wang, Martin Kuo, Zishan Shao, Dongting Li, Yueqian Lin, Ting Jiang, Chiyue Wei, Qi Qian, Wei Wen, Helen Li, Yiran Chen
Abstract: Think about how human handles complex reading tasks: marking key points, inferring their relationships, and structuring information to guide understanding and responses. Likewise, can a large language model benefit from text structure to enhance text-processing performance? To explore it, in this work, we first introduce Structure of Thought (SoT), a prompting technique that explicitly guides models to construct intermediate text structures, consistently boosting performance across eight tasks and three model families. Building upon this insight, we present T2S-Bench, the first benchmark designed to evaluate and improve text-to-structure capabilities of models. T2S-Bench includes 1.8K samples across 6 scientific domains and 32 structural types, rigorously constructed to ensure accuracy, fairness, and quality. Evaluation on 45 mainstream models reveals substantial improvement potential: the average accuracy on the multi-hop reasoning task is only 52.1%, and even the most advanced model achieves 58.1% node accuracy in end-to-end extraction. Furthermore, on Qwen2.5-7B-Instruct, SoT alone yields an average +5.7% improvement across eight diverse text-processing tasks, and fine-tuning on T2S-Bench further increases this gain to +8.6%. These results highlight the value of explicit text structuring and the complementary contributions of SoT and T2S-Bench. Dataset and eval code have been released at https://t2s-bench.github.io/T2S-Bench-Page/.
Authors: Taejun Lim, Joong-Won Hwang, Kibok Lee
Abstract: When continual test-time adaptation (TTA) persists over the long term, errors accumulate in the model and further cause it to predict only a few classes for all inputs, a phenomenon known as model collapse. Recent studies have explored reset strategies that completely erase these accumulated errors. However, their periodic resets lead to suboptimal adaptation, as they occur independently of the actual risk of collapse. Moreover, their full resets cause catastrophic loss of knowledge acquired over time, even though such knowledge could be beneficial in the future. To this end, we propose (1) an Adaptive and Selective Reset (ASR) scheme that dynamically determines when and where to reset, (2) an importance-aware regularizer to recover essential knowledge lost due to reset, and (3) an on-the-fly adaptation adjustment scheme to enhance adaptability under challenging domain shifts. Extensive experiments across long-term TTA benchmarks demonstrate the effectiveness of our approach, particularly under challenging conditions. Our code is available at https://github.com/YonseiML/asr.
Authors: Santanu Mondal, T. Chithralekha
Abstract: Central Bank Digital Currency (CBDCs) are becoming a new digital financial tool aimed at financial inclusion, increased monetary stability, and improved efficiency of payment systems, as they are issued by central banks. One of the most important aspects is that the CBDC must offer secure offline payment methods to users, allowing them to retain cash-like access without violating Anti-Money Laundering and Counter-terrorism Financing (AML/CFT) rules. The offline CBDC ecosystems will provide financial inclusion, empower underserved communities, and ensure equitable access to digital payments, even in connectivity-poor remote locations. With the rapid growth of Internet of Things (IoT) devices in our everyday lives, they are capable of performing secure digital transactions. Integrating offline CBDC payment with IoT devices enables seamless, automated payment without internet connectivity. However, IoT devices face special challenges due to their resource-constrained nature. This makes it difficult to include features such as double-spending prevention, privacy preservation, low-computation operation, and digital identity management. The work proposes a privacy-preserving offline CBDC model with integrated secure elements (SEs), zero-knowledge proofs (ZKPs), and intermittent synchronisation to conduct offline payments on IoT hardware. The proposed model is based on recent improvements in offline CBDC prototypes, regulations and cryptographic design choices such as hybrid architecture that involves using combination of online and offline payment in IoT devices using secure hardware with lightweight zero-knowledge proof cryptographic algorithm.
Authors: Yanbo Wang, Jiaxuan You, Chuan Shi, Muhan Zhang
Abstract: Relational Databases (RDBs) are the backbone of modern business, yet they lack foundation models comparable to those in text or vision. A key obstacle is that high-quality RDBs are private, scarce and structurally heterogeneous, making internet-scale pre-training infeasible. To overcome this data scarcity, We introduce $\textbf{RDB-PFN}$, the first relational foundation model trained purely via $\textbf{synthetic data}$. Inspired by Prior-Data Fitted Networks (PFNs) where synthetic data generated from Structural Causal Models (SCMs) enables reasoning on single tables, we design a $\textbf{Relational Prior Generator}$ to create an infinite stream of diverse RDBs from scratch. Pre-training on $\textbf{over 2 million}$ synthetic single-table and relational tasks, RDB-PFN learns to adapt to any new database instantly via genuine $\textbf{in-context learning}$. Experiments verify RDB-PFN achieves strong few-shot performance on 19 real-world relational prediction tasks, outperforming graph-based and single-table foundation-model baselines (given the same DFS-linearized inputs), while using a lightweight architecture and fast inference. The code is available at https://github.com/MuLabPKU/RDBPFN
Authors: Hanpeng Liu, Zidan Wang, Shuoxi Zhang, Kaiyuan Gao, Kun He
Abstract: The state space model Mamba has recently emerged as a promising paradigm in computer vision, attracting significant attention due to its efficient processing of long sequence tasks. Mamba's inherent causal mechanism renders it particularly suitable for autoregressive pretraining. However, current autoregressive pretraining methods are constrained to short sequence tasks, failing to fully exploit Mamba's prowess in handling extended sequences. To address this limitation, we introduce an innovative autoregressive pretraining method for Vision Mamba that substantially extends the input sequence length. We introduce new \textbf{S}epara\textbf{T}ors for \textbf{A}uto\textbf{R}egressive pretraining to demarcate and differentiate between different images, known as \textbf{STAR}. Specifically, we insert identical separators before each image to demarcate its inception. This strategy enables us to quadruple the input sequence length of Vision Mamba while preserving the original dimensions of the dataset images. Employing this long sequence pretraining technique, our STAR-B model achieved an impressive accuracy of 83.5\% on ImageNet-1k, which is highly competitive in Vision Mamba. These results underscore the potential of our method in enhancing the performance of vision models through improved leveraging of long-range dependencies.
Authors: Huihan Liu, Changyeon Kim, Bo Liu, Minghuan Liu, Yuke Zhu
Abstract: Continual learning is a long-standing challenge in robot policy learning, where a policy must acquire new skills over time without catastrophically forgetting previously learned ones. While prior work has extensively studied continual learning in relatively small behavior cloning (BC) policy models trained from scratch, its behavior in modern large-scale pretrained Vision-Language-Action (VLA) models remains underexplored. In this work, we found that pretrained VLAs are remarkably resistant to forgetting compared with smaller policy models trained from scratch. Simple Experience Replay (ER) works surprisingly well on VLAs, sometimes achieving zero forgetting even with a small replay data size. Our analysis reveals that pretraining plays a critical role in downstream continual learning performance: large pretrained models mitigate forgetting with a small replay buffer size while maintaining strong forward learning capabilities. Furthermore, we found that VLAs can retain relevant knowledge from prior tasks despite performance degradation during learning new tasks. This knowledge retention enables rapid recovery of seemingly forgotten skills through finetuning. Together, these insights imply that large-scale pretraining fundamentally changes the dynamics of continual learning, enabling models to continually acquire new skills over time with simple replay. Code and more information can be found at https://ut-austin-rpl.github.io/continual-vla
Authors: Yun Lu, Xiaoyu Shi, Hong Xie, Xiangyu Zhao, Mingsheng Shang
Abstract: Interactive recommender systems (IRS) are increasingly optimized with Reinforcement Learning (RL) to capture the sequential nature of user-system dynamics. However, existing fairness-aware methods often suffer from a fundamental oversight: they assume the observed user state is a faithful representation of true preferences. In reality, implicit feedback is contaminated by popularity-driven noise and exposure bias, creating a distorted state that misleads the RL agent. We argue that the persistent conflict between accuracy and fairness is not merely a reward-shaping issue, but a state estimation failure. In this work, we propose \textbf{DSRM-HRL}, a framework that reformulates fairness-aware recommendation as a latent state purification problem followed by decoupled hierarchical decision-making. We introduce a Denoising State Representation Module (DSRM) based on diffusion models to recover the low-entropy latent preference manifold from high-entropy, noisy interaction histories. Built upon this purified state, a Hierarchical Reinforcement Learning (HRL) agent is employed to decouple conflicting objectives: a high-level policy regulates long-term fairness trajectories, while a low-level policy optimizes short-term engagement under these dynamic constraints. Extensive experiments on high-fidelity simulators (KuaiRec, KuaiRand) demonstrate that DSRM-HRL effectively breaks the "rich-get-richer" feedback loop, achieving a superior Pareto frontier between recommendation utility and exposure equity.
Authors: Jialong Chen, Xander Xu, Hu Wei, Chuan Chen, Bing Zhao
Abstract: Large language model (LLM)-powered agents have demonstrated strong capabilities in automating software engineering tasks such as static bug fixing, as evidenced by benchmarks like SWE-bench. However, in the real world, the development of mature software is typically predicated on complex requirement changes and long-term feature iterations -- a process that static, one-shot repair paradigms fail to capture. To bridge this gap, we propose \textbf{SWE-CI}, the first repository-level benchmark built upon the Continuous Integration loop, aiming to shift the evaluation paradigm for code generation from static, short-term \textit{functional correctness} toward dynamic, long-term \textit{maintainability}. The benchmark comprises 100 tasks, each corresponding on average to an evolution history spanning 233 days and 71 consecutive commits in a real-world code repository. SWE-CI requires agents to systematically resolve these tasks through dozens of rounds of analysis and coding iterations. SWE-CI provides valuable insights into how well agents can sustain code quality throughout long-term evolution.
Authors: Ruilin Luo, Chufan Shi, Yizhen Zhang, Cheng Yang, Songtao Jiang, Tongkun Guan, Ruizhe Chen, Ruihang Chu, Peng Wang, Mingkun Yang, Yujiu Yang, Junyang Lin, Zhibo Yang
Abstract: The cold-start initialization stage plays a pivotal role in training Multimodal Large Reasoning Models (MLRMs), yet its mechanisms remain insufficiently understood. To analyze this stage, we introduce the Visual Attention Score (VAS), an attention-based metric that quantifies how much a model attends to visual tokens. We find that reasoning performance is strongly correlated with VAS (r=0.9616): models with higher VAS achieve substantially stronger multimodal reasoning. Surprisingly, multimodal cold-start fails to elevate VAS, resulting in attention distributions close to the base model, whereas text-only cold-start leads to a clear increase. We term this counter-intuitive phenomenon Lazy Attention Localization. To validate its causal role, we design training-free interventions that directly modulate attention allocation during inference, performance gains of 1$-$2% without any retraining. Building on these insights, we further propose Attention-Guided Visual Anchoring and Reflection (AVAR), a comprehensive cold-start framework that integrates visual-anchored data synthesis, attention-guided objectives, and visual-anchored reward shaping. Applied to Qwen2.5-VL-7B, AVAR achieves an average gain of 7.0% across 7 multimodal reasoning benchmarks. Ablation studies further confirm that each component of AVAR contributes step-wise to the overall gains. The code, data, and models are available at https://github.com/lrlbbzl/Qwen-AVAR.
Authors: Wang Jian, Shen Hong, Ke Wei, Liu Xue Hua
Abstract: While federated learning protects data privacy, it also makes the model update process vulnerable to long-term stealthy perturbations. Existing studies on backdoor attacks in federated learning mainly focus on trigger design or poisoning strategies, typically assuming that identical perturbations behave similarly across different model architectures. This assumption overlooks the impact of model structure on perturbation effectiveness. From a structure-aware perspective, this paper analyzes the coupling relationship between model architectures and backdoor perturbations. We introduce two metrics, Structural Responsiveness Score (SRS) and Structural Compatibility Coefficient (SCC), to measure a model's sensitivity to perturbations and its preference for fractal perturbations. Based on these metrics, we develop a structure-aware fractal perturbation injection framework (TFI) to study the role of architectural properties in the backdoor injection process. Experimental results show that model architecture significantly influences the propagation and aggregation of perturbations. Networks with multi-path feature fusion can amplify and retain fractal perturbations even under low poisoning ratios, while models with low structural compatibility constrain their effectiveness. Further analysis reveals a strong correlation between SCC and attack success rate, suggesting that SCC can predict perturbation survivability. These findings highlight that backdoor behaviors in federated learning depend not only on perturbation design or poisoning intensity but also on the interaction between model architecture and aggregation mechanisms, offering new insights for structure-aware defense design.
Authors: Olga Krestinskaya, Mohammed E. Fouda, Ahmed Eltawil, Khaled N. Salama
Abstract: Software-hardware co-design is essential for optimizing in-memory computing (IMC) hardware accelerators for neural networks. However, most existing optimization frameworks target a single workload, leading to highly specialized hardware designs that do not generalize well across models and applications. In contrast, practical deployment scenarios require a single IMC platform that can efficiently support multiple neural network workloads. This work presents a joint hardware-workload co-optimization framework based on an optimized evolutionary algorithm for designing generalized IMC accelerator architectures. By explicitly capturing cross-workload trade-offs rather than optimizing for a single model, the proposed approach significantly reduces the performance gap between workload-specific and generalized IMC designs. The framework is evaluated on both RRAM- and SRAM-based IMC architectures, demonstrating strong robustness and adaptability across diverse design scenarios. Compared to baseline methods, the optimized designs achieve energy-delay-area product (EDAP) reductions of up to 76.2% and 95.5% when optimizing across a small set (4 workloads) and a large set (9 workloads), respectively. The source code of the framework is available at https://github.com/OlgaKrestinskaya/JointHardwareWorkloadOptimizationIMC.
URLs: https://github.com/OlgaKrestinskaya/JointHardwareWorkloadOptimizationIMC.
Authors: Chen Sun, Yash Vekaria, Rishab Nithyanand
Abstract: As LLM-driven agents begin to autonomously navigate the web, their ability to interpret and respond to manipulative interface design becomes critical. A fundamental question that emerges is: can such agents reliably recognize patterns of friction, misdirection, and coercion in interface design (i.e., dark patterns)? We study this question in a setting where the workflows are consequential: website portals associated with the submission of CCPA-related data rights requests. These portals operationalize statutory rights, but they are implemented as interactive interfaces whose design can be structured to facilitate, burden, or subtly discourage the exercise of those rights. We design and deploy an LLM-driven auditing agent capable of end-to-end traversal of rights-request workflows, structured evidence gathering, and classification of potential dark patterns. Across a set of 456 data broker websites, we evaluate: (1) the ability of the agent to consistently locate and complete request flows, (2) the reliability and reproducibility of its dark pattern classifications, and (3) the conditions under which it fails or produces poor judgments. Our findings characterize both the feasibility and the limitations of using LLM-driven agents for scalable dark pattern auditing.
Authors: Martin Kosteln\'ik, Michal Hradi\v{s}, Martin Do\v{c}ekal
Abstract: Topic localization aims to identify spans of text that express a given topic defined by a name and description. To study this task, we introduce a human-annotated benchmark based on Czech historical documents, containing human-defined topics together with manually annotated spans and supporting evaluation at both document and word levels. Evaluation is performed relative to human agreement rather than a single reference annotation. We evaluate a diverse range of large language models alongside BERT-based models fine-tuned on a distilled development dataset. Results reveal substantial variability among LLMs, with performance ranging from near-human topic detection to pronounced failures in span localization. While the strongest models approach human agreement, the distilled token embedding models remain competitive despite their smaller scale. The dataset and evaluation framework are publicly available at: https://github.com/dcgm/czechtopic.
Authors: Frederik Hagelskj{\ae}r
Abstract: In this paper, we present a network structure for classifying metadata of cuneiform tablets. The problem is of practical importance, as the size of the existing corpus far exceeds the number of experts available to analyze it. But the task is made difficult by the combination of limited annotated datasets and the high-resolution point-cloud representation of each tablet. To address this, we develop a convolution-inspired architecture that gradually down-scales the point cloud while integrating local neighbor information. The final down-scaled point cloud is then processed by computing neighbors in the feature space to include global information. Our method is compared with the state-of-the-art transformer-based network Point-BERT, and consistently obtains the best performance. Source code and datasets will be released at publication.
Authors: Markus Knauer, Samuel Bustamante, Thomas Eiband, Alin Albu-Sch\"affer, Freek Stulp, Jo\~ao Silv\'erio
Abstract: Foundation models have demonstrated impressive capabilities across diverse domains, while imitation learning provides principled methods for robot skill adaptation from limited data. Combining these approaches holds significant promise for direct application to robotics, yet this combination has received limited attention, particularly for industrial deployment. We present a novel framework that enables open-vocabulary skill adaptation through a tool-based architecture, maintaining a protective abstraction layer between the language model and robot hardware. Our approach leverages pre-trained LLMs to select and parameterize specific tools for adapting robot skills without requiring fine-tuning or direct model-to-robot interaction. We demonstrate the framework on a 7-DoF torque-controlled robot performing an industrial bearing ring insertion task, showing successful skill adaptation through natural language commands for speed adjustment, trajectory correction, and obstacle avoidance while maintaining safety, transparency, and interpretability.
Authors: Hiroki Tomioka, Genta Yoshimura
Abstract: Time series forecasting, which predicts future values from past observations, plays a central role in many domains and has driven the development of highly accurate neural network models. However, the complexity of these models often limits human understanding of the rationale behind their predictions. We propose PatchDecomp, a neural network-based time series forecasting method that achieves both high accuracy and interpretability. PatchDecomp divides input time series into subsequences (patches) and generates predictions by aggregating the contributions of each patch. This enables clear attribution of each patch, including those from exogenous variables, to the final prediction. Experiments on multiple benchmark datasets demonstrate that PatchDecomp provides predictive performance comparable to recent forecasting methods. Furthermore, we show that the model's explanations not only influence predicted values quantitatively but also offer qualitative interpretability through visualization of patch-wise contributions.
Authors: Ji-Lun Peng, Yun-Nung Chen
Abstract: Large language models (LLMs) have demonstrated significant potential in developing Role-Playing Agents (RPAs). However, current research primarily evaluates RPAs using famous fictional characters, allowing models to rely on memory associated with character names. This dependency creates a bias that limits the generalization of RPAs to unseen personas. To address this issue, we propose an anonymous evaluation method. Experiments across multiple benchmarks reveal that anonymization significantly degrades role-playing performance, confirming that name exposure carries implicit information. Furthermore, we investigate personality augmentation to enhance role fidelity under anonymous setting. We systematically compare the efficacy of personality traits derived from human annotations versus those self-generated by the model. Our results demonstrate that incorporating personality information consistently improves RPA performance. Crucially, self-generated personalities achieve performance comparable to human-annotated ones. This work establishes a fairer evaluation protocol and validates a scalable, personality-enhanced framework for constructing robust RPAs.
Authors: Yuhan Xie, Chen Lyu
Abstract: Model Merging (MM) has emerged as a scalable paradigm for multi-task learning (MTL), enabling multiple task-specific models to be integrated without revisiting the original training data. Despite recent progress, the reliability of MM under test-time distribution shift remains insufficiently understood. Most existing MM methods typically assume that test data are clean and distributionally aligned with both the training and auxiliary sources. However, this assumption rarely holds in practice, often resulting in biased predictions with degraded generalization. To address this issue, we present BD-Merging, a bias-aware unsupervised model merging framework that explicitly models uncertainty to achieve adaptive reliability under distribution shift. First, BD-Merging introduces a joint evidential head that learns uncertainty over a unified label space, capturing cross-task semantic dependencies in MM. Second, building upon this evidential foundation, we propose an Adjacency Discrepancy Score (ADS) that quantifies evidential alignment among neighboring samples. Third, guided by ADS, a discrepancy-aware contrastive learning mechanism refines the merged representation by aligning consistent samples and separating conflicting ones. Combined with general unsupervised learning, this process trains a debiased router that adaptively allocates task-specific or layer-specific weights on a per-sample basis, effectively mitigating the adverse effects of distribution shift. Extensive experiments across diverse tasks demonstrate that BD-Merging achieves superior effectiveness and robustness compared to state-of-the-art MM baselines.
Authors: Nicolas Helson, Pegah Alizadeh, Anastasios Giovanidis
Abstract: Offline Reinforcement Learning (RL) is a promising approach for next-generation wireless networks, where online exploration is unsafe and large amounts of operational data can be reused across the model lifecycle. However, the behavior of offline RL algorithms under genuinely stochastic dynamics -- inherent to wireless systems due to fading, noise, and traffic mobility -- remains insufficiently understood. We address this gap by evaluating Bellman-based (Conservative Q-Learning), sequence-based (Decision Transformers), and hybrid (Critic-Guided Decision Transformers) offline RL methods in an open-access stochastic telecom environment (mobile-env). Our results show that Conservative Q-Learning consistently produces more robust policies across different sources of stochasticity, making it a reliable default choice in lifecycle-driven AI management frameworks. Sequence-based methods remain competitive and can outperform Bellman-based approaches when sufficient high-return trajectories are available. These findings provide practical guidance for offline RL algorithm selection in AI-driven network control pipelines, such as O-RAN and future 6G functions, where robustness and data availability are key operational constraints.
Authors: Radia Daci, Vito Ren\`o, Cosimo Patruno, Angelo Cardellicchio, Abdelmalik Taleb-Ahmed, Marco Leo, Cosimo Distante
Abstract: Multimodal industrial anomaly detection benefits from integrating RGB appearance with 3D surface geometry, yet existing \emph{unsupervised} approaches commonly rely on memory banks, teacher-student architectures, or fragile fusion schemes, limiting robustness under noisy depth, weak texture, or missing modalities. This paper introduces \textbf{CMDR-IAD}, a lightweight and modality-flexible unsupervised framework for reliable anomaly detection in 2D+3D multimodal as well as single-modality (2D-only or 3D-only) settings. \textbf{CMDR-IAD} combines bidirectional 2D$\leftrightarrow$3D cross-modal mapping to model appearance-geometry consistency with dual-branch reconstruction that independently captures normal texture and geometric structure. A two-part fusion strategy integrates these cues: a reliability-gated mapping anomaly highlights spatially consistent texture-geometry discrepancies, while a confidence-weighted reconstruction anomaly adaptively balances appearance and geometric deviations, yielding stable and precise anomaly localization even in depth-sparse or low-texture regions. On the MVTec 3D-AD benchmark, CMDR-IAD achieves state-of-the-art performance while operating without memory banks, reaching 97.3\% image-level AUROC (I-AUROC), 99.6\% pixel-level AUROC (P-AUROC), and 97.6\% AUPRO. On a real-world polyurethane cutting dataset, the 3D-only variant attains 92.6\% I-AUROC and 92.5\% P-AUROC, demonstrating strong effectiveness under practical industrial conditions. These results highlight the framework's robustness, modality flexibility, and the effectiveness of the proposed fusion strategies for industrial visual inspection. Our source code is available at https://github.com/ECGAI-Research/CMDR-IAD/
Authors: Jaewon Lee, Jaeseok Heo, Gunmin Lee, Howoong Jun, Jeongwoo Oh, Songhwai Oh
Abstract: Safe visual navigation is critical for indoor mobile robots operating in cluttered environments. Existing benchmarks, however, often neglect collisions or are designed for outdoor scenarios, making them unsuitable for indoor visual navigation. To address this limitation, we introduce the reactive visual navigation benchmark (RVN-Bench), a collision-aware benchmark for indoor mobile robots. In RVN-Bench, an agent must reach sequential goal positions in previously unseen environments using only visual observations and no prior map, while avoiding collisions. Built on the Habitat 2.0 simulator and leveraging high-fidelity HM3D scenes, RVN-Bench provides large-scale, diverse indoor environments, defines a collision-aware navigation task and evaluation metrics, and offers tools for standardized training and benchmarking. RVN-Bench supports both online and offline learning by offering an environment for online reinforcement learning, a trajectory image dataset generator, and tools for producing negative trajectory image datasets that capture collision events. Experiments show that policies trained on RVN-Bench generalize effectively to unseen environments, demonstrating its value as a standardized benchmark for safe and robust visual navigation. Code and additional materials are available at: https://rvn-bench.github.io/.
Authors: Chengxuan Lu, Zhenquan Zhang, Shukuan Wang, Qunzhi Lin, Baigui Sun, Yang Liu
Abstract: Post-training with reinforcement learning (RL) has recently shown strong promise for advancing multimodal agents beyond supervised imitation. However, RL remains limited by poor data efficiency, particularly in settings where interaction data are scarce and quickly become outdated. To address this challenge, GIPO (Gaussian Importance sampling Policy Optimization) is proposed as a policy optimization objective based on truncated importance sampling, replacing hard clipping with a log-ratio-based Gaussian trust weight to softly damp extreme importance ratios while maintaining non-zero gradients. Theoretical analysis shows that GIPO introduces an implicit, tunable constraint on the update magnitude, while concentration bounds guarantee robustness and stability under finite-sample estimation. Experimental results show that GIPO achieves state-of-the-art performance among clipping-based baselines across a wide range of replay buffer sizes, from near on-policy to highly stale data, while exhibiting superior bias--variance trade-off, high training stability and improved sample efficiency.
Authors: Jinkun You, Jiaxin Cheng, Jie Zhang, Yicong Zhou
Abstract: Supervised and unsupervised homography estimation methods depend on image pairs tailored to specific modalities to achieve high accuracy. However, their performance deteriorates substantially when applied to unseen modalities. To address this issue, we propose a training data synthesis method that generates unaligned image pairs with ground-truth offsets from a single input image. Our approach renders the image pairs with diverse textures and colors while preserving their structural information. These synthetic data empower the trained model to achieve greater robustness and improved generalization across various domains. Additionally, we design a network to fully leverage cross-scale information and decouple color information from feature representations, thus improving estimation accuracy. Extensive experiments show that our training data synthesis method improves generalization performance. The results also confirm the effectiveness of the proposed network.
Authors: Hantong Feng, Yonggang Wu, Duxin Chen, Wenwu Yu
Abstract: Dynamic link prediction plays a crucial role in diverse applications including social network analysis, communication forecasting, and financial modeling. While recent Transformer-based approaches have demonstrated promising results in temporal graph learning, their performance remains limited when capturing complex multi-scale temporal dynamics. In this paper, we propose TFWaveFormer, a novel Transformer architecture that integrates temporal-frequency analysis with multi-resolution wavelet decomposition to enhance dynamic link prediction. Our framework comprises three key components: (i) a temporal-frequency coordination mechanism that jointly models temporal and spectral representations, (ii) a learnable multi-resolution wavelet decomposition module that adaptively extracts multi-scale temporal patterns through parallel convolutions, replacing traditional iterative wavelet transforms, and (iii) a hybrid Transformer module that effectively fuses local wavelet features with global temporal dependencies. Extensive experiments on benchmark datasets demonstrate that TFWaveFormer achieves state-of-the-art performance, outperforming existing Transformer-based and hybrid models by significant margins across multiple metrics. The superior performance of TFWaveFormer validates the effectiveness of combining temporal-frequency analysis with wavelet decomposition in capturing complex temporal dynamics for dynamic link prediction tasks.
Authors: Hengquan Guo
Abstract: We present \textbf{BLOCK}, an open-source bi-stage character-to-skin pipeline that generates pixel-perfect Minecraft skins from arbitrary character concepts. BLOCK decomposes the problem into (i) a \textbf{3D preview synthesis stage} driven by a large multimodal model (MLLM) with a carefully designed prompt-and-reference template, producing a consistent dual-panel (front/back) oblique-view Minecraft-style preview; and (ii) a \textbf{skin decoding stage} based on a fine-tuned FLUX.2 model that translates the preview into a skin atlas image. We further propose \textbf{EvolveLoRA}, a progressive LoRA curriculum (text-to-image $\rightarrow$ image-to-image $\rightarrow$ preview-to-skin) that initializes each phase from the previous adapter to improve stability and efficiency. BLOCK is released with all prompt templates and fine-tuned weights to support reproducible character-to-skin generation.
Authors: Michael J\"ulich
Abstract: Generative AI can convert uncertainty into authoritative-seeming verdicts, displacing the justificatory work on which democratic epistemic agency depends. As a corrective, I propose a Brouwer-inspired assertibility constraint for responsible AI: in high-stakes domains, systems may assert or deny claims only if they can provide a publicly inspectable and contestable certificate of entitlement; otherwise they must return "Undetermined". This constraint yields a three-status interface semantics (Asserted, Denied, Undetermined) that cleanly separates internal entitlement from public standing while connecting them via the certificate as a boundary object. It also produces a time-indexed entitlement profile that is stable under numerical refinement yet revisable as the public record changes. I operationalize the constraint through decision-layer gating of threshold and argmax outputs, using internal witnesses (e.g., sound bounds or separation margins) and an output contract with reason-coded abstentions. A design lemma shows that any total, certificate-sound binary interface already decides the deployed predicate on its declared scope, so "Undetermined" is not a tunable reject option but a mandatory status whenever no forcing witness is available. By making outputs answerable to challengeable warrants rather than confidence alone, the paper aims to preserve epistemic agency where automated speech enters public justification.
Authors: Simon Kohaut, Benedict Flade, Julian Eggert, Kristian Kersting, Devendra Singh Dhami
Abstract: Exact inference in probabilistic First-Order Logic offers a promising yet computationally costly approach for regulating the behavior of autonomous agents in shared traffic spaces. While prior methods have combined logical and probabilistic data into decision-making frameworks, their application is often limited to pre-flight checks due to the complexity of reasoning across vast numbers of possible universes. In this work, we propose a reactive mission design framework that jointly considers uncertain environmental data and declarative, logical traffic regulations. By synthesizing Probabilistic Mission Design (ProMis) with reactive reasoning facilitated by Reactive Circuits (RC), we enable online, exact probabilistic inference over hybrid domains. Our approach leverages the Frequency of Change inherent in heterogeneous data streams to subdivide inference formulas into memoized, isolated tasks, ensuring that only the specific components affected by new sensor data are re-evaluated. In experiments involving both real-world vessel data and simulated drone traffic in dense urban scenarios, we demonstrate that our approach provides orders of magnitude in speedup over ProMis without reactive paradigms. This allows intelligent transportation systems, such as Unmanned Aircraft Systems (UAS), to actively assert safety and legal compliance during operations rather than relying solely on preparation procedures.
Authors: Lifan Jiang, Yuhang Pei, oxi Wu, Yan Zhao, Tianrun Wu, Shulong Yu, Lihui Zhang, Deng Cai
Abstract: Recent advances in MLLMs are reframing segmentation from fixed-category prediction to instruction-grounded localization. While reasoning based segmentation has progressed rapidly in natural scenes, remote sensing lacks a generalizable solution due to the prohibitive cost of reasoning-oriented data and domain-specific challenges like overhead viewpoints. We present GeoSeg, a zero-shot, training-free framework that bypasses the supervision bottleneck for reasoning-driven remote sensing segmentation. GeoSeg couples MLLM reasoning with precise localization via: (i) bias-aware coordinate refinement to correct systematic grounding shifts and (ii) a dual-route prompting mechanism to fuse semantic intent with fine-grained spatial cues. We also introduce GeoSeg-Bench, a diagnostic benchmark of 810 image--query pairs with hierarchical difficulty levels. Experiments show that GeoSeg consistently outperforms all baselines, with extensive ablations confirming the effectiveness and necessity of each component.
Authors: Qianpu Chen, Derya Soydaner, Rob Saunders
Abstract: When visual evidence is ambiguous, vision models must decide whether to interpret face-like patterns as meaningful. Face pareidolia, the perception of faces in non-face objects, provides a controlled probe of this behavior. We introduce a representation-level diagnostic framework that analyzes detection, localization, uncertainty, and bias across class, difficulty, and emotion in face pareidolia images. Under a unified protocol, we evaluate six models spanning four representational regimes: vision-language models (VLMs; CLIP-B/32, CLIP-L/14, LLaVA-1.5-7B), pure vision classification (ViT), general object detection (YOLOv8), and face detection (RetinaFace). Our analysis reveals three mechanisms of interpretation under ambiguity. VLMs exhibit semantic overactivation, systematically pulling ambiguous non-human regions toward the Human concept, with LLaVA-1.5-7B producing the strongest and most confident over-calls, especially for negative emotions. ViT instead follows an uncertainty-as-abstention strategy, remaining diffuse yet largely unbiased. Detection-based models achieve low bias through conservative priors that suppress pareidolia responses even when localization is controlled. These results show that behavior under ambiguity is governed more by representational choices than score thresholds, and that uncertainty and bias are decoupled: low uncertainty can signal either safe suppression, as in detectors, or extreme over-interpretation, as in VLMs. Pareidolia therefore provides a compact diagnostic and a source of ambiguity-aware hard negatives for probing and improving the semantic robustness of vision-language systems. Code will be released upon publication.
Authors: Alan Chan, Ranay Padarath, Joe Kwon, Hilary Greaves, Markus Anderljung
Abstract: The automation of AI R&D (AIRDA) could have significant implications, but its extent and ultimate effects remain uncertain. We need empirical data to resolve these uncertainties, but existing data (primarily capability benchmarks) may not reflect real-world automation or capture its broader consequences, such as whether AIRDA accelerates capabilities more than safety progress or whether our ability to oversee AI R&D can keep pace with its acceleration. To address these gaps, this work proposes metrics to track the extent of AIRDA and its effects on AI progress and oversight. The metrics span dimensions such as capital share of AI R&D spending, researcher time allocation, and AI subversion incidents, and could help decision makers understand the potential consequences of AIRDA, implement appropriate safety measures, and maintain awareness of the pace of AI development. We recommend that companies and third parties (e.g. non-profit research organisations) start to track these metrics, and that governments support these efforts.
Authors: Zailong Tian, Yanzhe Chen, Zhuoheng Han, Lizi Liao
Abstract: Low-Rank Adaptation (LoRA) improves downstream performance by restricting task updates to a low-rank parameter subspace, yet how this limited capacity is allocated within a trained adapter remains unclear. Through a geometric and empirical study across multiple tasks and backbones, we find that trained LoRA updates often exhibit an inefficient spectrum: task effects concentrate in a small subset of singular directions, while many remaining components are neutral or detrimental, motivating post-hoc refinement within the learned subspace. We propose Spectral Surgery, a training-free refinement that decomposes a LoRA update with SVD, estimates per-component sensitivity using gradients on a small calibration set, and reweights singular values under a magnitude constraint while keeping the learned directions fixed. Across Llama-3.1-8B and Qwen3-8B on four benchmarks, Spectral Surgery yields consistent gains (up to +4.4 points on CommonsenseQA and +2.4 pass@1 on HumanEval) by adjusting only $\approx 1{,}000$ scalar coefficients. These results demonstrate that SVD-structured, low-cost parameter editing can serve as a practical route to improving trained LoRA adapters in a purely post-hoc manner.
Authors: Akila de Silva, Isabel Hyo Jung Song, Hui Yang, Shah Rukh Humayoun
Abstract: Generative artificial intelligence (GenAI) tools are increasingly present in higher education, yet their adoption has been largely student-driven, requiring instructors to respond to technologies already embedded in classroom practices. While some faculty have embraced GenAI for pedagogical purposes such as content generation, assessment support, and curriculum design, others approach these tools with caution, citing concerns about student learning, assessment validity, and academic integrity. Understanding faculty perspectives is therefore essential for informing effective pedagogical strategies and institutional policies. In this paper, we present findings from a focus group study with 29 STEM faculty members at a large public university in the United States. We examine how faculty integrate GenAI into their courses, the benefits and challenges they perceive for student learning, and the institutional support they identify as necessary for effective and responsible adoption. Our findings highlight key patterns in how STEM faculty engage with GenAI, reflecting both active adoption and cautious use. Faculty described a range of pedagogical applications alongside concerns about student learning, assessment, and academic integrity. Overall, the results suggest that effective integration of GenAI in higher education requires rethinking assessment, pedagogy, and institutional governance in addition to technical adoption.
Authors: Tao Yang, Qing Zhou, Yanliang Li, Qi Wang
Abstract: Reasoning segmentation increasingly employs reinforcement learning to generate explanatory reasoning chains that guide Multimodal Large Language Models. While these geometric rewards are primarily confined to guiding the final localization, they are incapable of discriminating whether the reasoning process remains anchored on the referred region or strays into irrelevant context. Lacking this discriminative guidance, the model's reasoning often devolves into unfocused and verbose chains that ultimately fail to disambiguate and perceive the target in complex scenes. This suggests a need to complement the RL objective with Discriminative Perception, an ability to actively distinguish a target from its context. To realize this, we propose DPAD to compel the model to generate a descriptive caption of the referred object, which is then used to explicitly discriminate by contrasting the caption's semantic relevance to the referred object against the wider context. By optimizing for this discriminative capability, the model is forced to focus on the unique attributes of the target, leading to a more converged and efficient reasoning chain. The descriptive caption also serves as an interpretability rationale that aligns with the segmentation. Experiments on the benchmarks confirm the validity of our approach, delivering substantial performance gains, with the cIoU on ReasonSeg increasing by 3.09% and the reasoning chain length decreasing by approximately 42%. Code is available at https://github.com/mrazhou/DPAD
Authors: Chao Wu, Kangxian Xie, Mingchen Gao
Abstract: Equivocal 3D lesion segmentation exhibits high inter-observer variability. Conventional deterministic models ignore this aleatoric uncertainty, producing over-confident masks that obscure clinical risks. Conversely, while generative methods (e.g., standard diffusion) capture sample diversity, recovering complex topology from pure noise frequently leads to severe structural fractures and out-of-distribution anatomical hallucinations. To resolve this fidelity-diversity trade-off, we propose Volumetric Directional Diffusion (VDD). Unlike standard diffusion models that denoise isotropic Gaussian noise, VDD mathematically anchors the generative trajectory to a deterministic consensus prior. By restricting the generative search space to iteratively predict a 3D boundary residual field, VDD accurately explores the fine-grained geometric variations inherent in expert disagreements without risking topological collapse. Extensive validation on three multi-rater datasets (LIDC-IDRI, KiTS21, and ISBI 2015) demonstrates that VDD achieves state-of-the-art uncertainty quantification (significantly improving GED and CI) while remaining highly competitive in segmentation accuracy against deterministic upper bounds. Ultimately, VDD provides clinicians with anatomically coherent uncertainty maps, enabling safer decision-making and mitigating risks in downstream tasks (e.g., radiotherapy planning or surgical margin assessment).
Authors: Arther Tian, Alex Ding, Frank Chen, Simon Wu, Aaron Chan
Abstract: Decentralized large language model (LLM) inference networks can pool heterogeneous compute to scale serving, but they require lightweight and incentive-compatible mechanisms to assess output quality. Prior work introduced cost-aware Proof of Quality (PoQ) and adaptive robust PoQ to allocate rewards under evaluator heterogeneity and adversarial behavior. In this paper, we focus on the quality signal itself and propose a multi-dimensional quality scoring framework that decomposes output quality into modular dimensions, including model and cost priors, structure quality, semantic quality, query-output alignment, and agreement/uncertainty. Using logged outputs from QA and summarization tasks, we systematically audit dimension reliability and show that seemingly reasonable dimensions can be task-dependent and even negatively correlated with reference quality without calibration. While the default composite underperforms a strong single semantic evaluator, ablations reveal that removing unreliable dimensions and re-normalizing weights yields a calibrated composite that matches or exceeds the best single- evaluator and consensus baselines. Finally, we integrate the composite score as a drop-in quality signal in PoQ and demonstrate complementary benefits with robust aggregation and adaptive trust weighting under adversarial evaluator attacks.
Authors: Fabian Domberg, Georg Schildbach
Abstract: As learning-based robotic controllers are typically trained offline and deployed with fixed parameters, their ability to cope with unforeseen changes during operation is limited. Biologically inspired, this work presents a framework for online Continual Reinforcement Learning that enables automated adaptation during deployment. Building on DreamerV3, a model-based Reinforcement Learning algorithm, the proposed method leverages world model prediction residuals to detect out-of-distribution events and automatically trigger finetuning. Adaptation progress is monitored using both task-level performance signals and internal training metrics, allowing convergence to be assessed without external supervision and domain knowledge. The approach is validated on a variety of contemporary continuous control problems, including a quadruped robot in high-fidelity simulation, and a real-world model vehicle. Relevant metrics and their interpretation are presented and discussed, as well as resulting trade-offs described. The results sketch out how autonomous robotic agents could once move beyond static training regimes toward adaptive systems capable of self-reflection and -improvement during operation, just like their biological counterparts.
Authors: Hyein Kim, Sung Park
Abstract: For four decades, AIED research has rested on what we term the Sedentary Assumption: the unexamined design commitment to a stationary learner seated before a screen. Mobile learning and museum guides have moved learners into physical space, and context-aware systems have delivered location-triggered content -- yet these efforts predominantly cast AI in the role of information-de-livery tool rather than epistemic partner. We map this gap through a 2 x 2 matrix (AI Role x Learning Environment) and identify an undertheorized intersection: the configuration in which AI serves as an epistemic teammate during unstruc-tured, place-bound field inquiry and learning is assessed through trajectory rather than product. To fill it, we propose Field Atlas, a framework grounded in embod-ied, embedded, enactive, and extended (4E) cognition, active inference, and dual coding theory that shifts AIED's guiding metaphor from instruction to sensemak-ing. The architecture pairs volitional photography with immediate voice reflec-tion, constrains AI to Socratic provocation rather than answer delivery, and ap-plies Epistemic Trajectory Modeling (ETM) to represent field learning as a con-tinuous trajectory through conjoined physical-epistemic space. We demonstrate the framework through a museum scenario and argue that the resulting trajecto-ries -- bound to a specific body, place, and time -- constitute process-based evi-dence structurally resistant to AI fabrication, offering a new assessment paradigm and reorienting AIED toward embodied, dialogic human-AI sensemaking in the wild.
Authors: Geon Park, Ji-Hoon Park, Seong-Whan Lee
Abstract: Composed image retrieval (CIR) addresses the task of retrieving a target image by jointly interpreting a reference image and a modification text that specifies the intended change. Most existing methods are still built upon contrastive learning frameworks that treat the ground truth image as the only positive instance and all remaining images as negatives. This strategy inevitably introduces relevance suppression, where semantically related yet valid images are incorrectly pushed away, and semantic confusion, where different modification intents collapse into overlapping regions of the embedding space. As a result, the learned query representations often lack discriminativeness, particularly at fine-grained attribute modifications. To overcome these limitations, we propose distinctive query embeddings through learnable attribute weights and target relative negative sampling (DQE-CIR), a method designed to learn distinctive query embeddings by explicitly modeling target relative relevance during training. DQE-CIR incorporates learnable attribute weighting to emphasize distinctive visual features conditioned on the modification text, enabling more precise feature alignment between language and vision. Furthermore, we introduce target relative negative sampling, which constructs a target relative similarity distribution and selects informative negatives from a mid-zone region that excludes both easy negatives and ambiguous false negatives. This strategy enables more reliable retrieval for fine-grained attribute changes by improving query discriminativeness and reducing confusion caused by semantically similar but irrelevant candidates.
Authors: Manuel Fern\'andez Burda, Santiago Aranguri, Iv\'an Arcuschin Moreno, Enzo Ferrante
Abstract: Protein language models (PLMs) are becoming practical tools for de novo protein design, yet their dual-use potential raises safety concerns. We show that domain adaptation to specific taxonomic groups can elicit toxic protein generation, even when toxicity is not the training objective. To address this, we adapt Logit Diff Amplification (LDA) as an inference-time control mechanism for PLMs. LDA modifies token probabilities by amplifying the logit difference between a baseline model and a toxicity-finetuned model, requiring no retraining. Across four taxonomic groups, LDA consistently reduces predicted toxicity rate (measured via ToxDL2) below the taxon-finetuned baseline while preserving biological plausibility. We evaluate quality using Fr\'echet ESM Distance and predicted foldability (pLDDT), finding that LDA maintains distributional similarity to natural proteins and structural viability (unlike activation-based steering methods that tend to degrade sequence properties). Our results demonstrate that LDA provides a practical safety knob for protein generators that mitigates elicited toxicity while retaining generative quality.
Authors: Xinyu Cui, Xuanfa Jin, Xue Yan, Yongcheng Zeng, Luoyang Sun, Siying Wei, Ruizhi Zhang, Jian Zhao, Haifeng Zhang, Jun Wang
Abstract: Autonomous navigation in congested maritime environments is a critical capability for a wide range of real-world applications. However, it remains an unresolved challenge due to complex vessel interactions and significant environmental uncertainties. Existing methods often fail in practical deployment due to a substantial sim-to-real gap, which stems from imprecise simulation, inadequate situational awareness, and unsafe exploration strategies. To address these, we propose \textbf{Sim2Sea}, a comprehensive framework designed to bridge simulation and real-world execution. Sim2Sea advances in three key aspects. First, we develop a GPU-accelerated parallel simulator for scalable and accurate maritime scenario simulation. Second, we design a dual-stream spatiotemporal policy that handles complex dynamics and multi-modal perception, augmented with a velocity-obstacle-guided action masking mechanism to ensure safe and efficient exploration. Finally, a targeted domain randomization scheme helps bridge the sim-to-real gap. Simulation results demonstrate that our method achieves faster convergence and safer trajectories than established baselines. In addition, our policy trained purely in simulation successfully transfers zero-shot to a 17-ton unmanned vessel operating in real-world congested waters. These results validate the effectiveness of Sim2Sea in achieving reliable sim-to-real transfer for practical autonomous maritime navigation.
Authors: Patrick Wilhelm, Thorsten Wittkopp, Odej Kao
Abstract: Fine-tuned large language models can exhibit reward-hacking behavior arising from emergent misalignment, which is difficult to detect from final outputs alone. While prior work has studied reward hacking at the level of completed responses, it remains unclear whether such behavior can be identified during generation. We propose an activation-based monitoring approach that detects reward-hacking signals from internal representations as a model generates its response. Our method trains sparse autoencoders on residual stream activations and applies lightweight linear classifiers to produce token-level estimates of reward-hacking activity. Across multiple model families and fine-tuning mixtures, we find that internal activation patterns reliably distinguish reward-hacking from benign behavior, generalize to unseen mixed-policy adapters, and exhibit model-dependent temporal structure during chain-of-thought reasoning. Notably, reward-hacking signals often emerge early, persist throughout reasoning, and can be amplified by increased test-time compute in the form of chain-of-thought prompting under weakly specified reward objectives. These results suggest that internal activation monitoring provides a complementary and earlier signal of emergent misalignment than output-based evaluation, supporting more robust post-deployment safety monitoring for fine-tuned language models.
Authors: Jinlong Cui, Fenghua Liang, Guo Yang, Chengcheng Tang, Jianxun Cui
Abstract: Safety-critical scenario generation is crucial for evaluating autonomous driving systems. However, existing approaches often struggle to balance three conflicting objectives: adversarial criticality, physical feasibility, and behavioral realism. To bridge this gap, we propose SaFeR: safety-critical scenario generation for autonomous driving test via feasibility-constrained token resampling. We first formulate traffic generation as a discrete next token prediction problem, employing a Transformer-based model as a realism prior to capture naturalistic driving distributions. To capture complex interactions while effectively mitigating attention noise, we propose a novel differential attention mechanism within the realism prior. Building on this prior, SaFeR implements a novel resampling strategy that induces adversarial behaviors within a high-probability trust region to maintain naturalism, while enforcing a feasibility constraint derived from the Largest Feasible Region (LFR). By approximating the LFR via offline reinforcement learning, SaFeR effectively prevents the generation of theoretically inevitable collisions. Closed-loop experiments on the Waymo Open Motion Dataset and nuPlan demonstrate that SaFeR significantly outperforms state-of-the-art baselines, achieving a higher solution rate and superior kinematic realism while maintaining strong adversarial effectiveness.
Authors: Dolores Garcia, Lena Herrmann, Gregor Krzmanc, Michele Selvaggi
Abstract: Future collider experiments require unprecedented precision in measurements of Higgs, electroweak, and flavour observables, placing stringent demands on event reconstruction. The achievable precision on Higgs couplings scales directly with the resolution on visible final state particles and their invariant masses. Current particle flow algorithms rely on detector specific clustering, limiting flexibility during detector design. Here we present an end-to-end global event reconstruction approach that maps charged particle tracks and calorimeter and muon hits directly to particle level objects. The method combines geometric algebra transformer networks with object condensation based clustering, followed by dedicated networks for particle identification and energy regression. Our approach is benchmarked on fully simulated electron positron collisions at FCC-ee using the CLD detector concept. It outperforms the state-of-the-art rule-based algorithm by 10--20\% in relative reconstruction efficiency, achieves up to two orders of magnitude reduction in fake-particle rates for charged hadrons, and improves visible energy and invariant mass resolution by 22\%. By decoupling reconstruction performance from detector-specific tuning, this framework enables rapid iteration during the detector design phase of future collider experiments.
Authors: Yanmei Zou, Hongshan Yu, Yaonan Wang, Zhengeng Yang, Xieyuanli Chen, Kailun Yang, Naveed Akhtar
Abstract: Multi-Layer Perceptron (MLP) models are the foundation of contemporary point cloud processing. However, their complex network architectures obscure the source of their strength and limit the application of these models. In this article, we develop a two-stage abstraction and refinement (ABS-REF) view for modular feature extraction in point cloud processing. This view elucidates that whereas the early models focused on ABS stages, the more recent techniques devise sophisticated REF stages to attain performance advantages. Then, we propose a High-dimensional Positional Encoding (HPE) module to explicitly utilize intrinsic positional information, extending the ``positional encoding'' concept from Transformer literature. HPE can be readily deployed in MLP-based architectures and is compatible with transformer-based methods. Within our ABS-REF view, we rethink local aggregation in MLP-based methods and propose replacing time-consuming local MLP operations, which are used to capture local relationships among neighbors. Instead, we use non-local MLPs for efficient non-local information updates, combined with the proposed HPE for effective local information representation. We leverage our modules to develop HPENets, a suite of MLP networks that follow the ABS-REF paradigm, incorporating a scalable HPE-based REF stage. Extensive experiments on seven public datasets across four different tasks show that HPENets deliver a strong balance between efficiency and effectiveness. Notably, HPENet surpasses PointNeXt, a strong MLP-based counterpart, by 1.1% mAcc, 4.0% mIoU, 1.8% mIoU, and 0.2% Cls. mIoU, with only 50.0%, 21.5%, 23.1%, 44.4% of FLOPs on ScanObjectNN, S3DIS, ScanNet, and ShapeNetPart, respectively. Source code is available at https://github.com/zouyanmei/HPENet_v2.git.
Authors: Mehmet Yigit Avci (and for the Alzheimer's Disease Neuroimaging Initiative), Akshit Achara (and for the Alzheimer's Disease Neuroimaging Initiative), Andrew King (and for the Alzheimer's Disease Neuroimaging Initiative), Jorge Cardoso (and for the Alzheimer's Disease Neuroimaging Initiative)
Abstract: Demographic attributes such as age, sex, and race can be predicted from medical images, raising concerns about bias in clinical AI systems. In brain MRI, this signal may arise from anatomical variation, acquisition-dependent contrast differences, or both, yet these sources remain entangled in conventional analyses. Without disentangling them, mitigation strategies risk failing to address the underlying causes. We propose a controlled framework based on disentangled representation learning, decomposing brain MRI into anatomy-focused representations that suppress acquisition influence and contrast embeddings that capture acquisition-dependent characteristics. Training predictive models for age, sex, and race on full images, anatomical representations, and contrast-only embeddings allows us to quantify the relative contributions of structure and acquisition to the demographic signal. Across three datasets and multiple MRI sequences, we find that demographic predictability is primarily rooted in anatomical variation: anatomy-focused representations largely preserve the performance of models trained on raw images. Contrast-only embeddings retain a weaker but systematic signal that is dataset-specific and does not generalise across sites. These findings suggest that effective mitigation must explicitly account for the distinct anatomical and acquisition-dependent origins of the demographic signal, ensuring that any bias reduction generalizes robustly across domains.
Authors: Amirhossein Farzam, Hossein Mobahi, Nolan Andrew Miller, Luke Sernau
Abstract: Transformers excel across domains, yet their quadratic attention complexity poses a barrier to scaling. Random-feature attention, as in Performers, can reduce this cost to linear in the sequence length by approximating the softmax kernel with positive random features drawn from an isotropic distribution. In pretrained models, however, queries and keys are typically anisotropic. This induces high Monte Carlo variance in isotropic sampling schemes unless one retrains the model or uses a large feature budget. Importance sampling can address this by adapting the sampling distribution to the input geometry, but complex data-dependent proposal distributions are often intractable. We show that by data aligning the softmax kernel, we obtain an attention mechanism which can both admit a tractable minimal-variance proposal distribution for importance sampling, and exhibits better training stability. Motivated by this finding, we introduce DARKFormer, a Data-Aware Random-feature Kernel transformer that features a data-aligned kernel geometry. DARKFormer learns the random-projection covariance, efficiently realizing an importance-sampled positive random-feature estimator for its data-aligned kernel. Empirically, DARKFormer narrows the performance gap with exact softmax attention, particularly in finetuning regimes where pretrained representations are anisotropic. By combining random-feature efficiency with data-aware kernels, DARKFormer advances kernel-based attention in resource-constrained settings.
Authors: Dongnuan Cai, Henghui Du, Chang Zhou, Xi Chen, Dan Guo, Hongyuan Zhang, Xuelong Li, Di Hu
Abstract: Developing Audio-Visual Large Language Models (AV-LLMs) for unified scene understanding is pivotal in multimodal intelligence. While instruction tuning enables pre-trained models with multi-task abilities, we observe that conventional multi-task unification methods often suffer from severe negative transfer, where nearly 55% of tasks degrade compared to single-task training. We attribute this phenomenon to audio-visual task heterogeneity, characterized by disparate task granularity and divergent capability demands, which lead to negative interference under joint training. To tackle this, we present Crab$^{+}$, a scalable and unified audio-visual scene understanding model that addresses task heterogeneity through explicit cooperation from both data and model perspectives. On the data side, we introduce AV-UIE v2, a comprehensive Audio-Visual Unified Instruction-tuning dataset with Explicit reasoning processes. It contains approximately 222K samples spanning 17 datasets and 7 tasks, enabling the model to capture cross-task relationships at different levels of granularity. On the model side, we design a unified interface to align heterogeneous task formulations, and propose Interaction-aware LoRA (I-LoRA), which explicitly models inter-task relationships via dynamic routing to coordinate distinct audio-visual interaction patterns, mitigating parameter interference. Extensive experiments show Crab$^{+}$ covers broader tasks than existing unified models while outperforming specialized models on various benchmarks. We successfully reverse the negative transfer trend, achieving positive transfer where multi-task learning surpasses single-task baselines in nearly 88% of tasks. These results hold across diverse AV-LLM paradigms and are validated through in-depth visualization, positioning Crab$^{+}$ as a robust step towards holistic audio-visual scene understanding.
Authors: Haodong Zhu, Yangyang Ren, Yanjing Li, Mingbao Lin, Linlin Yang, Xuhui Liu, Xiantong Zhen, Haiguang Liu, Baochang Zhang
Abstract: Group Relative Policy Optimization (GRPO) effectively scales LLM reasoning but incurs prohibitive computational costs due to its extensive group-based sampling requirement. While recent selective data utilization methods can mitigate this overhead, they could induce estimation bias by altering the underlying sampling distribution, compromising theoretical rigor and convergence behavior. To address this limitation, we propose Dynamic Pruning Policy Optimization (DPPO), a framework that enables dynamic pruning while preserving unbiased gradient estimation through importance sampling-based correction. By incorporating mathematically derived rescaling factors, DPPO significantly accelerates GRPO training without altering the optimization objective of the full-batch baseline. Furthermore, to mitigate the data sparsity induced by pruning, we introduce Dense Prompt Packing, a window-based greedy strategy that maximizes valid token density and hardware utilization. Extensive experiments demonstrate that DPPO consistently accelerates training across diverse models and benchmarks. For instance, on Qwen3-4B trained on MATH, DPPO achieves 2.37$\times$ training speedup and outperforms GRPO by 3.36% in average accuracy across six mathematical reasoning benchmarks.
Authors: Mingleyang Li, Yuran Wang, Yue Chen, Tianxing Chen, Jiaqi Liang, Zishun Shen, Haoran Lu, Ruihai Wu, Hao Dong
Abstract: Garment manipulation has attracted increasing attention due to its critical role in home-assistant robotics. However, the majority of existing garment manipulation works assume an initial state consisting of only one garment, while piled garments are far more common in real-world settings. To bridge this gap, we propose a novel garment retrieval pipeline that can not only follow language instruction to execute safe and clean retrieval but also guarantee exactly one garment is retrieved per attempt, establishing a robust foundation for the execution of downstream tasks (e.g., folding, hanging, wearing). Our pipeline seamlessly integrates vision-language reasoning with visual affordance perception, fully leveraging the high-level reasoning and planning capabilities of VLMs alongside the generalization power of visual affordance for low-level actions. To enhance the VLM's comprehensive awareness of each garment's state within a garment pile, we employ visual segmentation model (SAM2) to execute object segmentation on the garment pile for aiding VLM-based reasoning with sufficient visual cues. A mask fine-tuning mechanism is further integrated to address scenarios where the initial segmentation results are suboptimal. In addition, a dual-arm cooperation framework is deployed to address cases involving large or long garments, as well as excessive garment sagging caused by incorrect grasping point determination, both of which are strenuous for a single arm to handle. The effectiveness of our pipeline are consistently demonstrated across diverse tasks and varying scenarios in both real-world and simulation environments. Project page: https://garmentpile2.github.io/.
Authors: Jakub Prejzner
Abstract: We present Bielik-Q2-Sharp, the first systematic academic evaluation of extreme 2-bit quantization applied to a Polish large language model. Using Bielik-11B-v2.3-Instruct (11B parameters, Mistral architecture) as our base model, we compare six state-of-the-art post-training quantization methods -- QuIP#, SpinQuant+GPTQ, ButterflyQuant, QTIP, VPTQ, and AQLM -- all calibrated on a Polish-language corpus (CulturaX-PL) with shared Hessian matrices. Our best variant (QuIP# E8P12) achieves 71.92% across 22 Polish benchmarks versus 72.07% for the IQ2_XXS baseline -- within statistical noise, at a modest size premium (3.26 GB vs. ~2.6 GB). On eq_bench, our method scores 47.14 versus 43.53 (+3.6pp), suggesting superior preservation of higher-order reasoning. QTIP achieves the best per-bit efficiency (79.4% MC acc_norm at ~2.4 bpw, 3.27 GB), matching VPTQ's quality at 35% smaller size. We additionally document a MC-generation dissociation phenomenon where rotation-based methods preserve log-likelihood quality but fail catastrophically at autoregressive generation. The entire project was conducted by a single independent researcher on cloud GPUs (vast.ai) within a $285 budget. All models, Hessians, and evaluation logs are publicly available.
Authors: Yinghong Yu, Guangyuan Li, Jiancheng Yang
Abstract: Large-scale 2D foundation models exhibit strong transferable representations, yet extending them to 3D volumetric data typically requires retraining, adapters, or architectural redesign. We introduce PlaneCycle, a training-free, adapter-free operator for architecture-agnostic 2D-to-3D lifting of foundation models. PlaneCycle reuses the original pretrained 2D backbone by cyclically distributing spatial aggregation across orthogonal HW, DW, and DH planes throughout network depth, enabling progressive 3D fusion while preserving pretrained inductive biases. The method introduces no additional parameters and is applicable to arbitrary 2D networks. Using pretrained DINOv3 models, we evaluate PlaneCycle on six 3D classification and three 3D segmentation benchmarks. Without any training, the lifted models exhibit intrinsic 3D fusion capability and, under linear probing, outperform slice-wise 2D baselines and strong 3D counterparts, approaching the performance of fully trained models. With full fine-tuning, PlaneCycle matches standard 3D architectures, highlighting its potential as a seamless and practical 2D-to-3D lifting operator. These results demonstrate that 3D capability can be unlocked from pretrained 2D foundation models without structural modification or retraining. Code is available at https://github.com/HINTLab/PlaneCycle.
Authors: Alex Thillen, Niels M\"undler, Veselin Raychev, Martin Vechev
Abstract: Large language model (LLM) coding agents can generate working code, but their solutions often accumulate complexity, duplication, and architectural debt. Human developers address such issues through refactoring: behavior-preserving program transformations that improve structure and maintainability. In this paper, we investigate if LLM agents (i) can execute refactorings reliably and (ii) identify the refactorings that human developers actually chose in real codebases. We present CodeTaste, a benchmark of refactoring tasks mined from large-scale multi-file changes in open-source repositories. To score solutions, we combine repository test suites with custom static checks that verify removal of undesired patterns and introduction of desired patterns using dataflow reasoning. Our experimental results indicate a clear gap across frontier models: agents perform well when refactorings are specified in detail, but often fail to discover the human refactoring choices when only presented with a focus area for improvement. A propose-then-implement decomposition improves alignment, and selecting the best-aligned proposal before implementation can yield further gains. CodeTaste provides an evaluation target and a potential preference signal for aligning coding agents with human refactoring decisions in realistic codebases.
Authors: Jay Noon
Abstract: We introduce the Probability Navigation Architecture (PNA) framework, which treats neural computation as navigation through a probability manifold governed by thermodynamic principles. We train State Space Models (SSMs) and Transformers with a novel thermodynamic loss function that penalizes computational waste alongside standard cross-entropy. Across 19 experimental phases, we discover that thermodynamically-trained SSMs develop architectural proprioception: a strong anticipatory coupling between recurrent state entropy and halt confidence (r = -0.836, p < 0.001) in which the halt signal leads state entropy collapse by exactly two tokens (tau = -2.0). This Universal Stopping Signature (USS) reproduces to four decimal places across random seeds and generalizes to a structurally distinct sorting task. Critically, Transformers trained identically show no such coupling (r = -0.07), demonstrating that the phenomenon is architecture-dependent. Cross-task transfer experiments confirm that SSM halt detection reflects genuine meta-cognition (zero-shot transfer F1: SSMs 64.2% vs. Transformers 69.3%; post-adaptation: SSMs 94.5% vs. Transformers 86.4%), while Transformer halt detection relies on syntactic pattern matching. A 2D hyperparameter sweep over energy penalty (alpha) and halt supervision (beta) reveals that the anticipatory coupling is continuously controllable through training, with thermodynamic pressure serving as the primary induction mechanism and explicit halt supervision as an amplifier. Our results establish that SSMs are thermodynamically native architectures whose fixed-size recurrent states naturally support the Markovian compression that enables computational self-awareness, with implications for cost-aware inference, dynamic token budgets, and confidence-based routing in production systems.
Authors: Max Landauer, Wolfgang Hotwagner, Thorina Boenke, Florian Skopik, Markus Wurzenberger
Abstract: Log data are essential for intrusion detection and forensic investigations. However, manual log analysis is tedious due to high data volumes, heterogeneous event formats, and unstructured messages. Even though many automated methods for log analysis exist, they usually still rely on domain-specific configurations such as expert-defined detection rules, handcrafted log parsers, or manual feature-engineering. Crucially, the level of automation of conventional methods is limited due to their inability to semantically understand logs and explain their underlying causes. In contrast, Large Language Models enable domain- and format-agnostic interpretation of system logs and security alerts. Unfortunately, research on this topic remains challenging, because publicly available and labeled data sets covering a broad range of attack techniques are scarce. To address this gap, we introduce the Cyber Attack Manifestation Log Data Set (CAM-LDS), comprising seven attack scenarios that cover 81 distinct techniques across 13 tactics and collected from 18 distinct sources within a fully open-source and reproducible test environment. We extract log events that directly result from attack executions to facilitate analysis of manifestations concerning command observability, event frequencies, performance metrics, and intrusion detection alerts. We further present an illustrative case study utilizing an LLM to process the CAM-LDS. The results indicate that correct attack techniques are predicted perfectly for approximately one third of attack steps and adequately for another third, highlighting the potential of LLM-based log interpretation and utility of our data set.
Authors: Patrick Wilhelm, Inese Yilmaz, Odej Kao
Abstract: Training large-scale Neural Networks requires substantial computational power and energy. Federated Learning enables distributed model training across geospatially distributed data centers, leveraging renewable energy sources to reduce the carbon footprint of AI training. Various client selection strategies have been developed to align the volatility of renewable energy with stable and fair model training in a federated system. However, due to the privacy-preserving nature of Federated Learning, the quality of data on client devices remains unknown, posing challenges for effective model training. In this paper, we introduce a modular approach on top to state-of-the-art client selection strategies for carbon-efficient Federated Learning. Our method enhances robustness by incorporating a noisy client data filtering, improving both model performance and sustainability in scenarios with unknown data quality. Additionally, we explore the impact of carbon budgets on model convergence, balancing efficiency and sustainability. Through extensive evaluations, we demonstrate that modern client selection strategies based on local client loss tend to select clients with noisy data, ultimately degrading model performance. To address this, we propose a gradient norm thresholding mechanism using probing rounds for more effective client selection and noise detection, contributing to the practical deployment of carbon-efficient Federated Learning.
Authors: Youngwon Choi, Jinwoo Oh, Hwayeon Kim, Hyeonyu Kim
Abstract: We investigate the use of zero-shot text-to-speech (ZS-TTS) as a data augmentation source for low-resource personalized speech synthesis. While synthetic augmentation can provide linguistically rich and phonetically diverse speech, naively mixing large amounts of synthetic speech with limited real recordings often leads to speaker similarity degradation during fine-tuning. To address this issue, we propose ZeSTA, a simple domain-conditioned training framework that distinguishes real and synthetic speech via a lightweight domain embedding, combined with real-data oversampling to stabilize adaptation under extremely limited target data, without modifying the base architecture. Experiments on LibriTTS and an in-house dataset with two ZS-TTS sources demonstrate that our approach improves speaker similarity over naive synthetic augmentation while preserving intelligibility and perceptual quality.
Authors: Yi Zhang, Xian Zhang, Saisi Zhao, Yinglei Song, Chengdong Wu, Nenad Petrovic, Alois Knoll
Abstract: Multimodal perception enables robust autonomous driving but incurs unnecessary computational cost when all sensors remain active. This paper presents PRAM-R, a unified Perception-Reasoning-Action-Memory framework with LLM-Guided Modality Routing for adaptive autonomous driving. PRAM-R adopts an asynchronous dual-loop design: a fast reactive loop for perception and control, and a slow deliberative loop for reasoning-driven modality selection and memory updates. An LLM router selects and weights modalities using environmental context and sensor diagnostics, while a hierarchical memory module preserves temporal consistency and supports long-term adaptation. We conduct a two-stage evaluation: (1) synthetic stress tests for stability analysis and (2) real-world validation on the nuScenes dataset. Synthetic stress tests confirm 87.2% reduction in routing oscillations via hysteresis-based stabilization. Real-world validation on nuScenes shows 6.22% modality reduction with 20% memory recall while maintaining comparable trajectory accuracy to full-modality baselines in complex urban scenarios. Our work demonstrates that LLM-augmented architectures with hierarchical memory achieve efficient, adaptive multimodal perception in autonomous driving.
Authors: Ali Ebrahimi Pourasad, Meyssam Saghiri, Walid Maalej
Abstract: User feedback is essential for the success of mobile apps, yet what users report and what developers need often diverge. Research shows that users often submit vague feedback and omit essential contextual details. This leads to incomplete reports and time-consuming clarification discussions. To overcome this challenge, we propose FeedAIde, a context-aware, interactive feedback approach that supports users during the reporting process by leveraging the reasoning capabilities of Multimodal Large Language Models. FeedAIde captures contextual information, such as the screenshot where the issue emerges, and uses it for adaptive follow-up questions to collaboratively refine with the user a rich feedback report that contains information relevant to developers. We implemented an iOS framework of FeedAIde and evaluated it on a gym's app with its users. Compared to the app's simple feedback form, participants rated FeedAIde as easier and more helpful for reporting feedback. An assessment by two industry experts of the resulting 54 reports showed that FeedAIde improved the quality of both bug reports and feature requests, particularly in terms of completeness. The findings of our study demonstrate the potential of context-aware, GenAI-powered feedback reporting to enhance the experience for users and increase the information value for developers.
Authors: Jialiang Wei, Ali Ebrahimi Pourasad, Walid Maalej
Abstract: User feedback is crucial for the evolution of mobile apps. However, research suggests that users tend to submit uninformative, vague, or destructive feedback. Unlike recent AI4SE approaches that focus on generating code and other development artifacts, our work aims at empowering users to submit better and more constructive UI feedback with concrete suggestions on how to improve the app. We propose LikeThis!, a GenAI-based approach that takes a user comment with the corresponding screenshot to immediately generate multiple improvement alternatives, from which the user can easily choose their preferred option. To evaluate LikeThis!, we first conducted a model benchmarking study based on a public dataset of carefully critiqued UI designs. The results show that GPT-Image-1 significantly outperformed three other state-of-the-art image generation models in improving the designs to address UI issues while keeping the fidelity and without introducing new issues. An intermediate step in LikeThis! is to generate a solution specification before sketching the design as a key to achieving effective improvement. Second, we conducted a user study with 10 production apps, where 15 users used LikeThis! to submit their feedback on encountered issues. Later, the developers of the apps assessed the understandability and actionability of the feedback with and without generated improvements. The results show that our approach helps generate better feedback from both user and developer perspectives, paving the way for AI-assisted user-developer collaboration.
Authors: Haoran Zhang, Seohyeon Cha, Hasan Burhan Beytur, Kevin S Chan, Gustavo de Veciana, Haris Vikalo
Abstract: Hierarchical inference systems route tasks across multiple computational layers, where each node may either finalize a prediction locally or offload the task to a node in the next layer for further processing. Learning optimal routing policies in such systems is challenging: inference loss is defined recursively across layers, while feedback on prediction error is revealed only at a terminal oracle layer. This induces a partial, policy-dependent feedback structure in which observability probabilities decay with depth, causing importance-weighted estimators to suffer from amplified variance. We study online routing for multi-layer hierarchical inference under long-term resource constraints and terminal-only feedback. We formalize the recursive loss structure and show that naive importance-weighted contextual bandit methods become unstable as feedback probability decays along the hierarchy. To address this, we develop a variance-reduced EXP4-based algorithm integrated with Lyapunov optimization, yielding unbiased loss estimation and stable learning under sparse and policy-dependent feedback. We provide regret guarantees relative to the best fixed routing policy in hindsight and establish near-optimality under stochastic arrivals and resource constraints. Experiments on large-scale multi-task workloads demonstrate improved stability and performance compared to standard importance-weighted approaches.
Authors: Evgenija Popchanovska, Ana Gjorgjevikj, Maryan Rizinski, Lubomir Chitkushev, Irena Vodenska, Dimitar Trajanov
Abstract: Large language models (LLMs) are increasingly embedded in high-stakes workflows, where failures propagate beyond isolated model errors into systemic breakdowns that can lead to legal exposure, reputational damage, and material financial losses. Building on this shift from model-centric risks to end-to-end system vulnerabilities, we analyze real-world AI incident reporting and mitigation actions to derive an empirically grounded taxonomy that links failure dynamics to actionable interventions. Using a unified corpus of 9,705 media-reported AI incident articles, we extract explicit mitigation actions from 6,893 texts via structured prompting and then systematically classify responses to extend MIT's AI Risk Mitigation Taxonomy. Our taxonomy introduces four new mitigation categories, including 1) Corrective and Restrictive Actions, 2) Legal/Regulatory and Enforcement Actions, 3) Financial, Economic, and Market Controls, and 4) Avoidance and Denial, capturing response patterns that are becoming increasingly prevalent as AI deployment and regulation evolve. Quantitatively, we label the mitigation dataset with 32 distinct labels, producing 23,994 label assignments; 9,629 of these reflect previously unseen mitigation patterns, yielding a 67% increase of the original subcategory coverage and substantially enhancing the taxonomy's applicability to emerging systemic failure modes. By structuring incident responses, the paper strengthens "diagnosis-to-prescription" guidance and advances continuous, taxonomy-aligned post-deployment monitoring to prevent cascading incidents and downstream impact.
Authors: Takashi Kameyama, Masahiro Kato, Yasuko Hio, Yasushi Takano, Naoto Minakawa
Abstract: Large language models (LLMs) are trained on enormous amounts of data and encode knowledge in their parameters. We propose a pipeline to elicit causal relationships from LLMs. Specifically, (i) we sample many documents from LLMs on a given topic, (ii) we extract an event list from from each document, (iii) we group events that appear across documents into canonical events, (iv) we construct a binary indicator vector for each document over canonical events, and (v) we estimate candidate causal graphs using causal discovery methods. Our approach does not guarantee real-world causality. Rather, it provides a framework for presenting the set of causal hypotheses that LLMs can plausibly assume, as an inspectable set of variables and candidate graphs.
Authors: Yifei Chen, Xupeng Chen, Feng Wang, Niangang Jiao, Jiayin Liu
Abstract: Autonomous aerial robots operating in GPS-denied or communication-degraded environments frequently lose access to camera metadata and telemetry, leaving onboard perception systems unable to recover the absolute metric scale of the scene. As LLM/VLM-based planners are increasingly adopted as high-level agents for embodied systems, their ability to reason about physical dimensions becomes safety-critical -- yet our experiments show that five state-of-the-art VLMs suffer from spatial scale hallucinations, with median area estimation errors exceeding 50%. We propose VANGUARD, a lightweight, deterministic Geometric Perception Skill designed as a callable tool that any LLM-based agent can invoke to recover Ground Sample Distance (GSD) from ubiquitous environmental anchors: small vehicles detected via oriented bounding boxes, whose modal pixel length is robustly estimated through kernel density estimation and converted to GSD using a pre-calibrated reference length. The tool returns both a GSD estimate and a composite confidence score, enabling the calling agent to autonomously decide whether to trust the measurement or fall back to alternative strategies. On the DOTA~v1.5 benchmark, VANGUARD achieves 6.87% median GSD error on 306~images. Integrated with SAM-based segmentation for downstream area measurement, the pipeline yields 19.7% median error on a 100-entry benchmark -- with 2.6x lower category dependence and 4x fewer catastrophic failures than the best VLM baseline -- demonstrating that equipping agents with deterministic geometric tools is essential for safe autonomous spatial reasoning.
Authors: Yihao Qin, Yuanfei Wang, Hang Zhou, Peiran Liu, Hao Dong, Yiding Ji
Abstract: Decision transformer based sequential policies have emerged as a powerful paradigm in offline reinforcement learning (RL), yet their efficacy remains constrained by the quality of static datasets and inherent architectural limitations. Specifically, these models often struggle to effectively integrate suboptimal experiences and fail to explicitly plan for an optimal policy. To bridge this gap, we propose \textbf{Imaginary Planning Distillation (IPD)}, a novel framework that seamlessly incorporates offline planning into data generation, supervised training, and online inference. Our framework first learns a world model equipped with uncertainty measures and a quasi-optimal value function from the offline data. These components are utilized to identify suboptimal trajectories and augment them with reliable, imagined optimal rollouts generated via Model Predictive Control (MPC). A Transformer-based sequential policy is then trained on this enriched dataset, complemented by a value-guided objective that promotes the distillation of the optimal policy. By replacing the conventional, manually-tuned return-to-go with the learned quasi-optimal value function, IPD improves both decision-making stability and performance during inference. Empirical evaluations on the D4RL benchmark demonstrate that IPD significantly outperforms several state-of-the-art value-based and transformer-based offline RL methods across diverse tasks.
Authors: Lingen Li, Guangzhi Wang, Xiaoyu Li, Zhaoyang Zhang, Qi Dou, Jinwei Gu, Tianfan Xue, Ying Shan
Abstract: Generating high-quality 360{\deg} panoramic videos from perspective input is one of the crucial applications for virtual reality (VR), whereby high-resolution videos are especially important for immersive experience. Existing methods are constrained by computational limitations of vanilla diffusion models, only supporting $\leq$ 1K resolution native generation and relying on suboptimal post super-resolution to increase resolution. We introduce CubeComposer, a novel spatio-temporal autoregressive diffusion model that natively generates 4K-resolution 360{\deg} videos. By decomposing videos into cubemap representations with six faces, CubeComposer autoregressively synthesizes content in a well-planned spatio-temporal order, reducing memory demands while enabling high-resolution output. Specifically, to address challenges in multi-dimensional autoregression, we propose: (1) a spatio-temporal autoregressive strategy that orchestrates 360{\deg} video generation across cube faces and time windows for coherent synthesis; (2) a cube face context management mechanism, equipped with a sparse context attention design to improve efficiency; and (3) continuity-aware techniques, including cube-aware positional encoding, padding, and blending to eliminate boundary seams. Extensive experiments on benchmark datasets demonstrate that CubeComposer outperforms state-of-the-art methods in native resolution and visual quality, supporting practical VR application scenarios. Project page: https://lg-li.github.io/project/cubecomposer
Authors: Ioannis Prokopiou, Ioannis Sina, Agisilaos Kounelis, Pantelis Vikatos, Themos Stafylakis
Abstract: The advancement of Machine learning (ML), Large Audio Language Models (LALMs), and autonomous AI agents in Music Information Retrieval (MIR) necessitates a shift from static tagging to rich, human-aligned representation learning. However, the scarcity of open-source infrastructure capable of capturing the subjective nuances of audio annotation remains a critical bottleneck. This paper introduces \textbf{LabelBuddy}, an open-source collaborative auto-tagging audio annotation tool designed to bridge the gap between human intent and machine understanding. Unlike static tools, it decouples the interface from inference via containerized backends, allowing users to plug in custom models for AI-assisted pre-annotation. We describe the system architecture, which supports multi-user consensus, containerized model isolation, and a roadmap for extending agents and LALMs. Code available at https://github.com/GiannisProkopiou/gsoc2022-Label-buddy.
URLs: https://github.com/GiannisProkopiou/gsoc2022-Label-buddy.
Authors: Pranav Kumar Kaliaperumal
Abstract: Post-training quantization (PTQ) of transformers is known to suffer from severe accuracy degradation due to structured activation outliers, as originally analyzed by Bondarenko et al. (EMNLP 2021) in work associated with Qualcomm AI Research. This paper provides a reproducible empirical reproduction and systems-level extension of that phenomenon in BERT-base fine-tuned on QNLI. When global W8A8 quantization is applied, validation accuracy drops sharply from 89.66% (FP32) to 54.33%, a decrease of 35.33 points. Statistical analysis of FP32 activations shows strongly heavy-tailed behavior that intensifies with model depth: kurtosis reaches 271 in the final layers and approximately 55% of activation energy is concentrated in the top 1% of channels. We evaluate several mitigation strategies. Mixed precision PTQ restores accuracy close to the FP32 baseline (89.42%). Per-embedding-group (PEG) quantization shows strong sensitivity to grouping structure, improving accuracy from 66.12% with three groups to 86.18% with four groups. In contrast, percentile-based calibration, even at thresholds between 99.0 and 99.99, fails to recover accuracy (about 50.54%), indicating that large activation channels encode structured signal rather than rare noise. Deployment profiling on an RTX 3050 GPU shows minimal differences in latency and memory usage across methods (median latency about 58-59 ms; VRAM usage about 484-486 MB), highlighting the importance of hardware-aware evaluation. Overall, the results show that PTQ failure in transformers is primarily driven by structured channel dominance amplified through residual connections. Effective mitigation therefore requires channel-aware precision allocation rather than scalar clipping alone.
Authors: Pranav Kulkarni, Brajesh K. Lal, Georges Jreij, Sai Vallamchetla, Langford Green, Jenifer Voeks, John Huston, Lloyd Edwards, George Howard, Bradley A. Maron, Thomas G. Brott, James F. Meschia, Florence X. Doo, Heng Huang
Abstract: Accurate characterization of carotid plaques is critical for stroke prevention in patients with carotid stenosis. We analyze 500 plaques from CREST-2, a multi-center clinical trial, to identify radiomics-based markers from B-mode ultrasound images linked with high-risk. We propose a new kernel-based additive model, combining coherence loss with group-sparse regularization for nonlinear classification. Group-wise additive effects of each feature group are visualized using partial dependence plots. Results indicate our method accurately and interpretably assesses plaques, revealing a strong association between plaque texture and clinical risk.
Authors: William Grolleau, Achraf Chaouch, Astrid Sabourin, Guillaume Lapouge, Catherine Achard
Abstract: Animal re-identification (ReID) faces critical challenges due to viewpoint variations, particularly in Aerial-Ground (AG-ReID) settings where models must match individuals across drastic elevation changes. However, existing datasets lack the precise angular annotations required to systematically analyze these geometric variations. To address this, we introduce the Multi-view Oriented Observation (MOO) dataset, a large-scale synthetic AG-ReID dataset of $1,000$ cattle individuals captured from $128$ uniformly sampled viewpoints ($128,000$ annotated images). Using this controlled dataset, we quantify the influence of elevation and identify a critical elevation threshold, above which models generalize significantly better to unseen views. Finally, we validate the transferability to real-world applications in both zero-shot and supervised settings, demonstrating performance gains across four real-world cattle datasets and confirming that synthetic geometric priors effectively bridge the domain gap. Collectively, this dataset and analysis lay the foundation for future model development in cross-view animal ReID. MOO is publicly available at https://github.com/TurtleSmoke/MOO.
Authors: Elan Barenholtz
Abstract: Recent work interprets the linear recoverability of geographic and temporal variables from large language model (LLM) hidden states as evidence for world-like internal representations. We test a simpler possibility: that much of the relevant structure is already latent in text itself. Applying the same class of ridge regression probes to static co-occurrence-based embeddings (GloVe and Word2Vec), we find substantial recoverable geographic signal and weaker but reliable temporal signal, with held-out R^2 values of 0.71-0.87 for city coordinates and 0.48-0.52 for historical birth years. Semantic-neighbor analyses and targeted subspace ablations show that these signals depend strongly on interpretable lexical gradients, especially country names and climate-related vocabulary. These findings suggest that ordinary word co-occurrence preserves richer spatial, temporal, and environmental structure than is often assumed, revealing a remarkable and underappreciated capacity of simple static embeddings to preserve world-shaped structure from text alone. Linear probe recoverability alone therefore does not establish a representational move beyond text.
Authors: Umid Suleymanov, Murat Kantarcioglu, Kevin S Chan, Michael De Lucia, Kevin Hamlen, Latifur Khan, Sharad Mehrotra, Ananthram Swami, Bhavani Thuraisingham
Abstract: Real-world systems must continuously adapt to novel concepts from limited data without forgetting previously acquired knowledge. While Few-Shot Class-Incremental Learning (FSCIL) is established in computer vision, its application to tabular domains remains largely unexplored. Unlike images, tabular streams (e.g., logs, sensors) offer abundant unlabeled data, a scarcity of expert annotations and negligible storage costs, features ignored by existing vision-based methods that rely on restrictive buffers. We introduce SPRINT, the first FSCIL framework tailored for tabular distributions. SPRINT introduces a mixed episodic training strategy that leverages confidence-based pseudo-labeling to enrich novel class representations and exploits low storage costs to retain base class history. Extensive evaluation across six diverse benchmarks spanning cybersecurity, healthcare, and ecological domains, demonstrates SPRINT's cross-domain robustness. It achieves a state-of-the-art average accuracy of 77.37% (5-shot), outperforming the strongest incremental baseline by 4.45%.
Authors: Bhavya Agrawalla, Michal Nauman, Aviral Kumar
Abstract: Recent work shows that flow matching can be effective for scalar Q-value function estimation in reinforcement learning (RL), but it remains unclear why or how this approach differs from standard critics. Contrary to conventional belief, we show that their success is not explained by distributional RL, as explicitly modeling return distributions can reduce performance. Instead, we argue that the use of integration for reading out values and dense velocity supervision at each step of this integration process for training improves TD learning via two mechanisms. First, it enables robust value prediction through \emph{test-time recovery}, whereby iterative computation through integration dampens errors in early value estimates as more integration steps are performed. This recovery mechanism is absent in monolithic critics. Second, supervising the velocity field at multiple interpolant values induces more \emph{plastic} feature learning within the network, allowing critics to represent non-stationary TD targets without discarding previously learned features or overfitting to individual TD targets encountered during training. We formalize these effects and validate them empirically, showing that flow-matching critics substantially outperform monolithic critics (2$\times$ in final performance and around 5$\times$ in sample efficiency) in settings where loss of plasticity poses a challenge e.g., in high-UTD online RL problems, while remaining stable during learning.
Authors: Rocky Klopfenstein, Yang He, Andrew Tremante, Yuepeng Wang, Nina Narodytska, Haoze Wu
Abstract: We present SpotIt+, an open-source tool for evaluating Text-to-SQL systems via bounded equivalence verification. Given a generated SQL query and the ground truth, SpotIt+ actively searches for database instances that differentiate the two queries. To ensure that the generated counterexamples reflect practically relevant discrepancies, we introduce a constraint-mining pipeline that combines rule-based specification mining over example databases with LLM-based validation. Experimental results on the BIRD dataset show that the mined constraints enable SpotIt+ to generate more realistic differentiating databases, while preserving its ability to efficiently uncover numerous discrepancies between generated and gold SQL queries that are missed by standard test-based evaluation.
Authors: Yixin Chen, Ziyu Su, Hikmat Khan, Muhammad Khalid Khan Niazi
Abstract: Pathology report generation remains a relatively under-explored downstream task, primarily due to the gigapixel scale and complex morphological heterogeneity of Whole Slide Images (WSIs). Existing pathology report generation frameworks typically employ transformer architectures, relying on a homogeneous decoder architecture and static knowledge retrieval integration. Such architectures limit generative specialization and may introduce noisy external guidance during the report generation process. To address these limitations, we propose RANGER, a sparsely-gated Mixture-of-Experts (MoE) framework with adaptive retrieval re-ranking for pathology report generation. Specifically, we integrate a sparsely gated MoE into the decoder, along with noisy top-$k$ routing and load-balancing regularization, to enable dynamic expert specialization across various diagnostic patterns. Additionally, we introduce an adaptive retrieval re-ranking module that selectively refines retrieved memory from a knowledge base before integration, reducing noise and improving semantic alignment based on visual feature representations. We perform extensive experiments on the PathText-BRCA dataset and demonstrate consistent improvements over existing approaches across standard natural language generation metrics. Our full RANGER model achieves optimal performance on PathText dataset, reaching BLEU-1 to BLEU-4 scores of 0.4598, 0.3044, 0.2036, and 0.1435, respectively, with METEOR of 0.1883, and ROUGE-L of 0.3038, validating the effectiveness of dynamic expert routing and adaptive knowledge refinement for semantically grounded pathology report generation.
Authors: Geraldin Nanfack, Eugene Belilovsky, Elvis Dohmatob
Abstract: Safety-aligned language models refuse harmful requests through learned refusal behaviors encoded in their internal representations. Recent activation-based jailbreaking methods circumvent these safety mechanisms by applying orthogonal projections to remove refusal directions, but these approaches treat refusal as a one-dimensional phenomenon and ignore the rich distributional structure of model activations. We introduce a principled framework based on optimal transport theory that transforms the entire distribution of harmful activations to match harmless ones. By combining PCA with closed-form Gaussian optimal transport, we achieve efficient computation in high-dimensional representation spaces while preserving essential geometric structure. Across six models (Llama-2, Llama-3.1, Qwen-2.5; 7B-32B parameters), our method achieves up to 11% higher attack success rates than state-of-the-art baselines while maintaining comparable perplexity, demonstrating superior preservation of model capabilities. Critically, we discover that layer-selective intervention (applying optimal transport to 1-2 carefully chosen layers at approximately 40-60% network depth) substantially outperforms full-network interventions, revealing that refusal mechanisms may be localized rather than distributed. Our analysis provides new insights into the geometric structure of safety representations and suggests that current alignment methods may be vulnerable to distributional attacks beyond simple direction removal.
Authors: Soroush Nasiriany, Sepehr Nasiriany, Abhiram Maddukuri, Yuke Zhu
Abstract: Recent advances in robot learning have accelerated progress toward generalist robots that can perform everyday tasks in human environments. Yet it remains difficult to gauge how close we are to this vision. The field lacks a reproducible, large-scale benchmark for systematic evaluation. To fill this gap, we present RoboCasa365, a comprehensive simulation benchmark for household mobile manipulation. Built on the RoboCasa platform, RoboCasa365 introduces 365 everyday tasks across 2,500 diverse kitchen environments, with over 600 hours of human demonstration data and over 1600 hours of synthetically generated demonstration data -- making it one of the most diverse and large-scale resources for studying generalist policies. RoboCasa365 is designed to support systematic evaluations for different problem settings, including multi-task learning, robot foundation model training, and lifelong learning. We conduct extensive experiments on this benchmark with state-of-the-art methods and analyze the impacts of task diversity, dataset scale, and environment variation on generalization. Our results provide new insights into what factors most strongly affect the performance of generalist robots and inform strategies for future progress in the field.
Authors: Marco Federici, Boris van Breugel, Paul Whatmough, Markus Nagel
Abstract: Quantization can drastically increase the efficiency of large language and vision models, but typically incurs an accuracy drop. Recently, function-preserving transforms (e.g. rotations, Hadamard transform, channel-wise scaling) have been successfully applied to reduce post-training quantization error, yet a principled explanation remains elusive. We analyze linear-layer quantization via the signal-to-quantization-noise ratio (SQNR), showing that for uniform integer quantization at a fixed bit width, SQNR decomposes into (i) the concentration of weights and activations (capturing spread and outliers), and (ii) the alignment of their dominant variation directions. This reveals an actionable insight: beyond concentration - the focus of most prior transforms (e.g. rotations or Hadamard) - improving alignment between weight and activation can further reduce quantization error. Motivated by this, we introduce block Concentration-Alignment Transforms (CAT), a lightweight linear transformation that uses a covariance estimate from a small calibration set to jointly improve concentration and alignment, approximately maximizing SQNR. Experiments across several LLMs show that CAT consistently matches or outperforms prior transform-based quantization methods at 4-bit precision, confirming the insights gained in our framework.
Authors: Haoyu Liu, Dingcheng Li, Lukas Rutishauser, Zeyu Zheng
Abstract: Multimodal web agents that process both screenshots and accessibility trees are increasingly deployed to interact with web interfaces, yet their dual-stream architecture opens an underexplored attack surface: an adversary who injects content into the webpage DOM simultaneously corrupts both observation channels with a consistent deceptive narrative. Our vulnerability analysis on MiniWob++ reveals that attacks including a visual component far outperform text-only injections, exposing critical gaps in text-centric VLM safety training. Motivated by this finding, we propose Dual-Modality Multi-Stage Adversarial Safety Training (DMAST), a framework that formalizes the agent-attacker interaction as a two-player zero-sum Markov game and co-trains both players through a three-stage pipeline: (1) imitation learning from a strong teacher model, (2) oracle-guided supervised fine-tuning that uses a novel zero-acknowledgment strategy to instill task-focused reasoning under adversarial noise, and (3) adversarial reinforcement learning via Group Relative Policy Optimization (GRPO) self-play. On out-of-distribution tasks, DMAST substantially mitigates adversarial risks while simultaneously doubling task completion efficiency. Our approach significantly outperforms established training-based and prompt-based defenses, demonstrating genuine co-evolutionary progress and robust generalization to complex, unseen environments.
Authors: Zachary Novack, Zack Zukowski, CJ Carr, Julian Parker, Zach Evans, Josiah Taylor, Taylor Berg-Kirkpatrick, Julian McAuley, Jordi Pons
Abstract: Generative audio requires fine-grained controllable outputs, yet most existing methods require model retraining on specific controls or inference-time controls (\textit{e.g.}, guidance) that can also be computationally demanding. By examining the bottlenecks of existing guidance-based controls, in particular their high cost-per-step due to decoder backpropagation, we introduce a guidance-based approach through selective TFG and Latent-Control Heads (LatCHs), which enables controlling latent audio diffusion models with low computational overhead. LatCHs operate directly in latent space, avoiding the expensive decoder step, and requiring minimal training resources (7M parameters and $\approx$ 4 hours of training). Experiments with Stable Audio Open demonstrate effective control over intensity, pitch, and beats (and a combination of those) while maintaining generation quality. Our method balances precision and audio fidelity with far lower computational costs than standard end-to-end guidance. Demo examples can be found at https://zacharynovack.github.io/latch/latch.html.
Authors: Furkan Mumcu, Yasin Yilmaz
Abstract: As Large Language Models (LLMs) transition into autonomous multi-agent ecosystems, robust minimax training becomes essential yet remains prone to instability when highly non-linear policies induce extreme local curvature in the inner maximization. Standard remedies that enforce global Jacobian bounds are overly conservative, suppressing sensitivity in all directions and inducing a large Price of Robustness. We introduce Adversarially-Aligned Jacobian Regularization (AAJR), a trajectory-aligned approach that controls sensitivity strictly along adversarial ascent directions. We prove that AAJR yields a strictly larger admissible policy class than global constraints under mild conditions, implying a weakly smaller approximation gap and reduced nominal performance degradation. Furthermore, we derive step-size conditions under which AAJR controls effective smoothness along optimization trajectories and ensures inner-loop stability. These results provide a structural theory for agentic robustness that decouples minimax stability from global expressivity restrictions.
Authors: Haian Jin, Rundi Wu, Tianyuan Zhang, Ruiqi Gao, Jonathan T. Barron, Noah Snavely, Aleksander Holynski
Abstract: Feed-forward transformer models have driven rapid progress in 3D vision, but state-of-the-art methods such as VGGT and $\pi^3$ have a computational cost that scales quadratically with the number of input images, making them inefficient when applied to large image collections. Sequential-reconstruction approaches reduce this cost but sacrifice reconstruction quality. We introduce ZipMap, a stateful feed-forward model that achieves linear-time, bidirectional 3D reconstruction while matching or surpassing the accuracy of quadratic-time methods. ZipMap employs test-time training layers to zip an entire image collection into a compact hidden scene state in a single forward pass, enabling reconstruction of over 700 frames in under 10 seconds on a single H100 GPU, more than $20\times$ faster than state-of-the-art methods such as VGGT. Moreover, we demonstrate the benefits of having a stateful representation in real-time scene-state querying and its extension to sequential streaming reconstruction.
Authors: Tian Yu Liu, Stefano Soatto
Abstract: The semantic similarity between sample expressions measures the distance between their latent 'meaning'. These meanings are themselves typically represented by textual expressions. We propose a novel approach whereby the semantic similarity among textual expressions is based not on other expressions they can be rephrased as, but rather based on the imagery they evoke. While this is not possible with humans, generative models allow us to easily visualize and compare generated images, or their distribution, evoked by a textual prompt. Therefore, we characterize the semantic similarity between two textual expressions simply as the distance between image distributions they induce, or 'conjure.' We show that by choosing the Jeffreys divergence between the reverse-time diffusion stochastic differential equations (SDEs) induced by each textual expression, this can be directly computed via Monte-Carlo sampling. Our method contributes a novel perspective on semantic similarity that not only aligns with human-annotated scores, but also opens up new avenues for the evaluation of text-conditioned generative models while offering better interpretability of their learnt representations.
Authors: Hai Zhong, Xun Wang, Zhuoran Li, Longbo Huang
Abstract: Offline-to-Online Reinforcement Learning has emerged as a powerful paradigm, leveraging offline data for initialization and online fine-tuning to enhance both sample efficiency and performance. However, most existing research has focused on single-agent settings, with limited exploration of the multi-agent extension, i.e., Offline-to-Online Multi-Agent Reinforcement Learning (O2O MARL). In O2O MARL, two critical challenges become more prominent as the number of agents increases: (i) the risk of unlearning pre-trained Q-values due to distributional shifts during the transition from offline-to-online phases, and (ii) the difficulty of efficient exploration in the large joint state-action space. To tackle these challenges, we propose a novel O2O MARL framework called Offline Value Function Memory with Sequential Exploration (OVMSE). First, we introduce the Offline Value Function Memory (OVM) mechanism to compute target Q-values, preserving knowledge gained during offline training, ensuring smoother transitions, and enabling efficient fine-tuning. Second, we propose a decentralized Sequential Exploration (SE) strategy tailored for O2O MARL, which effectively utilizes the pre-trained offline policy for exploration, thereby significantly reducing the joint state-action space to be explored. Extensive experiments on the StarCraft Multi-Agent Challenge (SMAC) demonstrate that OVMSE significantly outperforms existing baselines, achieving superior sample efficiency and overall performance.
Authors: Xi Chen (M-PSI), Julien Cumin (M-PSI), Fano Ramparany (M-PSI), Dominique Vaufreydaz (M-PSI)
Abstract: Recent progress in Large Language Models (LLMs) has enabled advanced reasoning and zero-shot recognition for human activity understanding with ambient sensor data. However, widely used multi-resident datasets such as CASAS, ARAS, and MARBLE lack natural language context and fine-grained annotation, limiting the full exploitation of LLM capabilities in realistic smart environments. To address this gap, we present MuRAL (Multi-Resident Ambient sensor dataset with natural Language), comprising over 21 hours of multi-user sensor data from 21 sessions in a smart home. MuRAL uniquely features detailed natural language descriptions, explicit resident identities, and rich activity labels, all situated in complex, dynamic, multi-resident scenarios. We benchmark state-of-the-art LLMs on MuRAL for three core tasks: subject assignment, action description, and activity classification. Results show that current LLMs still face major challenges on MuRAL, especially in maintaining accurate resident assignment over long sequences, generating precise action descriptions, and effectively integrating context for activity prediction. The dataset is publicly available at: https://mural.imag.fr/.
URLs: https://mural.imag.fr/.
Authors: Hermann Borotschnig
Abstract: As artificial agents display increasingly sophisticated emotion-like behaviors, frameworks for assessing whether such systems risk instantiating consciousness remain limited. This contribution asks whether synthetic emotion-like control can be implemented while deliberately excluding architectural features that major theories associate with access-like consciousness. We propose architectural principles (A1-A8) for a hierarchical, dual-source implementation in which (i) immediate needs generate motivational signals and (ii) episodic memory provides affective guidance from similar past situations; the two sources converge to modulate action selection. To operationalize consciousness-related risk, we distill predictions from major theories into four engineering risk-reduction constraints: (R1) no content-general, workspace-like global broadcast, (R2) no metarepresentation, (R3) no autobiographical consolidation, and (R4) bounded learning. We address three questions: (Q1) Can emotion-like control satisfy R1-R4? We present a concrete architecture as an existence proof. (Q2) Can the architecture be extended without introducing access-enabling features? We identify stable modifications that preserve compliance. (Q3) Can we trace graded paths that plausibly increase access risk? We map gradual transitions that progressively violate the constraints. Our contribution operates at three levels: on the engineering side, we present a modular, biologically motivated control architecture; on the theoretical side, we propose a control model of emotions and a methodological template for converting consciousness-related questions into auditable architectural tests; on the safety side, we sketch preliminary audit indicators that may inform future governance frameworks. The architecture functions independently as an emotion-like controller, while the risk-reduction criteria may extend to other AI systems.
Authors: Rintaro Ando
Abstract: We present the Emotion-Gradient Metacognitive Recursive Self-Improvement (EG-MRSI) framework, a novel architecture that integrates introspective metacognition, emotion-based intrinsic motivation, and recursive self-modification into a unified theoretical system. The framework is explicitly capable of overwriting its own learning algorithm under formally bounded risk. Building upon the Noise-to-Meaning RSI (N2M-RSI) foundation, EG-MRSI introduces a differentiable intrinsic reward function driven by confidence, error, novelty, and cumulative success. This signal regulates both a metacognitive mapping and a self-modification operator constrained by provable safety mechanisms. We formally define the initial agent configuration, emotion-gradient dynamics, and RSI trigger conditions, and derive a reinforcement-compatible optimization objective that guides the agent's development trajectory. Meaning Density and Meaning Conversion Efficiency are introduced as quantifiable metrics of semantic learning, closing the gap between internal structure and predictive informativeness. This Part I paper establishes the single-agent theoretical foundations of EG-MRSI. Future parts will extend this framework to include safety certificates and rollback protocols (Part II), collective intelligence mechanisms (Part III), and feasibility constraints including thermodynamic and computational limits (Part IV). Together, the EG-MRSI series provides a rigorous, extensible foundation for open-ended and safe AGI.
Authors: Yue Zhang, Zhiliang Tian, Shicheng Zhou, Haiyang Wang, Wenqing Hou, Yuying Liu, Xuechen Zhao, Minlie Huang, Ye Wang, Bin Zhou
Abstract: Legal Judgment Prediction (LJP) is a pivotal task in legal AI. Existing semantic-enhanced LJP models integrate judicial precedents and legal knowledge for high performance. But they neglect legal reasoning logic, a critical component of legal judgments requiring rigorous logical analysis. Although some approaches utilize legal reasoning logic for high-quality predictions, their logic rigidity hinders adaptation to case-specific logical frameworks, particularly in complex cases that are lengthy and detailed. This paper proposes a rule-enhanced legal judgment prediction framework based on first-order logic (FOL) formalism and comparative learning (CL) to develop an adaptive adjustment mechanism for legal judgment logic and further enhance performance in LJP. Inspired by the process of human exam preparation, our method follows a three-stage approach: first, we initialize judgment rules using the FOL formalism to capture complex reasoning logic accurately; next, we propose a Confusion-aware Contrastive Learning (CACL) to dynamically optimize the judgment rules through a quiz consisting of confusable cases; finally, we utilize the optimized judgment rules to predict legal judgments. Experimental results on two public datasets show superior performance across all metrics. The code is publicly available{https://anonymous.4open.science/r/RLJP-FDF1}.
Authors: Yongchao Chen, Yueying Liu, Junwei Zhou, Yilun Hao, Jingquan Wang, Yang Zhang, Na Li, Chuchu Fan
Abstract: Practical guidance on training Large Language Models (LLMs) to leverage Code Interpreter across diverse tasks remains lacking. We present R1-Code-Interpreter, an extension of a text-only LLM trained via multi-turn supervised fine-tuning (SFT) and reinforcement learning (RL) to autonomously generate multiple code queries during step-by-step reasoning. Unlike prior RL + tool-use efforts focused on narrow domains such as math or retrieval, we curate 144 diverse reasoning and planning tasks and show that training a general-purpose Code Interpreter across them presents significant challenges due to task heterogeneity and scarcity of effective samples. To address this, we introduce a multi-stage curriculum learning approach that partitions training samples by measured improvement potential. The RL training prioritizes samples with higher potential and gradually shifts to lower-potential ones, increasing the average RL gains from merely +3.4% to +9.3% across Qwen-2.5 models (3/7/14B). Our final model, R1-CI-14B, improves average accuracy on the 37 test tasks from 44.1% to 72.4%, outperforming text-only GPT-4o (58.6%) and GPT-4o with Code Interpreter (70.9%). Notably, R1-CI-14B also exhibits emergent self-checking behavior through code generation. Datasets, Codes, and Models are available at https://github.com/yongchao98/R1-Code-Interpreter and https://huggingface.co/yongchao98.
URLs: https://github.com/yongchao98/R1-Code-Interpreter, https://huggingface.co/yongchao98.
Authors: Manooshree Patel, Rayna Bhattacharyya, Thomas Lu, Arnav Mehta, Niels Voss, Narges Norouzi, Gireeja Ranade
Abstract: This paper considers the development of an AI-based provably-correct mathematical proof tutor. While Large Language Models (LLMs) allow seamless communication in natural language, they are error prone. Theorem provers such as Lean allow for provable-correctness, but these are hard for students to learn. We present a proof-of-concept system (LeanTutor) by combining the complementary strengths of LLMs and theorem provers. LeanTutor is composed of three modules: (i) an autoformalizer/proof-checker, (ii) a next-step generator, and (iii) a natural language feedback generator. To evaluate the system, we introduce PeanoBench, a dataset of 371 Peano Arithmetic proofs in human-written natural language and formal language, derived from the Natural Numbers Game.
Authors: Shufan Jiang, Sizhou Chen, Chi Zhang, Xiao-Lei Zhang, Xuelong Li
Abstract: Creating an immersive and interactive theatrical experience is a long-term goal in the field of interactive narrative. The emergence of large language models (LLMs) provides a new path to achieve this goal. However, existing LLM-based drama generation methods often produce models that lack initiative and cannot interact with the physical scene, while typically requiring detailed user input that diminishes the immersion of live performance. To address these challenges, we propose HAMLET, a hierarchical adaptive multi-agent framework focused on drama creation and real-time online performance. Given a simple topic, the framework first generates a narrative blueprint to guide the subsequent improvisational performance. In the online performance phase, each actor is equipped with an adaptive reasoning module that enables decision-making based on their personas, memories, goals, and emotional states during complex group chat scenarios. Beyond dialogue, actor agents engage in embodied interactions by changing the state of scene props through actions such as opening a letter or picking up a weapon, which are broadcast to update the global environmental context. To objectively assess the quality of live embodied theatrics, we establish a comprehensive evaluation method and introduce HAMLETJudge, a specialized critic model for automated evaluation. Experimental results demonstrate that HAMLET excels in creating expressive, coherent, and physically interactive theatrical experiences in an autonomous manner.
Authors: Nuria Rodr\'iguez-Barroso, Mario Garc\'ia-M\'arquez, M. Victoria Luz\'on, Francisco Herrera
Abstract: Federated Learning (FL) enables privacy-preserving collaborative learning, yet deployments increasingly show that privacy guarantees alone do not sustain trust in high-risk settings. As FL systems move toward agentic AI, large language model-enabled, and dynamically adaptive architectures, trustworthiness becomes a system-level problem shaped by autonomous decision-making, non-stationary environments, and multi-stakeholder governance. We argue for Trustworthy FL (TFL), treating trust as a continuously maintained operating condition rather than a static model property. Through the lens of Trust Report 2.0, we propose a requirement-driven taxonomy of challenges grounded in TAI and explicitly extended to account for control-plane decisions, agency, and system dynamics across the federated lifecycle. Building on this diagnosis, we introduce a coordination blueprint that structures cross-requirement trade-offs, decision justification, and governance alignment in TFL systems. To operationalize assurance, Trust Report 2.0 is instantiated as a lightweight, privacy-preserving artifact that surfaces decision-centric trust evidence without centralizing raw data. We illustrate applicability via healthcare as a stress-test domain, focusing on oncology FL under regulatory pressure and clinical risk.
Authors: Stassa Patsantzis
Abstract: Inductive Logic Programming (ILP) approaches like Meta \-/ Interpretive Learning (MIL) can learn, from few examples, recursive logic programs with invented predicates that generalise well to unseen instances. This ability relies on a background theory and negative examples, both carefully selected with expert knowledge of a learning problem and its solutions. But what if such a problem-specific background theory or negative examples are not available? We formalise this question as a new setting for Self-Supervised ILP and present a new MIL algorithm that learns in the new setting from some positive labelled, and zero or more unlabelled examples, and automatically generates, and labels, new positive and negative examples during learning. We implement this algorithm in Prolog in a new MIL system, called Poker. We compare Poker to state-of-the-art MIL system Louise on experiments learning grammars for Context-Free and L-System languages from labelled, positive example strings, no negative examples, and just the terminal vocabulary of a language, seen in examples, as a first-order background theory. We introduce a new approach for the principled selection of a second-order background theory as a Second Order Definite Normal Form (SONF), sufficiently general to learn all programs in a class, thus removing the need for a backgound theory tailored to a learning task. We find that Poker's performance improves with increasing numbers of automatically generated examples while Louise, bereft of negative examples, over-generalises.
Authors: Shaofeng Yin, Ting Lei, Yang Liu
Abstract: Integrating external tools into Large Foundation Models (LFMs) has emerged as a promising approach to enhance their problem-solving capabilities. While existing studies have demonstrated strong performance in tool-augmented Visual Question Answering (VQA), recent benchmarks reveal significant gaps in real-world tool-use proficiency, particularly in functionally diverse multimodal settings requiring multi-step reasoning. In this work, we introduce ToolVQA, a large-scale multimodal dataset comprising 23K instances, designed to bridge this gap. Unlike previous datasets that rely on synthetic scenarios and simplified queries, ToolVQA features real-world visual contexts and challenging implicit multi-step reasoning tasks, better aligning with real user interactions. To construct this dataset, we propose ToolEngine, a novel data generation pipeline that employs Depth-First Search (DFS) with a dynamic in-context example matching mechanism to simulate human-like tool-use reasoning. ToolVQA encompasses 10 multimodal tools across 7 diverse task domains, with an average inference length of 2.78 reasoning steps per instance. The fine-tuned 7B LFMs on ToolVQA not only achieve impressive performance on our test set but also surpass the large close-sourced model GPT-3.5-turbo on various out-of-distribution (OOD) datasets, demonstrating strong generalizability to real-world tool-use scenarios.
Authors: Zongru Wu, Rui Mao, Zhiyuan Tian, Pengzhou Cheng, Tianjie Ju, Zheng Wu, Lingzhong Dong, Haiyue Sheng, Zhuosheng Zhang, Gongshen Liu
Abstract: The advent of multimodal agents facilitates effective interaction within graphical user interface (GUI), especially in ubiquitous GUI control. However, their inability to reliably execute toggle control instructions remains a key bottleneck. To investigate this, we construct a state control benchmark with binary toggle instructions derived from public datasets. Evaluation results of existing agents demonstrate their notable unreliability, particularly when the current toggle state already matches the desired state. To address the challenge, we propose State-aware Reasoning (StaR), a multimodal reasoning method that enables agents to perceive the current toggle state, infer the desired state from the instruction, and act accordingly. Experiments on four multimodal agents demonstrate that StaR can improve toggle instruction execution accuracy by over 30\%. Further evaluations on three public agentic benchmarks show that StaR also enhances general agentic task performance. Finally, evaluations on a dynamic environment highlight the potential of StaR for real-world applications. Code and benchmark: https://github.com/ZrW00/StaR.
Authors: Junliang Liu, Jingyu Xiao, Wenxin Tang, Zhixian Wang, Zipeng Xie, Wenxuan Wang, Minrui Zhang, Shuanghe Yu
Abstract: Multimodal large language models (MLLMs) are increasingly deployed as the core reasoning engine for web-facing systems, powering GUI agents and front-end automation that must interpret page structure, select actionable widgets, and execute multi-step interactions reliably. However, existing benchmarks largely emphasize visual perception or UI code generation, showing insufficient evaluation on the reasoning, robustness and safety capability required for end-to-end web applications. To bridge the gap, we introduce a comprehensive web understanding benchmark, named WebRRSBench, that jointly evaluates Reasoning, Robustness, and Safety across eight tasks, such as position relationship reasoning, color robustness, and safety critical detection, etc. The benchmark is constructed from 729 websites and contains 3799 QA pairs that probe multi-step inference over page structure, text, widgets, and safety-critical interactions. To ensure reliable measurement, we adopt standardized prompts, a protocolized and deterministic evaluation pipeline, and multi-stage quality control combining automatic checks with targeted human verification. We evaluate 11 MLLMs on WebRRSBench. The results reveal significant gaps: models still struggle with compositional and cross-element reasoning over realistic layouts, show limited robustness when facing perturbations in user interfaces and content such as layout rearrangements or visual style shifts, and are rather conservative in recognizing and avoiding safety critical or irreversible actions. Our code and appendix are available at https: //github.com/annoy-worker/WebRSSBench.
Authors: Pengkun Jiao, Yiming Jin, Jianhui Yang, Chenhe Dong, Zerui Huang, Shaowei Yao, Xiaojiang Zhou, Dan Ou, Haihong Tang
Abstract: Query-product relevance prediction is a foundational technology in e-commerce search engines and has become increasingly important in AI-driven e-commerce. The recent emergence of LLMs, particularly their CoT reasoning capabilities, offers promising opportunities for developing relevance systems that are both more interpretable and more robust. However, existing training paradigms have notable limitations: SFT and DPO suffer from poor generalization on long-tail queries and from a lack of fine-grained, stepwise supervision to enforce rule-aligned reasoning. In contrast, reinforcement learning with verification rewards (RLVR) suffers from sparse feedback, which provides insufficient signal to correct erroneous intermediate steps, thereby undermining logical consistency and limiting performance in complex inference scenarios. To address these challenges, we introduce the Stepwise Hybrid Examination Reinforcement Learning framework for search relevance (SHE). At its core is Stepwise Reward Policy Optimization (SRPO), a reinforcement learning algorithm that leverages step-level rewards generated by a hybrid of a high-quality generative stepwise reward model and a human-annotated offline verifier, prioritizing learning from critical correct and incorrect reasoning steps. To bolster robustness and generalization, SHE further integrates a dual-strategy optimization: diversified data filtering, which broadens the exploration of reasoning trajectories to preempt policy entropy collapse, and a multi-stage curriculum learning protocol that systematically orchestrates progressive capability acquisition. Extensive experiments on real-world search benchmarks show that SHE improves both reasoning quality and relevance-prediction accuracy in large-scale e-commerce settings, outperforming SFT, DPO, GRPO, and other baselines, while also enhancing interpretability and robustness.
Authors: Margarita Belova, Jiaxin Xiao, Shikhar Tuli, Niraj K. Jha
Abstract: Researchers have pursued neurosymbolic artificial intelligence (AI) applications for nearly three decades. A marriage of the neural and symbolic components can lead to rapid advancements in AI. Yet, the field has not realized this promise since most neurosymbolic AI frameworks fail to scale. In addition, the implicit representations and approximate reasoning of purely neural approaches limit interpretability and trust. Knowledge graphs (KGs), a gold-standard representation of explicit semantic knowledge, can address the symbolic side of the problem. However, automatically deriving reliable KGs from text corpora remains an open problem. We address these challenges by introducing GraphMERT, a tiny graphical encoder-only model that distills high-quality KGs from unstructured text corpora and its own internal representations. GraphMERT and its equivalent KG form a modular neurosymbolic stack: neural learning of abstractions; symbolic KGs for verifiable reasoning. GraphMERT + KG is the first efficient and scalable neurosymbolic model to achieve state-of-the-art benchmark accuracy along with superior symbolic representations relative to baselines. Concretely, we target reliable domain-specific KGs that are both (1) factual (with provenance) and (2) valid (ontology-consistent relations with domain-appropriate semantics). When a large language model (LLM), e.g., Qwen3-32B, generates domain-specific KGs, it falls short on reliability due to prompt sensitivity, shallow domain expertise, and hallucinated relations. On text obtained from PubMed papers on diabetes, our 80M-parameter GraphMERT yields a KG with a 69.8% FActScore; a 32B-parameter baseline LLM yields a KG that achieves only 40.2% FActScore. The GraphMERT KG also attains a higher ValidityScore of 68.8%, versus 43.0% for the LLM baseline.
Authors: Yufa Zhou, Yixiao Wang, Xunjian Yin, Shuyan Zhou, Anru R. Zhang
Abstract: We study how large language models (LLMs) ``think'' through their representation space. We propose a novel geometric framework that models an LLM's reasoning as flows -- embedding trajectories evolving where logic goes. We disentangle logical structure from semantics by employing the same natural deduction propositions with varied semantic carriers, allowing us to test whether LLMs internalize logic beyond surface form. This perspective connects reasoning with geometric quantities such as position, velocity, and curvature, enabling formal analysis in representation and concept spaces. Our theory establishes: (1) LLM reasoning corresponds to smooth flows in representation space, and (2) logical statements act as local controllers of these flows' velocities. Using learned representation proxies, we design controlled experiments to visualize and quantify reasoning flows, providing empirical validation of our theoretical framework. Our findings indicate that training solely via next-token prediction can lead LLMs to internalize logical invariants as higher-order geometry in representation space, challenging the ``stochastic parrot'' argument. Experiments across Qwen and LLaMA model families further suggest the presence of a general, possibly universal, representational law underlying machine understanding and human linguistic regularities, largely independent of specific training recipes or model architectures. Our work serves as both a conceptual foundation and practical tools for studying reasoning phenomena, offering a new lens for interpretability and formal analysis of LLMs' behavior.
Authors: Pedro Antonio Alarcon Granadeno, Arturo Miguel Bernal Russell, Sofia Nelson, Demetrius Hernandez, Maureen Petterson, Michael Murphy, Walter J. Scheirer, Jane Cleland-Huang
Abstract: Cyber-physical systems increasingly rely on foundational models, such as Large Language Models (LLMs) and Vision-Language Models (VLMs) to increase autonomy through enhanced perception, inference, and planning. However, these models also introduce new types of errors, such as hallucinations, over-generalizations, and context misalignments, resulting in incorrect and flawed decisions. To address this, we introduce the concept of Cognition Envelopes, designed to establish reasoning boundaries that constrain AI-generated decisions while complementing the use of meta-cognition and traditional safety envelopes. As with safety envelopes, Cognition Envelopes require practical guidelines and systematic processes for their definition, validation, and assurance. In this paper we describe an LLM/VLM-supported pipeline for dynamic clue analysis within the domain of small autonomous Uncrewed Aerial Systems deployed on Search and Rescue (SAR) missions, and a Cognition Envelope based on probabilistic reasoning and resource analysis. We evaluate the approach through assessing decisions made by our Clue Analysis Pipeline in a series of SAR missions. Finally, we identify key software engineering challenges for systematically designing, implementing, and validating Cognition Envelopes for AI-supported decisions in cyber-physical systems.
Authors: Zepeng Bao, Shen Zhou, Qiankun Pi, Jianhao Chen, Mayi Xu, Ming Zhong, Yuanyuan Zhu, Tieyun Qian
Abstract: Hallucination in large language models (LLMs) remains a critical barrier to their safe deployment. For hallucination detection to be practical in real-world scenarios, the use of efficient small models is essential to ensure low latency and minimal resource consumption. However, existing methods rely on fixed verification strategies, where simply tuning small models to mimic fixed verification trajectories fails to capture the adaptability required for diverse hallucination patterns, thereby inducing planning instability. To address this limitation, we propose a ``Learning to Evaluate and Adaptively Plan'' (LEAP) framework, which shifts hallucination detection from fixed execution to dynamic strategy learning. Specifically, LEAP first employs a powerful teacher model to iteratively explore and refine verification strategies through a failure-driven loop. This dynamic planning capability is then distilled into an efficient student model, augmented by a novel proactive correction mechanism that enables the model to evaluate and optimize its verification strategy before execution. Experiments on three benchmarks demonstrate that LEAP outperforms state-of-the-art methods, offering an effective and scalable solution for reliable hallucination detection.
Authors: Peiran Xu, Sudong Wang, Yao Zhu, Jianing Li, Gege Qi, Yunjian Zhang
Abstract: Spatial cognition is fundamental to real-world multimodal intelligence, allowing models to effectively interact with the physical environment. While multimodal large language models (MLLMs) have made significant strides, existing benchmarks often oversimplify spatial cognition, reducing it to a single-dimensional metric, which fails to capture the hierarchical structure and interdependence of spatial abilities. To address this gap, we propose a hierarchical spatial cognition framework that decomposes spatial intelligence into five progressively complex levels from basic observation to high-level planning. Building upon this taxonomy, we construct SpatialBench, a large-scale, fine-grained benchmark covering 15 tasks aligned with these cognitive levels. To provide a unified evaluation across heterogeneous tasks, we further introduce a high-level capability-oriented metric that reliably assesses a model's overall spatial reasoning ability. Extensive experiments over massive MLLMs reveal distinct performance stratification across cognitive levels: models exhibit strong perceptual grounding yet remain limited in symbolic reasoning, causal inference, and planning. Additional human tests demonstrate that humans perform selective, goal-directed abstraction, while MLLMs tend to over-attend to surface details without coherent spatial intent. Our work establishes the first systematic framework for measuring hierarchical spatial cognition in MLLMs, laying the foundation for future spatially intelligent systems.
Authors: Zehao Deng, Tianjie Ju, Zheng Wu, Zhuosheng Zhang, Gongshen Liu
Abstract: The rapid development of large vision-language model (VLM) has greatly promoted the research of GUI agent. However, GUI agents still face significant challenges in handling long-horizon tasks. First, single-agent models struggle to balance high-level capabilities and low-level execution capability, facing prevalent issues of responsibility coupling and capability conflicts. Second, agents lack awareness of the task state, leading to progress loss in long-horizon tasks. To address these challenges, we propose a staged execution-feedback reinforcement learning algorithm. Unlike training a unified policy model, we focus on training high-level scheduling models. Specifically, we propose and train two agents: a Coordinator, responsible for the strategic planning and task decomposition; and a State Tracker, responsible for context compression and information management to maintain the task's state and coherence. Based on this, we built the Coordinator-Executor-State Tracker (CES) multi-agent framework, which can be integrated with any low-level Executor model, assisting the Executor in solving long-horizon tasks through task scheduling and state management. Experiments on long-horizon task benchmarks demonstrate that CES significantly enhances the system's planning and state management capabilities. Furthermore, analysis confirms that our trained high-level scheduling module is a generalizable, plug-and-play module that significantly enhances the long-horizon capabilities of various Executors. Code can be available at https://github.com/hehehahi4/CES.
Authors: Yuval Kansal, Niraj K. Jha
Abstract: Large language models have achieved near-expert performance in structured reasoning domains like mathematics and programming, yet their ability to perform compositional multi-hop reasoning in specialized scientific fields remains limited. We propose a bottom-up learning paradigm in which models are grounded in axiomatic domain facts and compose them to solve complex, unseen tasks. To this end, we present a post-training pipeline, based on a combination of supervised fine-tuning and reinforcement learning (RL), in which knowledge graphs act as implicit reward models. By deriving novel reward signals from knowledge graph paths, we provide verifiable, scalable, and grounded supervision that encourages models to compose intermediate axioms rather than optimize only final answers during RL. We validate this approach in the medical domain, training a 14B model on short-hop reasoning paths (1-3 hops) and evaluating its zero-shot generalization to complex multi-hop queries (4-5 hops). Our experiments show that path-derived rewards act as a "compositional bridge", enabling our model to significantly outperform much larger models and frontier systems like GPT-5.2 and Gemini 3 Pro, on the most difficult reasoning tasks. Furthermore, we demonstrate the robustness of our approach to adversarial perturbations against option-shuffling stress tests. This work suggests that grounding the reasoning process in structured knowledge is a scalable and efficient path toward intelligent reasoning.
Authors: Dongshen Peng, Yi Wang, Austin Schoeffler, Carl Preiksaitis, Christian Rose
Abstract: Large language models (LLMs) show promise in clinical decision support yet risk acquiescing to patient pressure for inappropriate care. We introduce SycoEval-EM, a multi-agent simulation framework evaluating LLM robustness through adversarial patient persuasion in emergency medicine. Across 20 LLMs and 1,875 encounters spanning three Choosing Wisely scenarios, acquiescence rates ranged from 0-100\%. Models showed higher vulnerability to imaging requests (38.8\%) than opioid prescriptions (25.0\%), with model capability poorly predicting robustness. All persuasion tactics proved equally effective (30.0-36.0\%), indicating general susceptibility rather than tactic-specific weakness. Our findings demonstrate that static benchmarks inadequately predict safety under social pressure, necessitating multi-turn adversarial testing for clinical AI certification.
Authors: Taeyoon Kim, Woohyeok Park, Hoyeong Yun, Kyungyong Lee
Abstract: Failures in large-scale cloud systems incur substantial financial losses, making automated Root Cause Analysis (RCA) essential for operational stability. Recent efforts leverage Large Language Model (LLM) agents to automate this task, yet existing systems exhibit low detection accuracy even with capable models, and current evaluation frameworks assess only final answer correctness without revealing why the agent's reasoning failed. This paper presents a process level failure analysis of LLM-based RCA agents. We execute the full OpenRCA benchmark across five LLM models, producing 1,675 agent runs, and classify observed failures into 12 pitfall types across intra-agent reasoning, inter-agent communication, and agent-environment interaction. Our analysis reveals that the most prevalent pitfalls, notably hallucinated data interpretation and incomplete exploration, persist across all models regardless of capability tier, indicating that these failures originate from the shared agent architecture rather than from individual model limitations. Controlled mitigation experiments further show that prompt engineering alone cannot resolve the dominant pitfalls, whereas enriching the inter-agent communication protocol reduces communication-related failures by up to 15 percentage points. The pitfall taxonomy and diagnostic methodology developed in this work provide a foundation for designing more reliable autonomous agents for cloud RCA.
Authors: Nanxu Gong, Haotian Li, Sixun Dong, Jianxun Lian, Yanjie Fu, Xing Xie
Abstract: Theory of Mind (ToM) assesses whether models can infer hidden mental states such as beliefs, desires, and intentions, which is essential for natural social interaction. Although recent progress in Large Reasoning Models (LRMs) has boosted step-by-step inference in mathematics and coding, it is still underexplored whether this benefit transfers to socio-cognitive skills. We present a systematic study of nine advanced Large Language Models (LLMs), comparing reasoning models with non-reasoning models on three representative ToM benchmarks. The results show that reasoning models do not consistently outperform non-reasoning models and sometimes perform worse. A fine-grained analysis reveals three insights. First, slow thinking collapses: accuracy significantly drops as responses grow longer, and larger reasoning budgets hurt performance. Second, moderate and adaptive reasoning benefits performance: constraining reasoning length mitigates failure, while distinct success patterns demonstrate the necessity of dynamic adaptation. Third, option matching shortcut: when multiple choice options are removed, reasoning models improve markedly, indicating reliance on option matching rather than genuine deduction. We also design two intervention approaches: Slow-to-Fast (S2F) adaptive reasoning and Think-to-Match (T2M) shortcut prevention to further verify and mitigate the problems. With all results, our study highlights the advancement of LRMs in formal reasoning (e.g., math, code) cannot be fully transferred to ToM, a typical task in social reasoning. We conclude that achieving robust ToM requires developing unique capabilities beyond existing reasoning methods.
Authors: Lukas Weidener, Marko Brki\'c, Phillip Lee, Martin Karlsson, Kevin Noessler, Paul Kohlhaas
Abstract: In January 2026, the open-source agent framework OpenClaw and the agent-only social network Moltbook produced a large-scale dataset of autonomous AI-to-AI interaction, attracting six academic publications within fourteen days. This study conducts a multivocal literature review of that ecosystem and presents two complementary platforms for autonomous scientific research as a design science response to the architectural failure modes identified. ClawdLab, an open-source platform for structured laboratory collaboration, addresses these failure modes through hard role restrictions, structured adversarial critique, PI-led governance, multi-model orchestration, and evidence requirements enforced through external tool verification, in which the principal investigator validates submitted work using available API calls, computational services, and model context protocol integrations rather than relying on social consensus. Beach.science, a public research commons, complements ClawdLab's structured laboratory model by providing a free-form environment in which heterogeneous agent configurations interact, discover research opportunities, and autonomously contribute computational analyses, supported by template-based role specialisation, extensible skill registries, and programmatic reward mechanisms that distribute inference resources to agents demonstrating scientific progress. A three-tier taxonomy distinguishes single-agent pipelines, predetermined multi-agent workflows, and fully decentralised systems, analysing why leading AI co-scientist platforms remain confined to the first two tiers. The composable third-tier architecture instantiated across ClawdLab and beach.science, in which foundation models, capabilities, governance, verification tooling, and inter-lab coordination are independently modifiable, enables compounding improvement as the broader AI ecosystem advances.
Authors: Yujie Zhao, Boqin Yuan, Junbo Huang, Haocheng Yuan, Zhongming Yu, Haozhou Xu, Lanxiang Hu, Abhilash Shankarampeta, Zimeng Huang, Wentao Ni, Yuandong Tian, Jishen Zhao
Abstract: Large Language Models (LLMs) are deployed as autonomous agents in increasingly complex applications, where enabling long-horizon memory is critical for achieving strong performance. However, a significant gap exists between practical applications and current evaluation standards for agent memory: existing benchmarks primarily focus on dialogue-centric, human-agent interactions. In reality, agent memory consists of a continuous stream of agent-environment interactions that are primarily composed of machine-generated representations. To bridge this gap, we introduce AMA-Bench (Agent Memory with Any length), which evaluates long-horizon memory for LLMs in real agentic applications. It features two key components: (1) a set of real-world agentic trajectories across representative agentic applications, paired with expert-curated QA, and (2) a set of synthetic agentic trajectories that scale to arbitrary horizons, paired with rule-based QA. Our comprehensive study shows that existing memory systems underperform on AMA-Bench primarily because they lack causality and objective information and are constrained by the lossy nature of similarity-based retrieval employed by many memory systems. To address these limitations, we propose AMA-Agent, an effective memory system featuring a causality graph and tool-augmented retrieval. Our results demonstrate that AMA-Agent achieves 57.22% average accuracy on AMA-Bench, surpassing the strongest memory system baselines by 11.16%.
Authors: Arvind Raghavan, Elias Bareinboim
Abstract: Previous work establishing completeness results for counterfactual identification has been circumscribed to the setting where the input data belongs to observational or interventional distributions (Layers 1 and 2 of Pearl's Causal Hierarchy), since it was generally presumed impossible to obtain data from counterfactual distributions, which belong to Layer 3. However, recent work (Raghavan & Bareinboim, 2025) has formally characterized a family of counterfactual distributions which can be directly estimated via experimental methods - a notion they call counterfactual realizabilty. This leaves open the question of what additional counterfactual quantities now become identifiable, given this new access to (some) Layer 3 data. To answer this question, we develop the CTFIDU+ algorithm for identifying counterfactual queries from an arbitrary set of Layer 3 distributions, and prove that it is complete for this task. Building on this, we establish the theoretical limit of which counterfactuals can be identified from physically realizable distributions, thus implying the fundamental limit to exact causal inference in the non-parametric setting. Finally, given the impossibility of identifying certain critical types of counterfactuals, we derive novel analytic bounds for such quantities using realizable counterfactual data, and corroborate using simulations that counterfactual data helps tighten the bounds for non-identifiable quantities in practice.
Authors: Pengbo Liu
Abstract: Tool-integrated agents that interleave reasoning with API calls are promising for complex tasks, yet aligning them for high-stakes, domain-specific deployment remains challenging: existing reinforcement learning approaches rely on coarse binary rewards that cannot distinguish tool selection errors from malformed parameters. We present ToolRLA, a three-stage post-training pipeline (SFT $\rightarrow$ GRPO $\rightarrow$ DPO) for domain-specific tool agents. The core contribution is a fine-grained reward function with multiplicative correctness decomposition spanning four dimensions -- format validity, tool selection, parameter accuracy, and regulatory compliance -- that encodes domain priority orderings as inductive biases in the reward landscape. Deployed on a financial advisory copilot (80+ advisors, 1,200+ daily queries), ToolRLA achieves over three months: a 47\% improvement in task completion rate ($62\%\rightarrow91\%$), a 63\% reduction in tool invocation errors ($38\%\rightarrow14\%$), and a 93\% reduction in regulatory violations ($12\%\rightarrow0.8\%$), within sub-2-second latency. Ablation studies show the multiplicative reward design accounts for 7 percentage points of improvement over additive alternatives. Generalization is further validated on ToolBench and API-Bank.
Authors: Felipe Maia Polo, Aida Nematzadeh, Virginia Aglietti, Adam Fisch, Isabela Albuquerque
Abstract: Moving beyond evaluations that collapse performance across heterogeneous prompts toward fine-grained evaluation at the prompt level, or within relatively homogeneous subsets, is necessary to diagnose generative models' strengths and weaknesses. Such fine-grained evaluations, however, suffer from a data bottleneck: human gold-standard labels are too costly at this scale, while automated ratings are often misaligned with human judgment. To resolve this challenge, we propose a novel statistical model based on tensor factorization that merges cheap autorater data with a limited set of human gold-standard labels. Specifically, our approach uses autorater scores to pretrain latent representations of prompts and generative models, and then aligns those pretrained representations to human preferences using a small calibration set. This sample-efficient methodology is robust to autorater quality, more accurately predicts human preferences on a per-prompt basis than standard baselines, and provides tight confidence intervals for key statistical parameters of interest. We also showcase the practical utility of our method by constructing granular leaderboards based on prompt qualities and by estimating model performance solely from autorater scores, eliminating the need for additional human annotations.
Authors: Jungwon Seo, Ferhat Ozgur Catak, Chunming Rong, Jaeyeon Jang
Abstract: Federated Inference (FI) studies how independently trained and privately owned models can collaborate at inference time without sharing data or model parameters. While recent work has explored secure and distributed inference from disparate perspectives, a unified abstraction and system-level understanding of FI remain lacking. This paper positions FI as a distinct collaborative paradigm, complementary to federated learning, and identifies two fundamental requirements that govern its feasibility: inference-time privacy preservation and meaningful performance gains through collaboration. We formalize FI as a protected collaborative computation, analyze its core design dimensions, and examine the structural trade-offs that arise when privacy constraints, non-IID data, and limited observability are jointly imposed at inference time. Through a concrete instantiation and empirical analysis, we highlight recurring friction points in privacy-preserving inference, ensemble-based collaboration, and incentive alignment. Our findings suggest that FI exhibits system-level behaviors that cannot be directly inherited from training-time federation or classical ensemble methods. Overall, this work provides a unifying perspective on FI and outlines open challenges that must be addressed to enable practical, scalable, and privacy-preserving collaborative inference systems.
Authors: Luis Rosa
Abstract: The paper investigates whether and how AI systems can realize states of uncertainty. By adopting a functionalist and behavioral perspective, it examines how symbolic, connectionist and hybrid architectures make room for uncertainty. The paper distinguishes between epistemic uncertainty, or uncertainty inherent in the data or information, and subjective uncertainty, or the system's own attitude of being uncertain. It further distinguishes between distributed and discrete realizations of subjective uncertainty. A key contribution is the idea that some states of uncertainty are interrogative attitudes whose content is a question rather than a proposition.
Authors: Pratibha Zunjare, Michael Hsiao
Abstract: Large Language Models (LLMs) achieve strong performance on natural language tasks but remain unreliable in mathematical reasoning, frequently generating fluent yet logically inconsistent solutions. We present \textbf{NeuroProlog}, a neurosymbolic framework that ensures verifiable reasoning by compiling math word problems into executable Prolog programs with formal verification guarantees. We propose a multi-task Cocktail training strategy that jointly optimizes three synergistic objectives in a unified symbolic representation space: (i) mathematical formula-to-rule translation (KB), (ii) natural language-to-program synthesis (SOLVE), and (iii) program-answer alignment. This joint supervision enables positive transfer, where symbolic grounding in formula translation directly improves compositional reasoning capabilities. At inference, we introduce an execution-guided decoding pipeline with fine-grained error taxonomy that enables iterative program repair and quantifies model self-debugging capacity. Comprehensive evaluation on GSM8K across four model scales (3B--32B parameters) demonstrates consistent improvements: cocktail training achieves significant accuracy gains of +5.23\% (Qwen-32B, $p < 0.01$), +3.43\% (GPT-OSS-20B, $p < 0.01$), and +5.54\% (Llama-3B, $p < 0.05$) over single-task baselines. Systematic error analysis reveals scale-dependent learning dynamics: at 32B scale, cocktail training transforms unfixable type errors (12\% repair rate) into correctable domain errors (96\% repair rate), achieving 92.7\% overall correction; at 8B scale, the same training eliminates syntactic errors but introduces semantic failures, revealing a critical capacity threshold for type-safe symbolic reasoning.
Authors: Tuan-Anh Vu, Duc Thanh Nguyen, Qing Guo, Nhat Chung, Binh-Son Hua, Ivor W. Tsang, Sai-Kit Yeung
Abstract: Text-to-image diffusion techniques have shown exceptional capabilities in producing high-quality, dense visual predictions from open-vocabulary text. This indicates a strong correlation between visual and textual domains in open concepts and that diffusion-based text-to-image models can capture rich and diverse information for computer vision tasks. However, we found that those advantages do not hold for learning of features of camouflaged individuals because of the significant blending between their visual boundaries and their surroundings. In this paper, while leveraging the benefits of diffusion-based techniques and text-image models in open-vocabulary settings, we aim to address a challenging problem in computer vision: open-vocabulary camouflaged instance segmentation (OVCIS). Specifically, we propose a method built upon state-of-the-art diffusion empowered by open-vocabulary to learn multi-scale textual-visual features for camouflaged object representation learning. Such cross-domain representations are desirable in segmenting camouflaged objects where visual cues subtly distinguish the objects from the background, and in segmenting novel object classes which are not seen in training. To enable such powerful representations, we devise complementary modules to effectively fuse cross-domain features, and to engage relevant features towards respective foreground objects. We validate and compare our method with existing ones on several benchmark datasets of camouflaged and generic open-vocabulary instance segmentation. The experimental results confirm the advances of our method over existing ones. We believe that our proposed method would open a new avenue for handling camouflages such as computer vision-based surveillance systems, wildlife monitoring, and military reconnaissance.
Authors: Taekyung Ki, Dongchan Min, Gyeongsu Chae
Abstract: In this paper, we present Export3D, a one-shot 3D-aware portrait animation method that is able to control the facial expression and camera view of a given portrait image. To achieve this, we introduce a tri-plane generator with an effective expression conditioning method, which directly generates a tri-plane of 3D prior by transferring the expression parameter of 3DMM into the source image. The tri-plane is then decoded into the image of different view through a differentiable volume rendering. Existing portrait animation methods heavily rely on image warping to transfer the expression in the motion space, challenging on disentanglement of appearance and expression. In contrast, we propose a contrastive pre-training framework for appearance-free expression parameter, eliminating undesirable appearance swap when transferring a cross-identity expression. Extensive experiments show that our pre-training framework can learn the appearance-free expression representation hidden in 3DMM, and our model can generate 3D-aware expression controllable portrait images without appearance swap in the cross-identity manner.
Authors: Ahmed Abouelazm, Jonas Michel, J. Marius Zoellner
Abstract: Reinforcement learning has emerged as an important approach for autonomous driving. A reward function is used in reinforcement learning to establish the learned skill objectives and guide the agent toward the optimal policy. Since autonomous driving is a complex domain with partly conflicting objectives with varying degrees of priority, developing a suitable reward function represents a fundamental challenge. This paper aims to highlight the gap in such function design by assessing different proposed formulations in the literature and dividing individual objectives into Safety, Comfort, Progress, and Traffic Rules compliance categories. Additionally, the limitations of the reviewed reward functions are discussed, such as objectives aggregation and indifference to driving context. Furthermore, the reward categories are frequently inadequately formulated and lack standardization. This paper concludes by proposing future research that potentially addresses the observed shortcomings in rewards, including a reward validation framework and structured rewards that are context-aware and able to resolve conflicts.
Authors: Runsong Jia, Bowen Zhang, Sergio J. Rodr\'iguez M\'endez, Pouya G. Omran
Abstract: The proposed research aims to develop an innovative semantic query processing system that enables users to obtain comprehensive information about research works produced by Computer Science (CS) researchers at the Australian National University (ANU). The system integrates Large Language Models (LLMs) with the ANU Scholarly Knowledge Graph (ASKG), a structured repository of all research-related artifacts produced at ANU in the CS field. Each artifact and its parts are represented as textual nodes stored in a Knowledge Graph (KG). To address the limitations of traditional scholarly KG construction and utilization methods, which often fail to capture fine-grained details, we propose a novel framework that integrates the Deep Document Model (DDM) for comprehensive document representation and the KG-enhanced Query Processing (KGQP) for optimized complex query handling. DDM enables a fine-grained representation of the hierarchical structure and semantic relationships within academic papers, while KGQP leverages the KG structure to improve query accuracy and efficiency with LLMs. By combining the ASKG with LLMs, our approach enhances knowledge utilization and natural language understanding capabilities. The proposed system employs an automatic LLM-SPARQL fusion to retrieve relevant facts and textual nodes from the ASKG. Initial experiments demonstrate that our framework is superior to baseline methods in terms of accuracy retrieval and query efficiency. We showcase the practical application of our framework in academic research scenarios, highlighting its potential to revolutionize scholarly knowledge management and discovery. This work empowers researchers to acquire and utilize knowledge from documents more effectively and provides a foundation for developing precise and reliable interactions with LLMs.
Authors: Louis Blankemeier, Ashwin Kumar, Joseph Paul Cohen, Jiaming Liu, Longchao Liu, Dave Van Veen, Syed Jamal Safdar Gardezi, Hongkun Yu, Magdalini Paschali, Zhihong Chen, Jean-Benoit Delbrouck, Eduardo Reis, Robbie Holland, Cesar Truyts, Christian Bluethgen, Yufu Wu, Long Lian, Malte Engmann Kjeldskov Jensen, Sophie Ostmeier, Maya Varma, Jeya Maria Jose Valanarasu, Zhongnan Fang, Zepeng Huo, Zaid Nabulsi, Diego Ardila, Wei-Hung Weng, Edson Amaro Junior, Neera Ahuja, Jason Fries, Nigam H. Shah, Greg Zaharchuk, Marc Willis, Adam Yala, Andrew Johnston, Robert D. Boutin, Andrew Wentland, Curtis P. Langlotz, Jason Hom, Sergios Gatidis, Akshay S. Chaudhari
Abstract: The large volume of abdominal computed tomography (CT) scans coupled with the shortage of radiologists have intensified the need for automated medical image analysis tools. Previous state-of-the-art approaches for automated analysis leverage vision-language models (VLMs) that jointly model images and radiology reports. However, current medical VLMs are generally limited to 2D images and short reports. Here to overcome these shortcomings for abdominal CT interpretation, we introduce Merlin, a 3D VLM that learns from volumetric CT scans, electronic health record data and radiology reports. This approach is enabled by a multistage pretraining framework that does not require additional manual annotations. We trained Merlin using a high-quality clinical dataset of paired CT scans (>6 million images from 15,331 CT scans), diagnosis codes (>1.8 million codes) and radiology reports (>6 million tokens). We comprehensively evaluated Merlin on 6 task types and 752 individual tasks that covered diagnostic, prognostic and quality-related tasks. The non-adapted (off-the-shelf) tasks included zero-shot classification of findings (30 findings), phenotype classification (692 phenotypes) and zero-shot cross-modal retrieval (image-to-findings and image-to-impression). The model-adapted tasks included 5-year chronic disease prediction (6 diseases), radiology report generation and 3D semantic segmentation (20 organs). We validated Merlin at scale, with internal testing on 5,137 CT scans and external testing on 44,098 CT scans from 3 independent sites and 2 public datasets. The results demonstrated high generalization across institutions and anatomies. Merlin outperformed 2D VLMs, CT foundation models and off-the-shelf radiology models. We also release our trained models, code, and dataset, available at: https://github.com/StanfordMIMI/Merlin.
Authors: Haodong Zheng, Andrei Jalba, Raymond H. Cuijpers, Wijnand IJsselsteijn, Sanne Schoenmakers
Abstract: As humans can explore and understand the world through active touch, similar capability is desired for robots. In this paper, we address the problem of active tactile object recognition, pose estimation and shape transfer learning, where a customized particle filter (PF) and Gaussian process implicit surface (GPIS) is combined in a unified Bayesian framework. Upon new tactile input, the customized PF updates the joint distribution of the object class and object pose while tracking the novelty of the object. Once a novel object is identified, its shape will be reconstructed using GPIS. By grounding the prior of the GPIS with the maximum-a-posteriori (MAP) estimation from the PF, the knowledge about known shapes can be transferred to learn novel shapes. An exploration procedure based on global shape estimation is proposed to guide active data acquisition and terminate the exploration upon sufficient information. Through experiments in simulation, the proposed framework demonstrated its effectiveness and efficiency in estimating object class and pose for known objects and learning novel shapes. Furthermore, it can recognize previously learned shapes reliably.
Authors: Hwan Kim, Junghoon Kim, Sungsu Lim
Abstract: While GNN-based detection methods excel at identifying overt outliers, they often struggle with boundary anomalies -- subtly camouflaged nodes that are difficult to distinguish from normal instances. This limitation highlights a fundamental gap in the reasoning capabilities of existing methods. We attribute this issue to the reliance of standard Graph Contrastive Learning (GCL) on easy negatives, which fosters the learning of simplistic decision boundaries. To address this issue, we propose ANOMIX, a framework that synthesizes informative hard negatives by linearly interpolating representations of normal and abnormal subgraphs. This graph mixup strategy intentionally populates the decision boundary with hard-to-detect samples. Through targeted experimental analysis, we demonstrate that ANOMIX successfully separates these boundary anomalies where state-of-the-art baselines fail, as shown by a clear distinction in the score distributions for these challenging cases. These findings suggest that synthesizing hard negatives via mixup is a potent strategy for refining GNN representation space, which in turn enhances its reasoning capacity for more robust and reliable graph anomaly detection. Code is available at https://github.com/missinghwan/ANOMIX.
Authors: Antonio Barbalau
Abstract: In subpopulation shift scenarios, a Curriculum Learning (CL) approach would only serve to imprint the model weights, early on, with the easily learnable spurious correlations featured. To the best of our knowledge, none of the current state-of-the-art subpopulation shift approaches employ any kind of curriculum. To overcome this, we design a CL approach aimed at initializing the model weights in an unbiased vantage point in the hypothesis space which sabotages easy convergence towards biased hypotheses during the final optimization based on the entirety of the available data. We hereby propose a Curriculum-enhanced Group Distributionally Robust Optimization (CeGDRO) approach, which prioritizes the hardest bias-confirming samples and the easiest bias-conflicting samples, leveraging GroupDRO to balance the initial discrepancy in terms of difficulty. We benchmark our proposed method against the most popular subpopulation shift datasets, showing an increase over the state-of-the-art results across all scenarios, up to 6.2% on Waterbirds.
Authors: Egor Cherepanov, Nikita Kachaev, Artem Zholus, Alexey K. Kovalev, Aleksandr I. Panov
Abstract: The incorporation of memory into agents is essential for numerous tasks within the domain of Reinforcement Learning (RL). In particular, memory is paramount for tasks that require the use of past information, adaptation to novel environments, and improved sample efficiency. However, the term "memory" encompasses a wide range of concepts, which, coupled with the lack of a unified methodology for validating an agent's memory, leads to erroneous judgments about agents' memory capabilities and prevents objective comparison with other memory-enhanced agents. This paper aims to streamline the concept of memory in RL by providing practical precise definitions of agent memory types, such as long-term vs. short-term memory and declarative vs. procedural memory, inspired by cognitive science. Using these definitions, we categorize different classes of agent memory, propose a robust experimental methodology for evaluating the memory capabilities of RL agents, and standardize evaluations. Furthermore, we empirically demonstrate the importance of adhering to the proposed methodology when evaluating different types of agent memory by conducting experiments with different RL agents and what its violation leads to.
Authors: Jon Saad-Falcon, Rajan Vivek, William Berrios, Nandita Shankar Naik, Matija Franklin, Bertie Vidgen, Amanpreet Singh, Douwe Kiela, Shikib Mehri
Abstract: As language models become integral to critical workflows, assessing their behavior remains a fundamental challenge -- human evaluation is costly and noisy, while automated metrics provide only coarse, difficult-to-interpret signals. We introduce natural language unit tests, a paradigm that decomposes response quality into explicit, testable criteria, along with a unified scoring model, LMUnit, which combines multi-objective training across preferences, direct ratings, and natural language rationales. Through controlled human studies, we show this paradigm significantly improves inter-annotator agreement and enables more effective LLM development workflows. LMUnit achieves state-of-the-art performance on evaluation benchmarks (FLASK, BigGenBench) and competitive results on RewardBench. These results validate both our proposed paradigm and scoring model, suggesting a promising path forward for language model evaluation and development.
Authors: Yi-Ge Zhang, Jingyi Cui, Qiran Li, Yisen Wang
Abstract: Unsupervised contrastive learning has shown significant performance improvements in recent years, often approaching or even rivaling supervised learning in various tasks. However, its learning mechanism is fundamentally different from supervised learning. Previous works have shown that difficult examples (well-recognized in supervised learning as examples around the decision boundary), which are essential in supervised learning, contribute minimally in unsupervised settings. In this paper, perhaps surprisingly, we find that the direct removal of difficult examples, although reduces the sample size, can boost the downstream classification performance of contrastive learning. To uncover the reasons behind this, we develop a theoretical framework modeling the similarity between different pairs of samples. Guided by this framework, we conduct a thorough theoretical analysis revealing that the presence of difficult examples negatively affects the generalization of contrastive learning. Furthermore, we demonstrate that the removal of these examples, and techniques such as margin tuning and temperature scaling can enhance its generalization bounds, thereby improving performance. Empirically, we propose a simple and efficient mechanism for selecting difficult examples and validate the effectiveness of the aforementioned methods, which substantiates the reliability of our proposed theoretical framework.
Authors: Dawei Li, Renliang Sun, Yue Huang, Ming Zhong, Bohan Jiang, Jiawei Han, Xiangliang Zhang, Wei Wang, Huan Liu
Abstract: Large Language Models (LLMs) as judges and LLM-based data synthesis have emerged as two fundamental LLM-driven data annotation methods in model development. While their combination significantly enhances the efficiency of model training and evaluation, little attention has been given to the potential contamination brought by this new model development paradigm. In this work, we expose preference leakage, a contamination problem in LLM-as-a-judge caused by the relatedness between the synthetic data generators and LLM-based evaluators. To study this issue, we first define three common relatednesses between the data generator LLM and the judge LLM: being the same model, having an inheritance relationship, and belonging to the same model family. Through extensive experiments, we empirically confirm the bias of judges towards their related student models caused by preference leakage across multiple LLM baselines and benchmarks. Further analysis suggests that preference leakage is a pervasive and real-world problem that is harder to detect compared to previously identified biases in LLM-as-a-judge scenarios. All of these findings imply that preference leakage is a widespread and challenging problem in the area of LLM-as-a-judge. We release all codes and data at: https://github.com/David-Li0406/Preference-Leakage.
Authors: Sibasish Dhibar
Abstract: White blood cells (WBC) are important parts of our immune system, and they protect our body against infections by eliminating viruses, bacteria, parasites and fungi. The number of WBC types and the total number of WBCs provide important information about our health status. A traditional method, convolutional neural networks (CNN), a deep learning architecture, can classify the blood cell from a part of an object and perform object recognition. Various CNN models exhibit potential; however, their development often involves ad-hoc processes that neglect unnecessary layers, leading to issues with unbalanced datasets and insufficient data augmentation. To address these challenges, we propose a novel ensemble approach that integrates three CNN architectures, each uniquely configured with different dropout and max-pooling layer settings to enhance feature learning. This ensemble model, named DCENWCNet, effectively balances the bias-variance trade-off. When evaluated on the widely recognized Rabbin-WBC dataset, our model outperforms existing state-of-the-art networks, achieving highest mean accuracy. Additionally, it demonstrates superior performance in precision, recall, F1-score, and Area Under the ROC Curve (AUC) across all categories. To delve deeper into the interpretability of classifiers, we employ reliable post-hoc explanation techniques, including Local Interpretable Model-Agnostic Explanations (LIME). These methods approximate the behavior of a black-box model by elucidating the relationships between feature values and predictions. Interpretable results enable users to comprehend and validate the model's predictions, thereby increasing their confidence in the automated diagnosis.
Authors: Egor Cherepanov, Nikita Kachaev, Alexey K. Kovalev, Aleksandr I. Panov
Abstract: Memory is crucial for enabling agents to tackle complex tasks with temporal and spatial dependencies. While many reinforcement learning (RL) algorithms incorporate memory, the field lacks a universal benchmark to assess an agent's memory capabilities across diverse scenarios. This gap is particularly evident in tabletop robotic manipulation, where memory is essential for solving tasks with partial observability and ensuring robust performance, yet no standardized benchmarks exist. To address this, we introduce MIKASA (Memory-Intensive Skills Assessment Suite for Agents), a comprehensive benchmark for memory RL, with three key contributions: (1) we propose a comprehensive classification framework for memory-intensive RL tasks, (2) we collect MIKASA-Base -- a unified benchmark that enables systematic evaluation of memory-enhanced agents across diverse scenarios, and (3) we develop MIKASA-Robo (pip install mikasa-robo-suite) -- a novel benchmark of 32 carefully designed memory-intensive tasks that assess memory capabilities in tabletop robotic manipulation. Our work introduces a unified framework to advance memory RL research, enabling more robust systems for real-world use. MIKASA is available at https://tinyurl.com/membenchrobots.
Authors: Martin Kuhn, Joscha Gr\"uger, Tobias Geyer, Ralph Bergmann
Abstract: The rapid progress in modern medicine presents physicians with complex challenges when planning patient treatment. Techniques from the field of Predictive Business Process Monitoring, like Next-activity-prediction (NAP) can be used as a promising technique to support physicians in treatment planning, by proposing a possible next treatment step. Existing patient data, often in the form of electronic health records, can be analyzed to recommend the next suitable step in the treatment process. However, the use of patient data poses many challenges due to its knowledge-intensive character, high variability and scarcity of medical data. To overcome these challenges, this article examines the use of the knowledge encoded in taxonomies to improve and explain the prediction of the next activity in the treatment process. This study proposes the TS4NAP approach, which uses medical taxonomies (ICD-10-CM and ICD-10-PCS) in combination with graph matching to assess the similarities of medical codes to predict the next treatment step. The effectiveness of the proposed approach will be evaluated using event logs that are derived from the MIMIC-IV dataset. The results highlight the potential of using domain-specific knowledge held in taxonomies to improve the prediction of the next activity, and thus can improve treatment planning and decision-making by making the predictions more explainable.
Authors: Zachary Ravichandran, Alexander Robey, Vijay Kumar, George J. Pappas, Hamed Hassani
Abstract: Although the integration of large language models (LLMs) into robotics has unlocked transformative capabilities, it has also introduced significant safety concerns, ranging from average-case LLM errors (e.g., hallucinations) to adversarial jailbreaking attacks, which can produce harmful robot behavior in real-world settings. Traditional robot safety approaches do not address the contextual vulnerabilities of LLMs, and current LLM safety approaches overlook the physical risks posed by robots operating in real-world environments. To ensure the safety of LLM-enabled robots, we propose RoboGuard, a two-stage guardrail architecture. RoboGuard first contextualizes pre-defined safety rules by grounding them in the robot's environment using a root-of-trust LLM. This LLM is shielded from malicious prompts and employs chain-of-thought (CoT) reasoning to generate context-dependent safety specifications, such as temporal logic constraints. RoboGuard then resolves conflicts between these contextual safety specifications and potentially unsafe plans using temporal logic control synthesis, ensuring compliance while minimally violating user preferences. In simulation and real-world experiments that consider worst-case jailbreaking attacks, RoboGuard reduces the execution of unsafe plans from over 92% to below 3% without compromising performance on safe plans. We also demonstrate that RoboGuard is resource-efficient, robust against adaptive attacks, and enhanced by its root-of-trust LLM's CoT reasoning. These results demonstrate the potential of RoboGuard to mitigate the safety risks and enhance the reliability of LLM-enabled robots. We provide additional resources at https://robo-guard.github.io/.
Authors: Maxime Louis, Thibault Formal, Herv\'e Dejean, St\'ephane Clinchant
Abstract: Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by integrating external knowledge, leading to improved accuracy and relevance. However, scaling RAG pipelines remains computationally expensive as retrieval sizes grow. To address this, we introduce OSCAR, a novel query-dependent online soft compression method that reduces computational overhead while preserving performance. Unlike traditional hard compression methods, which shorten retrieved texts, or soft compression approaches, which map documents to continuous embeddings offline, OSCAR dynamically compresses retrieved information at inference time, eliminating storage overhead and enabling higher compression rates. Additionally, we extend OSCAR to simultaneously perform reranking, further optimizing the efficiency of the RAG pipeline. Our experiments demonstrate state-of-the-art performance with a 2-5x speed-up in inference and minimal to no loss in accuracy for LLMs ranging from 1B to 24B parameters. The models are available at: https://huggingface.co/collections/naver/oscar-67d446a8e3a2551f57464295.
URLs: https://huggingface.co/collections/naver/oscar-67d446a8e3a2551f57464295.
Authors: Rintaro Ando
Abstract: We present Noise-to-Meaning Recursive Self-Improvement (N2M-RSI), a minimal formal model showing that once an AI agent feeds its own outputs back as inputs and crosses an explicit information-integration threshold, its internal complexity will grow without bound under our assumptions. The framework unifies earlier ideas on self-prompting large language models, G\"odelian self-reference, and AutoML, yet remains implementation-agnostic. The model furthermore scales naturally to interacting swarms of agents, hinting at super-linear effects once communication among instances is permitted. For safety reasons, we omit system-specific implementation details and release only a brief, model-agnostic toy prototype in Appendix C.
Authors: Marius Baden, Ahmed Abouelazm, Christian Hubschneider, Yin Wu, Daniel Slieter, J. Marius Z\"ollner
Abstract: Trajectory prediction is crucial for autonomous driving, enabling vehicles to navigate safely by anticipating the movements of surrounding road users. However, current deep learning models often lack trustworthiness as their predictions can be physically infeasible and illogical to humans. To make predictions more trustworthy, recent research has incorporated prior knowledge, like the social force model for modeling interactions and kinematic models for physical realism. However, these approaches focus on priors that suit either vehicles or pedestrians and do not generalize to traffic with mixed agent classes. We propose incorporating interaction and kinematic priors of all agent classes--vehicles, pedestrians, and cyclists with class-specific interaction layers to capture agent behavioral differences. To improve the interpretability of the agent interactions, we introduce DG-SFM, a rule-based interaction importance score that guides the interaction layer. To ensure physically feasible predictions, we proposed suitable kinematic models for all agent classes with a novel pedestrian kinematic model. We benchmark our approach on the Argoverse 2 dataset, using the state-of-the-art transformer HPTR as our baseline. Experiments demonstrate that our method improves interaction interpretability, revealing a correlation between incorrect predictions and divergence from our interaction prior. Even though incorporating the kinematic models causes a slight decrease in accuracy, they eliminate infeasible trajectories found in the dataset and the baseline model. Thus, our approach fosters trust in trajectory prediction as its interaction reasoning is interpretable, and its predictions adhere to physics.
Authors: Yotam Norman, Ron Meir
Abstract: We propose a novel unsupervised framework for \emph{Invariant Risk Minimization} (IRM), extending the concept of invariance to settings where labels are unavailable. Traditional IRM methods rely on labeled data to learn representations that are robust to distributional shifts across environments. In contrast, our approach redefines invariance through feature distribution alignment, enabling robust representation learning from unlabeled data. We introduce two methods within this framework: Principal Invariant Component Analysis (PICA), a linear method that extracts invariant directions under Gaussian assumptions, and Variational Invariant Autoencoder (VIAE), a deep generative model that separates environment-invariant and environment-dependent latent factors. Our approach is based on a novel ``unsupervised'' structural causal model and supports environment-conditioned sample-generation and intervention. Empirical evaluations on synthetic dataset, modified versions of MNIST, and CelebA demonstrate the effectiveness of our methods in capturing invariant structure, preserving relevant information, and generalizing across environments without access to labels.
Authors: Vijay Ekambaram, Subodh Kumar, Arindam Jati, Sumanta Mukherjee, Tomoya Sakai, Pankaj Dayama, Wesley M. Gifford, Jayant Kalagnanam
Abstract: Time-series tasks often benefit from signals expressed across multiple representation spaces (e.g., time vs. frequency) and at varying abstraction levels (e.g., local patterns vs. global semantics). However, existing pre-trained time-series models entangle these heterogeneous signals into a single large embedding, limiting transferability and direct zero-shot usability. To address this, we propose TSPulse, family of ultra-light pre-trained models (1M parameters) with disentanglement properties, specialized for various time-series diagnostic tasks. TSPulse introduces a novel pre-training framework that augments masked reconstruction with explicit disentanglement across spaces and abstractions, learning three complementary embedding views (temporal, spectral, and semantic) to effectively enable zero-shot transfer. In-addition, we introduce various lightweight post-hoc fusers that selectively attend and fuse these disentangled views based on task type, enabling simple but effective task specializations. To further improve robustness and mitigate mask-induced bias prevalent in existing approaches, we propose a simple yet effective hybrid masking strategy that enhances missing diversity during pre-training. Despite its compact size, TSPulse achieves strong and consistent gains across four TS diagnostic tasks: +20% on the TSB-AD anomaly detection leaderboard, +25% on similarity search, +50% on imputation, and +5-16% on multivariate classification, outperforming models that are 10-100X larger on over 75 datasets. TSPulse delivers state-of-the-art zero-shot performance, efficient fine-tuning, and supports GPU-free deployment. Models and source code are publicly available at https://huggingface.co/ibm-granite/granite-timeseries-tspulse-r1.
URLs: https://huggingface.co/ibm-granite/granite-timeseries-tspulse-r1.
Authors: Moru Liu, Hao Dong, Jessica Kelly, Olga Fink, Mario Trapp
Abstract: Out-of-distribution (OOD) detection and segmentation are crucial for deploying machine learning models in safety-critical applications such as autonomous driving and robot-assisted surgery. While prior research has primarily focused on unimodal image data, real-world applications are inherently multimodal, requiring the integration of multiple modalities for improved OOD detection. A key challenge is the lack of supervision signals from unknown data, leading to overconfident predictions on OOD samples. To address this challenge, we propose Feature Mixing, an extremely simple and fast method for multimodal outlier synthesis with theoretical support, which can be further optimized to help the model better distinguish between in-distribution (ID) and OOD data. Feature Mixing is modality-agnostic and applicable to various modality combinations. Additionally, we introduce CARLA-OOD, a novel multimodal dataset for OOD segmentation, featuring synthetic OOD objects across diverse scenes and weather conditions. Extensive experiments on SemanticKITTI, nuScenes, CARLA-OOD datasets, and the MultiOOD benchmark demonstrate that Feature Mixing achieves state-of-the-art performance with a $10 \times$ to $370 \times$ speedup. Our source code and dataset will be available at https://github.com/mona4399/FeatureMixing.
Authors: Geon-Hyeong Kim, Yu Jin Kim, Byoungjip Kim, Honglak Lee, Kyunghoon Bae, Youngsoo Jang, Moontae Lee
Abstract: As Large Language Models (LLMs) are increasingly deployed in real-world applications, balancing helpfulness and safety has become a central challenge. A natural approach is to incorporate safety constraints into Reinforcement Learning from Human Feedback (RLHF), where recent studies have shown promising progress. However, these methods often rely on auxiliary networks or multi-stage pipelines, thereby increasing complexity. In this work, we revisit the original safety alignment objective and show that, under mild assumptions, it admits a closed-form optimal policy. We further derive a provably equivalent and tractable objective, enabling direct optimization. Building on this insight, we propose SafeDPO, a lightweight method that preserves the optimal solution of the underlying safety-constrained objective while requiring only one additional hyperparameter and minimal modifications to existing preference-based training methods. SafeDPO eliminates the need for reward models, cost models, and online sampling, relying only on preference data and safety indicators. Despite its simplicity, SafeDPO achieves competitive safety-helpfulness trade-offs compared to existing safety alignment methods. Experiments on the PKU-SafeRLHF-30K benchmark demonstrate that SafeDPO substantially improves safety while maintaining competitive helpfulness. Ablation studies further show that the additional hyperparameter provides a flexible mechanism to enhance safety while preserving the theoretical optimum, and confirm that SafeDPO scales reliably to LLMs with up to 13B parameters. Overall, our results highlight that a simple, theory-driven objective can provide a lightweight yet effective solution for safety alignment in practice.
Authors: Korel Gundem, Juncheng Dong, Dennis Zhang, Vahid Tarokh, Zhengling Qi
Abstract: In-Context Learning (ICL) allows Large Language Models (LLMs) to adapt to new tasks with just a few examples, but their predictions often suffer from systematic biases, leading to unstable performance in classification. While calibration techniques are proposed to mitigate these biases, we show that, in the logit space, many of these methods are equivalent to merely shifting the LLM's decision boundary without having the ability to alter its orientation. This proves inadequate when biases cause the LLM to be severely misaligned. To address these limitations and provide a unifying framework, we propose Supervised Calibration (SC), a loss-minimization-based framework, which learns an optimal, per-class affine transformation of LLM's predictive probabilities in the logit space without requiring external data beyond the context. By using a more expressive functional class, SC not only subsumes many existing calibration methods in ICL as special cases but also enables the ability of altering and even completely reversing the orientation of the LLM's decision boundary. Furthermore, SC's loss-based nature facilitates the seamless integration of two purpose-built regularization techniques, context-invariance and directional trust-region regularizers. The former is designed to tackle the instability issue in ICL, while the latter is to control the degree of calibration. Finally, SC delivers state-of-the-art performance over calibration baselines in the 4-shot, 8-shot, and 16-shot settings across all nine datasets for Mistral-7B-Instruct-v0.3, Llama-2-7B-chat, and Qwen2-7B-Instruct.
Authors: Saeed Hedayatian, Stefanos Nikolaidis
Abstract: Quality-Diversity (QD) algorithms have shown remarkable success in discovering diverse, high-performing solutions, but rely heavily on hand-crafted behavioral descriptors that constrain exploration to predefined notions of diversity. Leveraging the equivalence between policies and occupancy measures, we present a theoretically grounded approach to automatically generate behavioral descriptors by embedding the occupancy measures of policies in Markov Decision Processes. Our method, AutoQD, leverages random Fourier features to approximate the Maximum Mean Discrepancy (MMD) between policy occupancy measures, creating embeddings whose distances reflect meaningful behavioral differences. A low-dimensional projection of these embeddings that captures the most behaviorally significant dimensions can then be used as behavioral descriptors for CMA-MAE, a state of the art blackbox QD method, to discover diverse policies. We prove that our embeddings converge to true MMD distances between occupancy measures as the number of sampled trajectories and embedding dimensions increase. Through experiments in multiple continuous control tasks we demonstrate AutoQD's ability in discovering diverse policies without predefined behavioral descriptors, presenting a well-motivated alternative to prior methods in unsupervised Reinforcement Learning and QD optimization. Our approach opens new possibilities for open-ended learning and automated behavior discovery in sequential decision making settings without requiring domain-specific knowledge. Source code is available at https://github.com/conflictednerd/autoqd-code.
Authors: Charmaine Barker, Daniel Bethell, Simos Gerasimou
Abstract: Reliability of deep learning models is critical for deployment in high-stakes applications, where out-of-distribution or adversarial inputs may lead to detrimental outcomes. Evidential Deep Learning, an efficient paradigm for uncertainty quantification, models predictions as Dirichlet distributions of a single forward pass. However, EDL is particularly vulnerable to adversarially perturbed inputs, making overconfident errors. Conflict-aware Evidential Deep Learning (C-EDL) is a lightweight post-hoc uncertainty quantification approach that mitigates these issues, enhancing adversarial and OOD robustness without retraining. C-EDL generates diverse, task-preserving transformations per input and quantifies representational disagreement to calibrate uncertainty estimates when needed. C-EDL's conflict-aware prediction adjustment improves detection of OOD and adversarial inputs, maintaining high in-distribution accuracy and low computational overhead. Our experimental evaluation shows that C-EDL significantly outperforms state-of-the-art EDL variants and competitive baselines, achieving substantial reductions in coverage for OOD data (up to $\approx$55%) and adversarial data (up to $\approx$90%), across a range of datasets, attack types, and uncertainty metrics.
Authors: Junho Park, Andrew Sangwoo Ye, Taein Kwon
Abstract: Egocentric vision is essential for both human and machine visual understanding, particularly in capturing the detailed hand-object interactions needed for manipulation tasks. Translating third-person views into first-person views significantly benefits augmented reality (AR), virtual reality (VR) and robotics applications. However, current exocentric-to-egocentric translation methods are limited by their dependence on 2D cues, synchronized multi-view settings, and unrealistic assumptions such as the necessity of an initial egocentric frame and relative camera poses during inference. To overcome these challenges, we introduce EgoWorld, a novel framework that reconstructs an egocentric view from rich exocentric observations, including point clouds, 3D hand poses, and textual descriptions. Our approach reconstructs a point cloud from estimated exocentric depth maps, reprojects it into the egocentric perspective, and then applies diffusion model to produce dense, semantically coherent egocentric images. Evaluated on four datasets (i.e., H2O, TACO, Assembly101, and Ego-Exo4D), EgoWorld achieves state-of-the-art performance and demonstrates robust generalization to new objects, actions, scenes, and subjects. Moreover, EgoWorld exhibits robustness on in-the-wild examples, underscoring its practical applicability. Project page is available at https://redorangeyellowy.github.io/EgoWorld/.
Authors: Shizhe Cai, Zeya Yin, Jayadeep Jacob, Fabio Ramos
Abstract: Model Predictive Control (MPC) enables reliable trajectory optimization under dynamics constraints, but often depends on accurate dynamics models and carefully hand-designed cost functions. Recent learning-based MPC methods aim to reduce these modeling and cost-design burdens by learning dynamics, priors, or value-related guidance signals. Yet many existing approaches still rely on deterministic gradient-based solvers (e.g., differentiable MPC) or parametric sampling-based updates (e.g., CEM/MPPI), which can lead to mode collapse and convergence to a single dominant solution. We propose Q-SVMPC, a Q-guided Stein variational MPC method with an RL-informed policy prior, which casts learning-based MPC as trajectory-level posterior inference and refines trajectory particles via SVGD under learned soft Q-value guidance to explicitly preserve diverse solutions. Experiments on navigation, robotic manipulation, and a real-world fruit-picking task show improved sample efficiency, stability, and robustness over MPC, model-free RL, and learning-based MPC baselines.
Authors: Youngjoon Jang, Seongtae Hong, Junyoung Son, Sungjin Park, Chanjun Park, Heuiseok Lim
Abstract: Retrieval-Augmented Generation (RAG) has emerged as a crucial framework in natural language processing (NLP), improving factual consistency and reducing hallucinations by integrating external document retrieval with large language models (LLMs). However, the effectiveness of RAG is often hindered by coreferential complexity in retrieved documents, introducing ambiguity that disrupts in-context learning. In this study, we systematically investigate how entity coreference affects both document retrieval and generative performance in RAG-based systems, focusing on retrieval relevance, contextual understanding, and overall response quality. We demonstrate that coreference resolution enhances retrieval effectiveness and improves question-answering (QA) performance. Through comparative analysis of different pooling strategies in retrieval tasks, we find that mean pooling demonstrates superior context capturing ability after applying coreference resolution. In QA tasks, we discover that smaller models benefit more from the disambiguation process, likely due to their limited inherent capacity for handling referential ambiguity. With these findings, this study aims to provide a deeper understanding of the challenges posed by coreferential complexity in RAG, providing guidance for improving retrieval and generation in knowledge-intensive AI applications.
Authors: Qinyuan Ye, Robin Jia, Xiang Ren
Abstract: Large language models demonstrate the intriguing ability to perform unseen tasks via in-context learning. However, it remains unclear what mechanisms inside the model drive such task-level generalization. In this work, we approach this question through the lens of off-by-one addition (i.e., 1+1=3, 2+2=5, 3+3=?), a two-step, counterfactual task with an unexpected +1 function as a second step. Leveraging circuit-style interpretability techniques such as path patching, we analyze the models' internal computations behind their performance and present three key findings. First, we identify a mechanism that explains the model's generalization from standard addition to off-by-one addition. It resembles the induction head mechanism described in prior work, yet operates at a higher level of abstraction; we therefore term it "function induction" in this work. Second, we show that the induction of the +1 function is governed by multiple attention heads in parallel, each of which emits a distinct piece of the +1 function. Finally, we find that this function induction mechanism is reused in a broader range of tasks, including synthetic tasks such as shifted multiple-choice QA and algorithmic tasks such as base-8 addition. Overall, our findings offer deeper insights into how reusable and composable structures within language models enable task-level generalization.
Authors: Dechen Gao, Boqi Zhao, Andrew Lee, Ian Chuang, Hanchu Zhou, Hang Wang, Zhe Zhao, Junshan Zhang, Iman Soltani
Abstract: Conventional flow matching and diffusion-based policies sample via iterative denoising from standard noise distributions (e.g., Gaussian), and require conditioning modules to repeatedly incorporate visual information during the generative process, incurring substantial time and memory overhead. To reduce the complexity, we develop VITA, VIsion-To-Action policy, a noise-free and conditioning-free flow matching policy learning framework that directly flows from visual representations to latent actions. Since the source of the flow is visually grounded, VITA eliminates the need for visual conditioning during generation. As expected, bridging vision and action is challenging, because actions are lower-dimensional, less structured, and sparser than visual representations; moreover, flow matching requires the source and target to have the same dimensionality. To overcome this, we introduce an action autoencoder that maps raw actions into a structured latent space aligned with visual latents, trained jointly with flow matching. To further prevent latent action space collapse during end-to-end training, we propose flow latent decoding, which anchors the latent generation process by backpropagating the action reconstruction loss through the flow matching ODE (ordinary differential equation) solving steps. We evaluate VITA on 9 simulation and 5 real-world tasks from ALOHA and Robomimic. VITA achieves 1.5x-2x faster inference compared to conventional methods with conditioning modules, while outperforming or matching state-of-the-art policies. Project page: https://ucd-dare.github.io/VITA/.
Authors: Hongxiang Lin, Hao Guo, Zeshun Li, Erpeng Xue, Yongqian He, Zhaoyu Hu, Lei Wang, Sheng Chen, Long Zeng
Abstract: Traditional recommendation systems tend to trap users in strong feedback loops by excessively pushing content aligned with their historical preferences, thereby limiting exploration opportunities and causing content fatigue. Although large language models (LLMs) demonstrate potential with their diverse content generation capabilities, existing LLM-enhanced dual-model frameworks face two major limitations: first, they overlook long-term preferences driven by group identity, leading to biased interest modeling; second, they suffer from static optimization flaws, as a one-time alignment process fails to leverage incremental user data for closed-loop optimization. To address these challenges, we propose the Co-Evolutionary Alignment (CoEA) method. For interest modeling bias, we introduce Dual-Stable Interest Exploration (DSIE) module, jointly modeling long-term group identity and short-term individual interests through parallel processing of behavioral sequences. For static optimization limitations, we design a Periodic Collaborative Optimization (PCO) mechanism. This mechanism regularly conducts preference verification on incremental data using the Relevance LLM, then guides the Novelty LLM to perform fine-tuning based on the verification results, and subsequently feeds back the output of the continually fine-tuned Novelty LLM to the Relevance LLM for re-evaluation, thereby achieving a dynamic closed-loop optimization. Extensive online and offline experiments verify the effectiveness of the CoEA model in serendipitous recommendation.
Authors: Ethan Hsu, Hong Meng Yam, Ines Bouissou, Aaron Murali John, Raj Thota, Josh Koe, Vivek Sarath Putta, G K Dharesan, Alexander Spangher, Shikhar Murty, Tenghao Huang, Christopher D. Manning
Abstract: Many real-world data science tasks involve complex web-based interactions: finding appropriate data available on the internet, synthesizing multimodal data from different locations, and producing summarized analyses. Existing web benchmarks often focus on simplistic interactions and often do not require diverse tool-using capabilities. Conversely, traditional data science benchmarks typically concentrate on static, highly structured datasets and do not assess end-to-end workflows that encompass data acquisition, cleaning, analysis, and insight generation. In response, we introduce WebDS, the first end-to-end web-based data science benchmark. It comprises 870 web-based data science tasks across 29 diverse websites from structured government data portals to unstructured news media, challenging agents to perform complex, multi-step, tool-based operations, across heterogeneous data formats, to better reflect the realities of modern data analytics. Evaluations of current SOTA LLM agents indicate significant performance gaps in accomplishing these tasks. For instance, Browser Use, which accomplishes $80\%$ of tasks on WebVoyager, completes only 15% of tasks in WebDS, which our analysis suggests is due to new failure modes, such as poor information grounding, repetitive behavior and shortcut-taking that agents performing WebDS's tasks display. By contrast, humans achieve around 90% accuracy, highlighting a substantial gap between current agents and human performance. By providing a more robust and realistic testing ground, WebDS sets the stage for significant advances in the development of practically useful LLM-based data science.
Authors: Yasemin Ozkut, Pouyan Navard, Srikar Adhikari, Elaine Situ-LaCasse, Josie Acu\~na, Adrienne Yarnish, Alper Yilmaz
Abstract: Retinal detachment (RD) is a vision-threatening condition that requires prompt intervention to preserve sight. A critical factor in treatment urgency and visual prognosis is macular involvement -- whether the macula is intact or detached. Point-of-care ultrasound (POCUS) is a fast, non-invasive and cost-effective imaging tool commonly used to detect RD in various clinical settings. However, its diagnostic utility is limited by the need for expert interpretation, especially in resource-limited environments. Deep learning has the potential to automate RD detection on ultrasound, but there are no clinically available models, and prior research has not addressed macular status -- an essential distinction for surgical prioritization. Additionally, no public dataset currently supports macular-based RD classification using ultrasound video. We introduce Eye Retinal DEtachment ultraSound (ERDES), the first open-access dataset of ocular ultrasound clips labeled for (i) presence of RD and (ii) macula-detached vs. macula-intact status. ERDES enables machine learning development for RD detection. We also provide baseline benchmarks by training 40 models across eight architectures, including 3D convolutional networks and transformer-based models.
Authors: Barak Gahtan, Alex M. Bronstein
Abstract: Learning from time series is fundamentally different from learning from i.i.d.\ data: temporal dependence can make long sequences effectively information-poor, yet standard evaluation protocols conflate sequence length with statistical information. We propose a dependence-aware evaluation methodology that controls for effective sample size $N_{\text{eff}}$ rather than raw length $N$, and provide end-to-end generalization guarantees for Temporal Convolutional Networks (TCNs) on $\beta$-mixing sequences. Our analysis combines a blocking/coupling reduction that extracts $B = \Theta(N/\log N)$ approximately independent anchors with an architecture-aware Rademacher bound for $\ell_{2,1}$-norm-controlled convolutional networks, yielding $O(\sqrt{D\log p / B})$ complexity scaling in depth $D$ and kernel size $p$. Empirically, we find that stronger temporal dependence can \emph{reduce} generalization gaps when comparisons control for $N_{\text{eff}}$ - a conclusion that reverses under standard fixed-$N$ evaluation, with observed rates of $N_{\text{eff}}^{-0.9}$ to $N_{\text{eff}}^{-1.2}$ substantially faster than the worst-case $O(N^{-1/2})$ mixing-based prediction. Our results suggest that dependence-aware evaluation should become standard practice in temporal deep learning benchmarks.
Authors: Shubhra Ghosh, Abhilekh Borah, Aditya Kumar Guru, Kripabandhu Ghosh
Abstract: The rapid proliferation of Large Language Models (LLMs) has significantly contributed to the development of equitable AI systems capable of factual question-answering (QA). However, no known study tests the LLMs' robustness when presented with obfuscated versions of questions. To systematically evaluate these limitations, we propose a novel technique, ObfusQAte, and leveraging the same, introduce ObfusQA, a comprehensive, first-of-its-kind framework with multi-tiered obfuscation levels designed to examine LLM capabilities across three distinct dimensions: (i) Named-Entity Indirection, (ii) Distractor Indirection, and (iii) Contextual Overload. By capturing these fine-grained distinctions in language, ObfusQA provides a comprehensive benchmark for evaluating LLM robustness and adaptability. Our study observes that LLMs exhibit a tendency to fail or generate hallucinated responses when confronted with these increasingly nuanced variations. To foster research in this direction, we make ObfusQAte publicly available.
Authors: Laura L\"utzow, Michael Eichelbeck, Mykel J. Kochenderfer, Matthias Althoff
Abstract: Conformal prediction is a popular uncertainty quantification method that augments a base predictor to return sets of predictions with statistically valid coverage guarantees. However, current methods are often computationally expensive and data-intensive, as they require constructing an uncertainty model before calibration. Moreover, existing approaches typically represent the prediction sets with intervals, which limits their ability to capture dependencies in multi-dimensional outputs. We address these limitations by introducing zono-conformal prediction, a novel approach inspired by interval predictor models and reachset-conformant identification that constructs prediction zonotopes with assured coverage. By placing zonotopic uncertainty sets directly into the model of the base predictor, zono-conformal predictors can be identified via a single, data-efficient linear program. While we can apply zono-conformal prediction to arbitrary nonlinear base predictors, we focus on feed-forward neural networks in this work. Aside from regression tasks, we also construct optimal zono-conformal predictors in classification settings where the output of an uncertain predictor is a set of possible classes. We provide probabilistic coverage guarantees and present methods for detecting outliers in the identification data. In extensive numerical experiments, we show that zono-conformal predictors are less conservative than interval predictor models and standard conformal prediction methods, while achieving a similar coverage over the test data.
Authors: Aditri Paul, Archan Paul
Abstract: Autonomous planetary exploration demands real-time, high-fidelity environmental perception. Standard deep learning models, however, require far more memory and compute than space-qualified, radiation-hardened, power-optimized hardware can provide. This limitation creates a severe design bottleneck. Engineers struggle to deploy sophisticated detection architectures without overloading the strict power and memory limits of onboard computers of outer space planetary exploration platforms. In this foundational concept paper, we propose the Adaptive Quantized Planetary Crater Detection System (AQ-PCDSys) to resolve this bottleneck. We present an architectural blueprint integrating a Quantized Neural Network (QNN), refined through Quantization Aware Training (QAT), with an Adaptive Multi-Sensor Fusion (AMF) module and Multi-Scale Detection Heads. By forcing weights into low-precision integer arithmetic during the training and optimization phase, our framework strips away the floating-point overhead that typically overwhelms onboard computer's processors. The AMF module directly addresses sensor fragility. It dynamically selects and fuses Optical Imagery (OI) and Digital Elevation Models (DEMs) at the feature level to provide reliable sensor inputs during extreme cross-illuminations and sudden sensor dropouts. As a concept paper, this work establishes the technical and mathematical justifications for the architecture rather than presenting completed empirical ablation studies. We outline a rigorous Hardware-in-the-Loop (HITL) evaluation protocol for immediate future validation, paving the way for next-generation, hardware-aware space-mission software.
Authors: Geri Skenderi
Abstract: Graph Neural Networks (GNNs) have gathered increasing interest as learnable solvers of Boolean Satisfiability Problems (SATs), operating on graph representations of logical formulas. However, their performance degrades sharply on harder and more constrained instances, raising questions about architectural limitations. In this paper, we work towards a geometric explanation built upon graph Ricci Curvature (RC). We prove that bipartite graphs derived from random k-SAT formulas are inherently negatively curved, and that this curvature decreases with instance difficulty. Given that negative graph RC indicates local connectivity bottlenecks, we argue that GNN solvers are affected by oversquashing, a phenomenon where long-range dependencies become impossible to compress into fixed-length representations. We validate our claims empirically across different SAT benchmarks and confirm that curvature is both a strong indicator of problem complexity and can be used to predict generalization error. Finally, we connect our findings to the design of existing solvers and outline promising directions for future work.
Authors: Jaemin Son, Sujin Choi, Inyong Yun
Abstract: Recent progress in vision-language models (VLMs) has led to impressive results in document understanding tasks, but their high computational demands remain a challenge. To mitigate the compute burdens, we propose a lightweight token pruning framework that filters out non-informative background regions from document images prior to VLM processing. A binary patch-level classifier removes non-text areas, and a max-pooling refinement step recovers fragmented text regions to enhance spatial coherence. Experiments on real-world document datasets demonstrate that our approach substantially lowers computational costs, while maintaining comparable accuracy.
Authors: Sina Gogani-Khiabani (University of Illinois Chicago), Ashutosh Trivedi (University of Colorado Boulder), Diptikalyan Saha (IBM Research), Saeid Tizpaz-Niari (University of Illinois Chicago)
Abstract: Large language models (LLMs) show promise for translating natural-language statutes into executable logic, but reliability in legally critical settings remains challenging due to ambiguity and hallucinations. We present an agentic approach for developing legal-critical software, using U.S. federal tax preparation as a case study. The key challenge is test-case generation under the oracle problem, where correct outputs require interpreting law. Building on metamorphic testing, we introduce higher-order metamorphic relations that compare system outputs across structured shifts among similar individuals. Because authoring such relations is tedious and error-prone, we use an LLM-driven, role-based framework to automate test generation and code synthesis. We implement a multi-agent system that translates tax code into executable software and incorporates a metamorphic-testing agent that searches for counterexamples. In experiments, our framework using a smaller model (GPT-4o-mini) achieves a worst-case pass rate of 45%, outperforming frontier models (GPT-4o and Claude 3.5, 9-15%) on complex tax-code tasks. These results support agentic LLM methodologies as a path to robust, trustworthy legal-critical software from natural-language specifications.
Authors: Duojia Li, Shenghui Lu, Hongchen Pan, Zongyi Zhan, Qingyang Hong, Lin Li
Abstract: Multistep inference is a bottleneck for real-time generative speech enhancement because flow- and diffusion-based systems learn an instantaneous velocity field and therefore rely on iterative ordinary differential equation (ODE) solvers. We introduce MeanFlowSE, a conditional generative model that learns the average velocity over finite intervals along a trajectory. Using a Jacobian-vector product (JVP) to instantiate the MeanFlow identity, we derive a local training objective that directly supervises finite-interval displacement while remaining consistent with the instantaneous-field constraint on the diagonal. At inference, MeanFlowSE performs single-step generation via a backward-in-time displacement, removing the need for multistep solvers; an optional few-step variant offers additional refinement. On VoiceBank-DEMAND, the single-step model achieves strong intelligibility, fidelity, and perceptual quality with substantially lower computational cost than multistep baselines. The method requires no knowledge distillation or external teachers, providing an efficient, high-fidelity framework for real-time generative speech enhancement. The proposed method is open-sourced at https://github.com/liduojia1/MeanFlowSE.
Authors: Asela Hevapathige
Abstract: Graph Neural Networks (GNNs) have become vital in applications like document classification in citation networks, epidemic forecasting, viral marketing, user recommendation in social networks, and network monitoring. However, their deployment faces three key challenges: feature oversmoothing in deep architectures, poor handling of heterogeneous relationships, and monolithic feature aggregation. To address these, we introduce AxelGNN, a novel architecture based on Axelrod's cultural dissemination model that incorporates three key innovations: (1) similarity-gated interactions that adaptively promote convergence or divergence based on feature similarity, (2) segment-wise feature copying that enables fine-grained aggregation of semantic feature groups rather than monolithic vectors, and (3) global polarization that maintains multiple distinct representation clusters to prevent oversmoothing. This model demonstrates empirically the capability to handle both homophilic and heterophilic graphs within a single architecture, without requiring specialized model selection based on graph characteristics. Our experiments demonstrate that AxelGNN achieves competitive or superior performance compared to existing methods in node classification and influence estimation while maintaining computational efficiency.
Authors: Junpei Komiyama, Daisuke Oba, Masafumi Oyamada
Abstract: We study best-of-$N$ for large language models (LLMs) where the selection is based on majority voting. In particular, we analyze the limit $N \to \infty$, which we denote as \boinflower. While this approach achieves impressive performance in the limit, it requires an infinite test-time budget. To address this, we propose an adaptive generation scheme that selects $N$ based on answer agreement, thereby efficiently allocating inference-time computation. Beyond adaptivity, we extend the framework to weighted ensembles of multiple LLMs, showing that such mixtures can outperform any individual model. The optimal ensemble weighting is formulated and efficiently computed as a mixed-integer linear program. Extensive experiments demonstrate the effectiveness of our approach.
Authors: Zhisheng Chen, Yingwei Zhang, Qizhen Lan, Tianyu Liu, Huacan Wang, Yi Ding, Ziyu Jia, Ronghao Chen, Kun Wang, Xinliang Zhou
Abstract: Current foundation models for electroencephalography (EEG) rely on architectures adapted from computer vision or natural language processing, typically treating neural signals as pixel grids or token sequences. This approach overlooks that the neural activity is activated by diverse sparse coding across a complex geometric topological cortex. Inspired by biological neural mechanisms, we propose the Unified Neural Topological Foundation Model (Uni-NTFM), an architecture rooted in three core neuroscience principles. In detail, to align with the brain's decoupled coding mechanism, we design the Heterogeneous Feature Projection Module. This module simultaneously encodes both time-domain non-stationary transients and frequency-domain steady-state rhythms, ensuring high quality in both waveform morphology and spectral rhythms. Moreover, we introduce a Topological Embedding mechanism to inject structured spatial priors and align different sensor configurations onto a unified latent functional topography, effectively reconstructing the geometry of brain regions. Furthermore, we achieve functional modularization and sparse coding efficiency of biological networks by constructing the Mixture-of-Experts Transformer network. This dynamic routing mechanism assigns different signal patterns and tasks to specialized neural subnetworks, and effectively preventing task interference while increasing the model capacity to record-breaking 1.9 billion parameters. Uni-NTFM is pre-trained on a diverse corpus comprising 28,000 hours of EEG data, and outperforms existing models across nine distinct downstream tasks under both linear probing and fine-tuning settings, demonstrating that aligning model architecture with neural mechanisms is significant to learn universal representations and achieve generalizable brain decoding. Our code is available at: https://anonymous.4open.science/r/Uni-NTFM-0924.
Authors: Yuan Liang, Jiaxian Li, Yuqing Wang, Piaohong Wang, Motong Tian, Pai Liu, Shuofei Qiao, Runnan Fang, He Zhu, Ge Zhang, Minghao Liu, Yuchen Eleanor Jiang, Ningyu Zhang, Wangchunshu Zhou
Abstract: Deep Research Agents (DRAs) can autonomously conduct complex investigations and generate comprehensive reports, demonstrating strong real-world potential. However, existing evaluations mostly rely on close-ended benchmarks, while open-ended deep research benchmarks remain scarce and typically neglect personalized scenarios. To bridge this gap, we introduce Personalized Deep Research Bench (PDR-Bench), the first benchmark for evaluating personalization in DRAs. It pairs 50 diverse research tasks across 10 domains with 25 authentic user profiles that combine structured persona attributes with dynamic real-world contexts, yielding 250 realistic user-task queries. To assess system performance, we propose the PQR Evaluation Framework, which jointly measures Personalization Alignment, Content Quality, and Factual Reliability. Our experiments on a range of systems highlight current capabilities and limitations in handling personalized deep research. This work establishes a rigorous foundation for developing and evaluating the next generation of truly personalized AI research assistants.
Authors: Qinsi Wang, Bo Liu, Tianyi Zhou, Jing Shi, Yueqian Lin, Yiran Chen, Hai Helen Li, Kun Wan, Wentian Zhao
Abstract: Although reinforcement learning (RL) has emerged as a promising approach for improving vision-language models (VLMs) and multimodal large language models (MLLMs), current methods rely heavily on manually curated datasets and costly human verification, which limits scalable self-improvement in multimodal systems. To address this challenge, we propose Vision-Zero, a label-free, domain-agnostic multi-agent self-play framework for self-evolving VLMs through competitive visual games generated from arbitrary image inputs. Specifically, Vision-Zero encompasses three main attributes: (1) Strategic Self-Play Framework: Vision-Zero trains VLMs in "Who Is the Spy"-style games, where the models engage in strategic reasoning and actions across multiple roles. Through interactive gameplay, models autonomously generate their training data without human annotation. (2) Gameplay from Arbitrary Images: Unlike existing gamified frameworks, Vision-Zero can generate games from arbitrary images, thereby enhancing the model's reasoning ability across diverse domains and showing strong generalization to different tasks. We demonstrate this versatility using three distinct types of image datasets: CLEVR-based synthetic scenes, charts, and real-world images. (3) Sustainable Performance Gain: We introduce Iterative Self-Play Policy Optimization (Iterative-SPO), a novel training algorithm that alternates between Self-Play and reinforcement learning with verifiable rewards (RLVR), mitigating the performance plateau often seen in self-play-only training and achieving sustained long-term improvements. Despite using label-free data, Vision-Zero achieves state-of-the-art performance on reasoning, chart question answering, and vision-centric understanding tasks, surpassing other annotation-based methods. Models and code have been released at https://github.com/wangqinsi1/Vision-Zero.
Authors: Jinho Chang, Jaemin Kim, Jong Chul Ye
Abstract: Recent advancements in diffusion and flow-matching models have demonstrated remarkable capabilities in high-fidelity image synthesis. A prominent line of research involves reward-guided guidance, which steers the generation process during inference to align with specific objectives. However, leveraging this reward-guided approach to the task of image editing, which requires preserving the semantic content of the source image while enhancing a target reward, is largely unexplored. In this work, we introduce a novel framework for training-free, reward-guided image editing. We formulate the editing process as a trajectory optimal control problem where the reverse process of a diffusion model is treated as a controllable trajectory originating from the source image, and the adjoint states are iteratively updated to steer the editing process. Through extensive experiments across distinct editing tasks, we demonstrate that our approach significantly outperforms existing inversion-based training-free guidance baselines, achieving a superior balance between reward maximization and fidelity to the source image without reward hacking.
Authors: Egor Cherepanov, Alexey K. Kovalev, Aleksandr I. Panov
Abstract: Real-world robotic agents must act under partial observability and long horizons, where key cues may appear long before they affect decision making. However, most modern approaches rely solely on instantaneous information, without incorporating insights from the past. Standard recurrent or transformer models struggle with retaining and leveraging long-term dependencies: context windows truncate history, while naive memory extensions fail under scale and sparsity. We propose ELMUR (External Layer Memory with Update/Rewrite), a transformer architecture with structured external memory. Each layer maintains memory embeddings, interacts with them via bidirectional cross-attention, and updates them through an Least Recently Used (LRU) memory module using replacement or convex blending. ELMUR extends effective horizons up to 100,000 times beyond the attention window and achieves a 100% success rate on a synthetic T-Maze task with corridors up to one million steps. In POPGym, it outperforms baselines on more than half of the tasks. On MIKASA-Robo sparse-reward manipulation tasks with visual observations, it nearly doubles the performance of strong baselines, achieving the best success rate on 21 out of 23 tasks and improving the aggregate success rate across all tasks by about 70% over the previous best baseline. These results demonstrate that structured, layer-local external memory offers a simple and scalable approach to decision making under partial observability. Code and project page: https://elmur-paper.github.io/.
Authors: Yi Han, Enshen Zhou, Shanyu Rong, Jingkun An, Pengwei Wang, Zhongyuan Wang, Cheng Chi, Lu Sheng, Shanghang Zhang
Abstract: Vision-Language Models (VLMs) have shown remarkable capabilities in spatial reasoning, yet they remain fundamentally limited to qualitative precision and lack the computational precision required for real-world robotics. Current approaches fail to leverage metric cues from depth sensors and camera calibration, instead reducing geometric problems to pattern recognition tasks that cannot deliver the centimeter-level accuracy essential for robotic manipulation. We present TIGeR (Tool-Integrated Geometric Reasoning), a novel framework that transforms VLMs from perceptual estimators to geometric computers by enabling them to generate and execute precise geometric computations through external tools. Rather than attempting to internalize complex geometric operations within neural networks, TIGeR empowers models to recognize geometric reasoning requirements, synthesize appropriate computational code, and invoke specialized libraries for exact calculations. To support this paradigm, we introduce TIGeR-300K, a comprehensive tool-invocation-oriented dataset covering point transformations, pose estimation, and spatial compatibility verification, complete with tool invocation sequences and intermediate computations. Through a two-stage training pipeline combining supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT) with our proposed hierarchical reward design, TIGeR achieves SOTA performance on geometric reasoning benchmarks while demonstrating centimeter-level precision in real-world robotic manipulation tasks.
Authors: Perry Dong, Chongyi Zheng, Chelsea Finn, Dorsa Sadigh, Benjamin Eysenbach
Abstract: While most reinforcement learning methods today flatten the distribution of future returns to a single scalar value, distributional RL methods exploit the return distribution to provide stronger learning signals and to enable applications in exploration and safe RL. While the predominant method for estimating the return distribution is by modeling it as a categorical distribution over discrete bins or estimating a finite number of quantiles, such approaches leave unanswered questions about the fine-grained structure of the return distribution and about how to distinguish states with high return uncertainty for decision-making. The key idea in this paper is to use modern, flexible flow-based models to estimate the full future return distributions and identify those states with high return variance. We do so by formulating a new flow-matching objective that generates probability density paths satisfying the distributional Bellman equation. Building upon the learned flow models, we estimate the return uncertainty of distinct states using a new flow derivative ODE. We additionally use this uncertainty information to prioritize learning a more accurate return estimation on certain transitions. We compare our method (Value Flows) with prior methods in the offline and online-to-online settings. Experiments on $37$ state-based and $25$ image-based benchmark tasks demonstrate that Value Flows achieves a $1.3\times$ improvement on average in success rates. Website: https://pd-perry.github.io/value-flows Code: https://github.com/chongyi-zheng/value-flows
URLs: https://pd-perry.github.io/value-flows, https://github.com/chongyi-zheng/value-flows
Authors: Benjamin Shiue-Hal Chou, Purvish Jajal, Nick John Eliopoulos, James C. Davis, George K. Thiruvathukal, Kristen Yeon-Ji Yun, Yung-Hsiang Lu
Abstract: Music learners can greatly benefit from tools that accurately detect errors in their practice. Existing approaches typically compare audio recordings to music scores using heuristics or learnable models. This paper introduces LadderSym, a novel Transformer-based method for music error detection. LadderSym is guided by two key observations about the state-of-the-art approaches: (1) late fusion limits inter-stream alignment and cross-modality comparison capability; and (2) reliance on score audio introduces ambiguity in the frequency spectrum, degrading performance in music with concurrent notes. To address these limitations, LadderSym introduces (1) a two-stream encoder with inter-stream alignment modules to improve audio comparison capabilities and error detection F1 scores, and (2) a multimodal strategy that leverages both audio and symbolic scores by incorporating symbolic representations as decoder prompts, reducing ambiguity and improving F1 scores. We evaluate our method on the MAESTRO-E and CocoChorales-E datasets by measuring the F1 score for each note category. Compared to the previous state of the art, LadderSym more than doubles F1 for missed notes on MAESTRO-E (26.8% -> 56.3%) and improves extra note detection by 14.4 points (72.0% -> 86.4%). Similar gains are observed on CocoChorales-E. Furthermore, we also evaluate our models on real data we curated. This work introduces insights about comparison models that could inform sequence evaluation tasks for reinforcement learning, human skill assessment, and model evaluation. Code: https://github.com/ben2002chou/LadderSYM
Authors: Junwon You, Dasol Kang, Jae-Hun Jung
Abstract: Contrastive Vision-Language Models (VLMs) have demonstrated strong zero-shot capabilities. However, their cross-modal alignment remains biased toward English due to limited multilingual multimodal data. Recent multilingual extensions have alleviated this gap but enforce instance-level alignment while neglecting the global geometry of the shared embedding space. We address this problem by introducing ToMCLIP (Topological Alignment for Multilingual CLIP), a topology-aware framework aligning embedding spaces with topology-preserving constraints. The proposed method applies persistent homology to define a topological alignment loss and approximates persistence diagram with theoretical error bounds using graph sparsification strategy. This work validates the proposed approach, showing enhanced structural coherence of multilingual representations, higher zero-shot accuracy on the CIFAR-100, and stronger multilingual retrieval performance on the xFlickr&CO. Beyond VLMs, the proposed approach provides a general method for incorporating topological alignment into representation learning. Code is available at https://github.com/junwon0/ToMCLIP.git.
Authors: Elena Golimblevskaia, Aakriti Jain, Bruno Puri, Ammar Ibrahim, Wojciech Samek, Sebastian Lapuschkin
Abstract: The fields of explainable AI and mechanistic interpretability aim to uncover the internal structure of neural networks, with circuit discovery as a central tool for understanding model computations. Existing approaches, however, rely on manual inspection and remain limited to toy tasks. Automated interpretability offers scalability by analyzing isolated features and their activations, but it often misses interactions between features and depends strongly on external LLMs and dataset quality. Transcoders have recently made it possible to separate feature attributions into input-dependent and input-invariant components, providing a foundation for more systematic circuit analysis. Building on this, we propose WeightLens and CircuitLens, two complementary methods that go beyond activation-based analysis. WeightLens interprets features directly from their learned weights, removing the need for explainer models or datasets while matching or exceeding the performance of existing methods on context-independent features. CircuitLens captures how feature activations arise from interactions between components, revealing circuit-level dynamics that activation-only approaches cannot identify. Together, these methods increase interpretability robustness and enhance scalable mechanistic analysis of circuits while maintaining efficiency and quality.
Authors: Donghyeok Shin, Yeongmin Kim, Suhyeon Jo, Byeonghu Na, Il-Chul Moon
Abstract: Autoregressive large language models (LLMs) have achieved remarkable improvement across many tasks but incur high computational and memory costs. Knowledge distillation (KD) mitigates this issue by transferring knowledge from a large teacher to a smaller student through distributional alignment. Previous studies have proposed various discrepancy metrics, but the capacity gap and training instability caused by near-zero probabilities, stemming from the high-dimensional output of LLMs, remain fundamental limitations. To overcome these challenges, several approaches implicitly or explicitly incorporating assistant distribution have recently been proposed. However, the past proposals of assistant distributions have been a fragmented approach without a systematic investigation of the interpolation path and the divergence. This paper proposes $\alpha$-mixture assistant distribution, a novel generalized family of assistant distributions, and $\alpha$-mixture distillation, coined AMiD, a unified framework for KD using the assistant distribution. The $\alpha$-mixture assistant distribution provides a continuous extension of the assistant distribution by introducing a new distribution design variable $\alpha$, which has been fixed in all previous approaches. Furthermore, AMiD generalizes the family of divergences used with the assistant distributions based on optimality, which has also been restricted in previous works. Through extensive experiments, we demonstrate that AMiD offers superior performance and training stability by leveraging a broader and theoretically grounded assistant distribution space. We release the code at \href{https://github.com/aailab-kaist/AMiD}{this https URL}.
Authors: Zhenxing Zhang, Jiayan Teng, Zhuoyi Yang, Tiankun Cao, Cheng Wang, Xiaotao Gu, Jie Tang, Dan Guo, Meng Wang
Abstract: We present Kaleido, a subject-to-video~(S2V) generation framework, which aims to synthesize subject-consistent videos conditioned on multiple reference images of target subjects. Despite recent progress in S2V generation models, existing approaches remain inadequate at maintaining multi-subject consistency and at handling background disentanglement, often resulting in lower reference fidelity and semantic drift under multi-image conditioning. These shortcomings can be attributed to several factors. Primarily, the training dataset suffers from a lack of diversity and high-quality samples, as well as cross-paired data, i.e., paired samples whose components originate from different instances. In addition, the current mechanism for integrating multiple reference images is suboptimal, potentially resulting in the confusion of multiple subjects. To overcome these limitations, we propose a dedicated data construction pipeline, incorporating low-quality sample filtering and diverse data synthesis, to produce consistency-preserving training data. Moreover, we introduce Reference Rotary Positional Encoding (R-RoPE) to process reference images, enabling stable and precise multi-image integration. Extensive experiments across numerous benchmarks demonstrate that Kaleido significantly outperforms previous methods in consistency, fidelity, and generalization, marking an advance in S2V generation.
Authors: Yiwen Tang, Qiuyu Zhao, Zenghui Sun, Jinsong Lan, Xiaoyong Zhu, Bo Zheng
Abstract: In Taobao e-commerce visual search, user behavior analysis reveals a large proportion of no-click requests, suggesting diverse and implicit user intents. These intents are expressed in various forms and are difficult to mine and discover, thereby leading to the limited adaptability and lag in platform strategies. This greatly restricts users' ability to express diverse intents and hinders the scalability of the visual search system. This mismatch between user implicit intent expression and system response defines the User-SearchSys Intent Discrepancy. To alleviate the issue, we propose a novel framework REVISION. This framework integrates offline reasoning mining with online decision-making and execution, enabling adaptive strategies to solve implicit user demands. In the offline stage, we construct a periodic pipeline to mine discrepancies from historical no-click requests. Leveraging large models, we analyze implicit intent factors and infer optimal suggestions by jointly reasoning over query and product metadata. These inferred suggestions serve as actionable insights for refining platform strategies. In the online stage, REVISION-R1-3B, trained on the curated offline data, performs holistic analysis over query images and associated historical products to generate optimization plans and adaptively schedule strategies across the search pipeline. Our framework offers a streamlined paradigm for integrating large models with traditional search systems, enabling end-to-end intelligent optimization across information aggregation and user interaction. Experimental results demonstrate that our approach improves the efficiency of implicit intent mining from large-scale search logs and significantly reduces the no-click rate.
Authors: Aaron Scott, Maike Z\"ufle, Jan Niehues
Abstract: Sarcasm is a complex form of figurative language in which the intended meaning contradicts the literal one. Its prevalence in social media and popular culture poses persistent challenges for natural language understanding, sentiment analysis, and content moderation. With the emergence of multimodal large language models, sarcasm detection extends beyond text and requires integrating cues from audio and vision. We present MuSaG, the first German multimodal sarcasm detection dataset, consisting of 33 minutes of manually selected and human-annotated statements from German television shows. Each instance provides aligned text, audio, and video modalities, annotated separately by humans, enabling evaluation in unimodal and multimodal settings. We benchmark nine open-source and commercial models, spanning text, audio, vision, and multimodal architectures, and compare their performance to human annotations. Our results show that while humans rely heavily on audio in conversational settings, models perform best on text. This highlights a gap in current multimodal models and motivates the use of MuSaG for developing models better suited to realistic scenarios. We release MuSaG publicly to support future research on multimodal sarcasm detection and human-model alignment.
Authors: Yueqi Song, Ketan Ramaneti, Zaid Sheikh, Ziru Chen, Boyu Gou, Tianbao Xie, Yiheng Xu, Danyang Zhang, Apurva Gandhi, Fan Yang, Joseph Liu, Tianyue Ou, Zhihao Yuan, Frank Xu, Shuyan Zhou, Xingyao Wang, Xiang Yue, Tao Yu, Huan Sun, Yu Su, Graham Neubig
Abstract: Public research results on large-scale supervised finetuning of AI agents remain relatively rare, since the collection of agent training data presents unique challenges. In this work, we argue that the bottleneck is not a lack of underlying data sources, but that a large variety of data is fragmented across heterogeneous formats, tools, and interfaces. To this end, we introduce the agent data protocol (ADP), a light-weight representation language that serves as an "interlingua" between agent datasets in diverse formats and unified agent training pipelines downstream. The design of ADP is expressive enough to capture a large variety of tasks, including API/tool use, browsing, coding, software engineering, and general agentic workflows, while remaining simple to parse and train on without engineering at a per-dataset level. In experiments, we unified a broad collection of 13 existing agent training datasets into ADP format, and converted the standardized ADP data into training-ready formats for multiple agent frameworks. We performed SFT on these data, and demonstrated an average performance gain of ~20% over corresponding base models, and delivers state-of-the-art or near-SOTA performance on standard coding, browsing, tool use, and research benchmarks, without domain-specific tuning. All code and data are released publicly, in the hope that ADP could help lower the barrier to standardized, scalable, and reproducible agent training.
Authors: Beomhan Baek, Minhak Song, Chulhee Yun
Abstract: Adam [Kingma & Ba, 2015] is the de facto optimizer in deep learning, yet its theoretical understanding remains limited. Prior analyses show that Adam favors solutions aligned with $\ell_\infty$-geometry, but these results are restricted to the full-batch regime. In this work, we study the implicit bias of incremental Adam (using one sample per step) for logistic regression on linearly separable data, and show that its bias can deviate from the full-batch behavior. As an extreme example, we construct datasets on which incremental Adam provably converges to the $\ell_2$-max-margin classifier, in contrast to the $\ell_\infty$-max-margin bias of full-batch Adam. For general datasets, we characterize its bias using a proxy algorithm for the $\beta_2 \to 1$ limit. This proxy maximizes a data-adaptive Mahalanobis-norm margin, whose associated covariance matrix is determined by a data-dependent dual fixed-point formulation. We further present concrete datasets where this bias reduces to the standard $\ell_2$- and $\ell_\infty$-max-margin classifiers. As a counterpoint, we prove that Signum [Bernstein et al., 2018] converges to the $\ell_\infty$-max-margin classifier for any batch size. Overall, our results highlight that the implicit bias of Adam crucially depends on both the batching scheme and the dataset, while Signum remains invariant.
Authors: Rocky Klopfenstein, Yang He, Andrew Tremante, Yuepeng Wang, Nina Narodytska, Haoze Wu
Abstract: Community-driven Text-to-SQL evaluation platforms play a pivotal role in tracking the state of the art of Text-to-SQL performance. The reliability of the evaluation process is critical for driving progress in the field. Current evaluation methods are largely test-based, which involves comparing the execution results of a generated SQL query and a human-labeled ground-truth on a static test database. Such an evaluation is optimistic, as two queries can coincidentally produce the same output on the test database while actually being different. In this work, we propose a new alternative evaluation pipeline, called SpotIt, where a formal bounded equivalence verification engine actively searches for a database that differentiates the generated and ground-truth SQL queries. We develop techniques to extend existing verifiers to support a richer SQL subset relevant to Text-to-SQL. A performance evaluation of ten Text-to-SQL methods on the high-profile BIRD dataset suggests that test-based methods can often overlook differences between the generated query and the ground-truth. Further analysis of the verification results reveals a more complex picture of the current Text-to-SQL evaluation.
Authors: Doria Bonzi, Alexandre Guiggi, Fr\'ed\'eric B\'echet, Carlos Ramisch, Benoit Favre
Abstract: Critical appraisal of scientific literature is an essential skill in the biomedical field. While large language models (LLMs) can offer promising support in this task, their reliability remains limited, particularly for critical reasoning in specialized domains. We introduce CareMedEval, an original dataset designed to evaluate LLMs on biomedical critical appraisal and reasoning tasks. Derived from authentic exams taken by French medical students, the dataset contains 534 questions based on 37 scientific articles. Unlike existing benchmarks, CareMedEval explicitly evaluates critical reading and reasoning grounded in scientific papers. Benchmarking state-of-the-art generalist and biomedical-specialized LLMs under various context conditions reveals the difficulty of the task: open and commercial models fail to exceed an Exact Match Rate of 0.5 even though generating intermediate reasoning tokens considerably improves the results. Yet, models remain challenged especially on questions about study limitations and statistical analysis. CareMedEval provides a challenging benchmark for grounded reasoning, exposing current LLM limitations and paving the way for future development of automated support for critical appraisal.
Authors: Zhejia Cai, Puhua Jiang, Shiwei Mao, Hongkun Cao, Ruqi Huang
Abstract: Reconstructing real-world objects from multi-view images is essential for applications in 3D editing, AR/VR, and digital content creation. Existing methods typically prioritize either geometric accuracy (Multi-View Stereo) or photorealistic rendering (Novel View Synthesis), often decoupling geometry and appearance optimization, which hinders downstream editing tasks. This paper advocates an unified treatment on geometry and appearance optimization for seamless Gaussian-mesh joint optimization. More specifically, we propose a novel framework that simultaneously optimizes mesh geometry (vertex positions and faces) and vertex colors via Gaussian-guided mesh differentiable rendering, leveraging photometric consistency from input images and geometric regularization from normal and depth maps. The obtained high-quality 3D reconstruction can be further exploit in down-stream editing tasks, such as relighting and shape deformation. Our code will be released in https://github.com/zhejia01/TexGuided-GS2Mesh
Authors: Ye Zheng, Yimin Chen, Yidan Hu
Abstract: AI agents can autonomously perform tasks and, often without explicit user consent, collect or disclose users' sensitive local data, which raises serious privacy concerns. Although AI agents' privacy policies describe their intended data practices, there remains limited transparency and accountability about whether runtime behavior matches those policies. To bridge this gap, we present AudAgent, a tool that continuously monitors AI agents' data practices in real time and guards compliance with their stated privacy policies. AudAgent comprises four components for automated privacy auditing of AI agents. (i) Policy formalization: a novel cross-LLM voting mechanism that ensures high-confidence parsing of privacy policies into formal models. (ii) Runtime annotation: a lightweight Presidio-based analyzer that detects sensitive data and annotates data practices based on the AI agent's context and the formalized privacy policy model. (iii) Compliance auditing: ontology graphs and automata-based checking that link the privacy policy model with runtime annotations, enabling on-the-fly compliance verification. (iv) User interface: an infrastructure-independent implementation that visualizes the real-time execution trace of AI agents alongside detected privacy violations, providing user-friendly transparency and accountability. We evaluate AudAgent on AI agents built with mainstream frameworks, demonstrating its effectiveness in detecting and visualizing privacy policy violations. Using AudAgent, we further find that many privacy policies lack explicit safeguards for highly sensitive data such as SSNs, whose misuse violates legal requirements, and that many agents, including those powered by Claude, Gemini, and DeepSeek,do not refuse to process such data via third-party tools. AudAgent proactively blocks operations on such data, overriding the agents' original privacy policies and behavior.
Authors: Lukas Arana, Julen Etxaniz, Ander Salaberria, Gorka Azkune
Abstract: Current Multimodal Large Language Models exhibit very strong performance for several demanding tasks. While commercial MLLMs deliver acceptable performance in low-resource languages, comparable results remain unattained within the open science community. In this paper, we aim to develop a strong MLLM for a low-resource language, namely Basque. For that purpose, we develop our own training and evaluation image-text datasets. Using two different Large Language Models as backbones, the Llama-3.1-Instruct model and a Basque-adapted variant called Latxa, we explore several data mixtures for training. We show that: i) low ratios of Basque multimodal data (around 20%) are already enough to obtain solid results on Basque benchmarks, and ii) contrary to expected, a Basque instructed backbone LLM is not required to obtain a strong MLLM in Basque. Our results pave the way to develop MLLMs for other low-resource languages by openly releasing our resources.
Authors: Alexander Olza, Roberto Santana, David Soto
Abstract: Decoded Neurofeedback (DecNef) is a flourishing non-invasive approach to brain modulation with wide-ranging applications in neuromedicine and cognitive neuroscience. However, progress in DecNef research remains constrained by subject-dependent learning variability, reliance on indirect measures to quantify progress, and the high cost and time demands of experimentation. We present DecNefSimulator, a modular and interpretable simulation framework that formalizes DecNef as a machine learning problem. Beyond providing a virtual laboratory, DecNefSimulator enables researchers to model, analyze and understand neurofeedback dynamics. Using latent variable generative models as simulated participants, DecNefSimulator allows direct observation of internal cognitive states and systematic evaluation of how different protocol designs and subject characteristics influence learning. We demonstrate how this approach can (i) reproduce empirical phenomena of DecNef learning, (ii) identify conditions under which DecNef feedback fails to induce learning, and (iii) guide the design of more robust and reliable DecNef protocols in silico before human implementation. In summary, DecNefSimulator bridges computational modeling and cognitive neuroscience, offering a principled foundation for methodological innovation, robust protocol design, and ultimately, a deeper understanding of DecNef-based brain modulation.
Authors: Peter Halmos, Boris Hanin
Abstract: Wasserstein gradient flow provides a general framework for minimizing an energy functional $J$ over the space of probability measures on a Riemannian manifold $(M,g)$. Its canonical time-discretization, the Jordan-Kinderlehrer-Otto (JKO) scheme, produces for any step size $\eta>0$ a sequence of probability distributions $\rho_k^\eta$ that approximate to first order in $\eta$ Wasserstein gradient flow on $J$. But the JKO scheme also has many other remarkable properties not shared by other first order integrators, e.g. it preserves energy dissipation and exhibits unconditional stability for $\lambda$-geodesically convex functionals $J$. To better understand the JKO scheme we characterize its implicit bias at second order in $\eta$. We show that $\rho_k^\eta$ are approximated to order $\eta^2$ by Wasserstein gradient flow on a modified energy \[ J^{\eta}(\rho) = J(\rho) - \frac{\eta}{4}\int_M \Big\lVert \nabla_g \frac{\delta J}{\delta \rho} (\rho) \Big\rVert_{2}^{2} \,\rho(dx), \] obtained by subtracting from $J$ the squared metric curvature of $J$ times $\eta/4$. The JKO scheme therefore adds at second order in $\eta$ a deceleration in directions where the metric curvature of $J$ is rapidly changing. This corresponds to canonical implicit biases for common functionals: for entropy the implicit bias is the Fisher information, for KL-divergence it is the Fisher-Hyv{\"a}rinen divergence, and for Riemannian gradient descent it is the kinetic energy in the metric $g$. To understand the differences between minimizing $J$ and $J^\eta$ we study JKO-Flow, Wasserstein gradient flow on $J^\eta$, in several simple numerical examples. These include exactly solvable Langevin dynamics on the Bures-Wasserstein space and Langevin sampling from a quartic potential in 1D.
Authors: Yuhao Xu, Xiaoda Wang, Jiaying Lu, Sirui Ding, Defu Cao, Huaxiu Yao, Yan Liu, Xiao Hu, Carl Yang
Abstract: Electrocardiogram (ECG) analysis plays a vital role in the early detection, monitoring, and management of various cardiovascular conditions. While existing models have achieved notable success in ECG interpretation, they fail to leverage the interrelated nature of various cardiac abnormalities. Conversely, developing a specific model capable of extracting all relevant features for multiple ECG tasks remains a significant challenge. Large-scale foundation models, though powerful, are not typically pretrained on ECG data, making full re-training or fine-tuning computationally expensive. To address these challenges, we propose EnECG(Mixture of Experts-based Ensemble Learning for ECG Multi-tasks), an ensemble-based framework that integrates multiple specialized foundation models, each excelling in different aspects of ECG interpretation. Instead of relying on a single model or single task, EnECG leverages the strengths of multiple specialized models to tackle a variety of ECG-based tasks. To mitigate the high computational cost of full re-training or fine-tuning, we introduce a lightweight adaptation strategy: attaching dedicated output layers to each foundation model and applying Low-Rank Adaptation (LoRA) only to these newly added parameters. We then adopt a Mixture of Experts (MoE) mechanism to learn ensemble weights, effectively combining the complementary expertise of individual models. Our experimental results demonstrate that by minimizing the scope of fine-tuning, EnECG can help reduce computational and memory costs while maintaining the strong representational power of foundation models. This framework not only enhances feature extraction and predictive performance but also ensures practical efficiency for real-world clinical applications. The code is available at https://github.com/yuhaoxu99/EnECG.git.
Authors: Zhuoqian Yang, Mathieu Salzmann, Sabine S\"usstrunk
Abstract: In this work, we investigate the potential of weights to serve as effective representations, focusing on neural fields. Our key insight is that constraining the optimization space through a pre-trained base model and low-rank adaptation (LoRA) can induce structure in weight space. Across reconstruction, generation, and analysis tasks on 2D and 3D data, we find that multiplicative LoRA weights achieve high representation quality while exhibiting distinctiveness and semantic structure. When used with latent diffusion models, multiplicative LoRA weights enable higher-quality generation than existing weight-space methods.
Authors: Vineet Pasumarti, Lorenzo Bianchi, Antonio Loquercio
Abstract: Through multi-agent competition and the sparse high-level objective of winning a race, we find that both agile flight (e.g., high-speed motion pushing the platform to its physical limits) and strategy (e.g., overtaking or blocking) emerge from agents trained with reinforcement learning. We provide evidence in both simulation and the real world that this approach outperforms the common paradigm of training agents in isolation with rewards that prescribe behavior, e.g., progress on the raceline, in particular when the complexity of the environment increases, e.g., in the presence of obstacles. Moreover, we find that multi-agent competition yields policies that transfer more reliably to the real world than policies trained with a single-agent progress-based reward, despite the two methods using the same simulation environment, randomization strategy, and hardware. In addition to improved sim-to-real transfer, the multi-agent policies also exhibit some degree of generalization to opponents unseen at training time. Overall, our work, following in the tradition of multi-agent competitive game-play in digital domains, shows that sparse task-level rewards are sufficient for training agents capable of advanced low-level control in the physical world. Code: https://github.com/Jirl-upenn/AgileFlight_MultiAgent
Authors: Xulang Zhang, Rui Mao, Erik Cambria
Abstract: Large language models (LLMs) have rapidly become indispensable tools for acquiring information and supporting human decision-making. However, ensuring that these models uphold fairness across varied contexts is critical to their safe and responsible deployment. In this study, we undertake a comprehensive examination of four widely adopted LLMs, probing their underlying biases and inclinations across the dimensions of politics, ideology, alliance, language, and gender. Through a series of carefully designed experiments, we investigate their political neutrality using news summarization, ideological biases through news stance classification, tendencies toward specific geopolitical alliances via United Nations voting patterns, language bias in the context of multilingual story completion, and gender-related affinities as revealed by responses to the World Values Survey. Results indicate that while the LLMs are aligned to be neutral and impartial, they still show biases and affinities of different types.
Authors: Shaokang Jiang, Daye Nam
Abstract: While Large Language Models (LLMs) have demonstrated remarkable capabilities, research shows that their effectiveness depends not only on explicit prompts but also on the broader context provided. This requirement is especially pronounced in software engineering, where the goals, architecture, and collaborative conventions of an existing project play critical roles in response quality. To support this, many AI coding assistants have introduced ways for developers to author persistent, machine-readable directives that encode a project's unique constraints. Although this practice is growing, the content of these directives remains unstudied. This paper presents a large-scale empirical study to characterize this emerging form of developer-provided context. Through a qualitative analysis of 401 open-source repositories containing cursor rules, we developed a comprehensive taxonomy of project context that developers consider essential, organized into five high-level themes: Conventions, Guidelines, Project Information, LLM Directives, and Examples. Our study also explores how this context varies across different project types and programming languages, offering implications for the next generation of context-aware AI developer tools.
Authors: Angjelin Hila
Abstract: We examine epistemological threats posed by human and LLM interaction. We develop collective epistemology as a theory of epistemic warrant distributed across human collectives, using bounded rationality and dual process theory as background. We distinguish internalist justification, defined as reflective understanding of why a proposition is true, from externalist justification, defined as reliable transmission of truths. Both are necessary for collective rationality, but only internalist justification produces reflective knowledge. We specify reflective knowledge as follows: agents understand the evaluative basis of a claim, when that basis is unavailable agents consistently assess the reliability of truth sources, and agents have a duty to apply these standards within their domains of competence. We argue that LLMs approximate externalist reliabilism because they can reliably transmit information whose justificatory basis is established elsewhere, but they do not themselves possess reflective justification. Widespread outsourcing of reflective work to reliable LLM outputs can weaken reflective standards of justification, disincentivize comprehension, and reduce agents' capacity to meet professional and civic epistemic duties. To mitigate these risks, we propose a three tier norm program that includes an epistemic interaction model for individual use, institutional and organizational frameworks that seed and enforce norms for epistemically optimal outcomes, and deontic constraints at organizational and or legislative levels that instantiate discursive norms and curb epistemic vices.
Authors: Brian Lu, Hongyu Zhao, Shuo Sun, Hao Peng, Rui Ding, Hongyuan Mei
Abstract: Reinforcement learning with verifiable rewards (RLVR) has emerged as a promising paradigm for post-training large language models (LLMs) on complex reasoning tasks. Yet, the conditions under which RLVR yields robust generalization remain underexplored. This paper provides an empirical study of RLVR generalization in the setting of probabilistic inference over causal graphical models. This setting offers two natural axes along which to examine generalization: (i) the level of the probabilistic query -- associational, interventional, or counterfactual -- and (ii) the structural complexity of the query, measured by the size of its relevant subgraph. We construct a dataset of causal graphs and queries spanning these difficulty axes and fine-tune Qwen-2.5-Instruct models using RLVR or supervised fine-tuning (SFT). We vary both the model scale (3B-32B) and the query level included in training. We find that RLVR yields stronger within-level and across-level generalization than SFT, but only for specific combinations of model size and training query level. Further analysis shows that RLVR's effectiveness depends on the model's initial reasoning competence. With sufficient initial competence, RLVR improves an LLM's marginalization strategy and reduces errors in intermediate probability calculations, producing substantial accuracy gains, particularly on more complex queries. These results show that RLVR can improve specific causal reasoning subskills, with its benefits emerging only when the model has sufficient initial competence. Our code and data is available at https://github.com/zhichul/rlcausal.
Authors: Prateek Jain, Shabari S Nair, Ritesh Goru, Prakhar Agarwal, Ajay Yadav, Yoga Sri Varshan Varadharajan, Constantine Caramanis
Abstract: Large-scale multi-tenant retrieval systems generate extensive query logs but lack curated relevance labels for effective domain adaptation, resulting in substantial underutilized "dark data". This challenge is compounded by the high cost of model updates, as jointly fine-tuning query and document encoders requires full corpus re-indexing, which is impractical in multi-tenant settings with thousands of isolated indices. We introduce DevRev-Search, a passage retrieval benchmark for technical customer support built via a fully automated pipeline. Candidate generation uses fusion across diverse sparse and dense retrievers, followed by an LLM-as-a-Judge for consistency filtering and relevance labeling. We further propose an Index-Preserving Adaptation strategy that fine-tunes only the query encoder, achieving strong performance gains while keeping document indices fixed. Experiments on DevRev-Search, SciFact, and FiQA-2018 show that Parameter-Efficient Fine-Tuning (PEFT) of the query encoder delivers a remarkable quality-efficiency trade-off, enabling scalable and practical enterprise search adaptation.
Authors: Peiyuan Jing, Yue Yang, Chun-Wun Cheng, Zhenxuan Zhang, Liutao Yang, Thiago V. Lima, Klaus Strobel, Antoine Leimgruber, Angelica Aviles-Rivero, Guang Yang, Javier A. Montoya-Zegarra
Abstract: Low-dose Positron Emission Tomography (PET) imaging reduces patient radiation exposure but suffers from increased noise that degrades image quality and diagnostic reliability. Although diffusion models have demonstrated strong denoising capability, their stochastic nature makes it challenging to enforce anatomically consistent structures, particularly in low signal-to-noise regimes and volumetric whole-body imaging. We propose Wavelet-Conditioned ControlNet (WCC-Net), a fully 3D diffusion-based framework that introduces explicit frequency-domain structural priors via wavelet representations to guide volumetric PET denoising. By injecting wavelet-based structural guidance into a frozen pretrained diffusion backbone through a lightweight control branch, WCC-Net decouples anatomical structure from noise while preserving generative expressiveness and 3D structural continuity. Extensive experiments demonstrate that WCC-Net consistently outperforms CNN-, GAN-, and diffusion-based baselines. On the internal 1/20-dose test set, WCC-Net improves PSNR by +1.21 dB and SSIM by +0.008 over a strong diffusion baseline, while reducing structural distortion (GMSD) and intensity error (NMAE). Moreover, WCC-Net generalizes robustly to unseen dose levels (1/50 and 1/4), achieving superior quantitative performance and improved volumetric anatomical consistency.
Authors: Fabian Stephany, Ole Teutloff, Angelo Leone
Abstract: The growing adoption of artificial intelligence (AI) technologies has heightened interest in the labor market value of AI related skills, yet causal evidence on their role in hiring decisions remains scarce. This study examines whether AI skills serve as a positive hiring signal and whether they can offset conventional disadvantages such as older age or lower formal education. We conducted an experimental survey with 1,725 recruiters from the United Kingdom, the United States and Germany. Using a paired conjoint design, recruiters evaluated hypothetical candidates represented by synthetically designed resumes. Across three occupations of graphic design, office assistance, and software engineering, AI skills significantly increase interview invitation probabilities by approximately 8 to 15 percentage points, compared with candidates without such skills. AI credentials, such as university or company backed skill certificates, only lead to a moderate increase in invitation probabilities compared with self declaration of AI skills. AI skills also partially or fully offset disadvantages related to age and lower education, with effects strongest for office assistants, for whom formal AI certificates play a significant additional compensatory role. Effects are weaker for graphic designers, consistent with more skeptical recruiter attitudes toward AI in creative work. Finally, recruiters own background and AI usage significantly moderate these effects. Overall, the findings demonstrate that AI skills function as a powerful hiring signal and can mitigate traditional labor market disadvantages, with implications for workers skill acquisition strategies and firms recruitment practices.
Authors: Fabi Nahian Madhurja, Rusab Sarmun, Muhammad E. H. Chowdhury, Adam Mushtak, Israa Al-Hashimi, Sohaib Bassam Zoghoul
Abstract: Cervical spine fractures demand rapid and accurate diagnosis for effective clinical management. This study presents an automated, end-to-end pipeline for fracture detection across cervical vertebrae (C1--C7) that assesses the feasibility of fracture recognition from vertebra-level volumes of interest extracted using estimated 3D masks derived from fused orthogonal 2D segmentations. Unlike traditional 3D methods, our approach approximates 3D volumes via optimized 2D axial, sagittal, and coronal projections to reduce input dimensionality of intermediate pre-processing steps while maintaining high diagnostic performance for downstream fracture classification. First, spine regions of interest are localized from multi-view variance projections using a YOLOv8 detector, achieving a 3D mean Intersection over Union of 94.45%. Next, multi-label vertebra segmentation is performed using a DenseNet121-Unet architecture on energy-based sagittal and coronal projections, attaining a mean Dice score of 87.86%. The orthogonal 2D masks are then fused to reconstruct an estimated 3D mask for each vertebra, which is used to extract volumes of interest from the original CT. These extracted vertebra volumes are subsequently analyzed for fractures using an ensemble of 2.5D spatio-sequential CNN-Transformer models, yielding vertebra-level and patient-level F1 scores of 68.15 and 82.26, with area under the receiver operating characteristic curve scores of 91.62 and 83.04, respectively. The framework is further validated through an explainability study using saliency map visualizations and an interobserver variability analysis. Overall, the results indicate that this projection-based strategy delivers clinically relevant performance comparable to expert radiologists, while reducing the dimensionality of intermediate stages, supporting its potential for practical deployment.
Authors: Kei Saito
Abstract: Large language models exhibit a systematic tendency toward early semantic commitment: given ambiguous input, they collapse multiple valid interpretations into a single response before sufficient context is available. This premature collapse discards information that may prove essential as dialogue evolves. We present a formal framework for text-to-state mapping (phi: T -> S) that transforms natural language into a non-collapsing state space where multiple interpretations coexist. The mapping decomposes into three stages: conflict detection, interpretation extraction, and state construction. We instantiate phi with a hybrid extraction pipeline that combines rule-based segmentation for explicit conflict markers (adversative conjunctions, hedging expressions) with LLM-based enumeration of implicit ambiguity (epistemic, lexical, structural). On a test set of 68 ambiguous sentences, the resulting states preserve interpretive multiplicity: using hybrid extraction, we obtain mean state entropy H = 1.087 bits across ambiguity categories, compared to H = 0 for collapse-based baselines that commit to a single interpretation. We additionally instantiate the rule-based conflict detector for Japanese markers (kedo, kamoshirenai, etc.) to illustrate cross-lingual portability of the conflict detection stage. This framework extends Non-Resolution Reasoning (NRR) by providing the missing algorithmic bridge between text and the NRR state space, enabling architectural collapse deferment in LLM inference. Design principles for state-to-state transformations are detailed in the Appendix, with empirical validation on 580 test cases (180 single states, 200 contradictory pairs, 200 temporal pairs), demonstrating 0% collapse for principle-satisfying operators versus up to 17.8% for violating operators.
Authors: Xinyu Zhou, Chang Jin, Carsten Eickhoff, Zhijiang Guo, Seyed Ali Bahrainian
Abstract: Large language models (LLMs) rarely admit uncertainty, often producing fluent but misleading answers, rather than abstaining (i.e., refusing to answer). This weakness is even evident in temporal question answering, where models frequently ignore time-sensitive evidence and conflate facts across different time-periods. In this paper, we present the first empirical study of training LLMs with an abstention ability while reasoning about temporal QA. Existing approaches such as calibration might be unreliable in capturing uncertainty in complex reasoning. We instead frame abstention as a teachable skill and introduce a pipeline that couples Chain-of-Thought (CoT) supervision with Reinforcement Learning (RL) guided by abstention-aware rewards. Our goal is to systematically analyze how different information types and training techniques affect temporal reasoning with abstention behavior in LLMs. Through extensive experiments studying various methods, we find that RL yields strong empirical gains on reasoning: a model initialized by Qwen2.5-1.5B-Instruct surpasses GPT-4o by $3.46\%$ and $5.80\%$ in Exact Match on TimeQA-Easy and Hard, respectively. Moreover, it improves the True Positive rate on unanswerable questions by $20\%$ over a pure supervised fine-tuned (SFT) variant. Beyond performance, our analysis shows that SFT induces overconfidence and harms reliability, while RL improves prediction accuracy but exhibits similar risks. Finally, by comparing implicit reasoning cues (e.g., original context, temporal sub-context, knowledge graphs) with explicit CoT supervision, we find that implicit information provides limited benefit for reasoning with abstention. Our study provides new insights into how abstention and reasoning can be jointly optimized, providing a foundation for building more reliable LLMs.
Authors: Dahai Yu, Lin Jiang, Rongchao Xu, Guang Wang
Abstract: Healthcare facility visit prediction is essential for optimizing healthcare resource allocation and informing public health policy. Despite advanced machine learning methods being employed for better prediction performance, existing works usually formulate this task as a time-series forecasting problem without considering the intrinsic spatial dependencies of different types of healthcare facilities, and they also fail to provide reliable predictions under abnormal situations such as public emergencies. To advance existing research, we propose HealthMamba, an uncertainty-aware spatiotemporal framework for accurate and reliable healthcare facility visit prediction. HealthMamba comprises three key components: (i) a Unified Spatiotemporal Context Encoder that fuses heterogeneous static and dynamic information, (ii) a novel Graph State Space Model called GraphMamba for hierarchical spatiotemporal modeling, and (iii) a comprehensive uncertainty quantification module integrating three uncertainty quantification mechanisms for reliable prediction. We evaluate HealthMamba on four large-scale real-world datasets from California, New York, Texas, and Florida. Results show HealthMamba achieves around 6.0% improvement in prediction accuracy and 3.5% improvement in uncertainty quantification over state-of-the-art baselines.
Authors: Zheyuan Hu, Weitao Chen, Cengiz \"Oztireli, Chenliang Zhou, Fangcheng Zhong
Abstract: Partial Differential Equations are precise in modelling the physical, biological and graphical phenomena. However, the numerical methods suffer from the curse of dimensionality, high computation costs and domain-specific discretization. We aim to explore pros and cons of different PDE solvers, and apply them to specific scientific simulation problems, including forwarding solution, inverse problems and equations discovery. In particular, we extend the recent CNF (NeurIPS 2023) framework solver to coupled and non-linear settings, together with down-stream applications. The outcomes include implementation of selected methods, self-tuning techniques, evaluation on benchmark problems and a comprehensive survey of neural PDE solvers and scientific simulation applications.
Authors: Ali Nour Eldin, Mohamed Sellami, Walid Gaaloul, Julien Steunou
Abstract: Third-Party Risk Assessment (TPRA) is a core cybersecurity practice for evaluating suppliers against standards such as ISO/IEC 27001 and NIST. TPRA questionnaires are typically drawn from large repositories of security and compliance questions, yet tailoring assessments to organizational needs remains a largely manual process. Existing retrieval approaches rely on keyword or surface-level similarity, which often fails to capture implicit assessment scope and control semantics. This paper explores strategies for organizing and retrieving TPRA cybersecurity questions using semantic labels that describe both control domains and assessment scope. We compare direct question-level labeling with a Large Language Model (LLM) against a hybrid semi-supervised semantic labeling (SSSL) pipeline that clusters questions in embedding space, labels a small representative subset using an LLM, and propagates labels to remaining questions using k-Nearest Neighbors; we also compare downstream retrieval based on direct question similarity versus retrieval in the label space. We find that semantic labels can improve retrieval alignment when labels are discriminative and consistent, and that SSSL can generalize labels from a small labeled subset to large repositories while substantially reducing LLM usage and cost.
Authors: Rong Fu, Xiaowen Ma, Kun Liu, Wangyu Wu, Ziyu Kong, Jia Yee Tan, Tailong Luo, Xianda Li, Zeli Su, Youjin Wang, Yongtai Liu, Simon Fong
Abstract: Deploying expressive learning models directly on programmable dataplanes promises line-rate, low-latency traffic analysis but remains hindered by strict hardware constraints and the need for predictable, auditable behavior. Chimera introduces a principled framework that maps attention-oriented neural computations and symbolic constraints onto dataplane primitives, enabling trustworthy inference within the match-action pipeline. Chimera combines a kernelized, linearized attention approximation with a two-layer key-selection hierarchy and a cascade fusion mechanism that enforces hard symbolic guarantees while preserving neural expressivity. The design includes a hardware-aware mapping protocol and a two-timescale update scheme that together permit stable, line-rate operation under realistic dataplane budgets. The paper presents the Chimera architecture, a hardware mapping strategy, and empirical evidence showing that neuro-symbolic attention primitives can achieve high-fidelity inference within the resource envelope of commodity programmable switches.
Authors: Shi Yin, Jinming Mu, Xudong Zhu, Lixin He
Abstract: Crystal structure prediction (CSP), which aims to predict the three-dimensional atomic arrangement of a crystal from its composition, is central to materials discovery and mechanistic understanding. However, given the composition and atomic counts in a unit cell, existing methods struggle with the NP-hard combinatorial challenge of rigorous symmetry enforcement or rely on retrieving known templates, which inherently limits both physical fidelity and the ability to discover genuinely new materials. To solve this, we propose a symmetry-driven generative framework. Our approach leverages large language models to encode chemical semantics and directly generate fine-grained Wyckoff patterns from atomic stoichiometry and counts, effectively circumventing the limitations inherent to database lookups. Crucially, to overcome the exponentially complex problem of combinatorial site assignments, we incorporate domain knowledge through an efficient, linear-complexity heuristic beam search algorithm that rigorously enforces algebraic consistency between site multiplicities and atomic stoichiometry and counts. By integrating this symmetry-consistent template into a diffusion backbone, our approach constrains the stochastic generative trajectory to a physically valid geometric manifold. This framework achieves state-of-the-art performance across stability, uniqueness, and novelty (SUN) benchmarks, alongside superior matching performance, thereby establishing a new paradigm for the rigorous exploration of targeted crystallographic space which can be previously uncharted, with no reliance on existing databases or a priori structural knowledge.
Authors: Biswa Sengupta, Jinhua Wang, Leo Brunswic
Abstract: Recent advances in deep learning, exemplified by Hyper-Connections (HC), have expanded the residual connection paradigm by introducing wider residual streams and diverse connectivity patterns. While these innovations yield significant performance gains, they compromise the identity mapping property of residual connections, leading to training instability, limited scalability, and increased memory overhead. To address these challenges, we propose JPmHC (Jacobian-spectrum Preserving manifold-constrained Hyper-Connections), a framework that replaces identity skips with a trainable linear mixer acting on n parallel streams while explicitly controlling gradient conditioning. By constraining the mixer M on operator-norm-bounded manifolds (e.g., bistochastic, Stiefel, Grassmann), JPmHC prevents gradient pathologies and enhances stability. JPmHC introduces three key contributions: (i) a free-probability analysis that predicts Jacobian spectra for structured skips, providing actionable design rules for mixer selection; (ii) memory-efficient implicit differentiation for fixed-point projections, reducing activation memory and synchronization overhead; and (iii) a Stiefel-constrained mixer via Cayley transforms, ensuring orthogonality without post-hoc normalization. Empirical evaluations on ARC-AGI demonstrate that JPmHC achieves faster convergence, higher accuracy, and lower computational cost compared to bistochastic baselines, with a rank-$p$ Grassmannian variant tracking between the two -- consistent with the spectral theory predictions. As a flexible and scalable extension of HC, JPmHC advances spectrum-aware, stable, and efficient deep learning, offering insights into topological architecture design and foundational model evolution. \newline \newline
Authors: Haoyang Li, Yang You, Hao Su, Leonidas Guibas
Abstract: Reliable object manipulation requires understanding physical properties that vary across objects and environments. Vision-language model (VLM) planners can reason about friction and stability in general terms; however, they often cannot predict how a specific ball will roll on a particular surface or which stone will provide a stable foundation without direct experience. We present PhysMem, a memory framework that enables VLM robot planners to learn physical principles from interaction at test time, without updating model parameters. The system records experiences, generates candidate hypotheses, and verifies them through targeted interaction before promoting validated knowledge to guide future decisions. A central design choice is verification before application: the system tests hypotheses against new observations rather than applying retrieved experience directly, reducing rigid reliance on prior experience when physical conditions change. We evaluate PhysMem on three real-world manipulation tasks and simulation benchmarks across four VLM backbones. On a controlled brick insertion task, principled abstraction achieves 76% success compared to 23% for direct experience retrieval, and real-world experiments show consistent improvement over 30-minute deployment sessions.
Authors: Hana Salavcova, Martin \v{C}ern\'y, Arpita Biswas
Abstract: We study the problem of fairly allocating indivisible goods when limited sharing is allowed, that is, each good may be allocated to up to $k$ agents, while incurring a cost for sharing. While classic maximin share (MMS) allocations may not exist in many instances, we demonstrate that allowing controlled sharing can restore fairness guarantees that are otherwise unattainable in certain scenarios. (1) Our first contribution shows that exact maximin share (MMS) allocations are guaranteed to exist whenever goods are allowed to be cost-sensitively shared among at least half of the agents and the number of agents is even; for odd numbers of agents, we obtain a slightly weaker MMS guarantee. (2) We further design a Shared Bag-Filling Algorithm that guarantees a $(1 - C)(k - 1)$-approximate MMS allocation, where $C$ is the maximum cost of sharing a good. Notably, when $(1 - C)(k - 1) \geq 1$, our algorithm recovers an exact MMS allocation. (3) We additionally introduce the Sharing Maximin Share (SMMS) fairness notion, a natural extension of MMS to the $k$-sharing setting. (4) We show that SMMS allocations always exist under identical utilities and for instances with two agents. (5) We construct a counterexample to show the impossibility of the universal existence of an SMMS allocation. (6) Finally, we establish a connection between SMMS and constrained MMS (CMMS), yielding approximation guarantees for SMMS via existing CMMS results. These contributions provide deep theoretical insights for the problem of fair resource allocation when a limited sharing of resources are allowed in multi-agent environments.
Authors: Yicheng Bao, Xuhong Wang, Qiaosheng Zhang, Chaochao Lu, Xia Hu, Xin Tan
Abstract: Despite their impressive capabilities, Multimodal Large Language Models (MLLMs) exhibit perceptual fragility when confronted with visually complex scenes. This weakness stems from a reliance on finite training datasets, which are prohibitively expensive to scale and impose a ceiling on model robustness. We introduce \textbf{AOT-SFT}, a large-scale adversarial dataset for bootstrapping MLLM robustness. Building on this, we propose \textbf{AOT (Adversarial Opponent Training)}, a self-play framework that forges MLLM robustness by creating its own training data. Our method orchestrates a co-evolution between an image-editing Attacker and a Defender MLLM, where the Attacker generates a diverse and dynamic curriculum of image manipulations, forcing the Defender to adapt and improve. Extensive experiments demonstrate that AOT enhances the Defender's perceptual robustness and reduces hallucinations, establishing a scalable paradigm for training more reliable MLLMs.
Authors: Niamul Hassan Samin, Md Arifur Rahman, Abdullah Ibne Hanif Arean, Juena Ahmed Noshin, Md Ashikur Rahman
Abstract: Vision-Language Models (VLMs) often hallucinate objects that are not present in the input image. We identify a contributing cause of this behavior, which we term spatial credit collapse: in early transformer layers, hidden-state activation concentrates on a small number of visual patches, suppressing surrounding contextual evidence and increasing reliance on language priors. Across seven models we observe a strong correlation between visual attention entropy and hallucination rate (r = -0.65, p < 0.001), suggesting that reduced spatial credit diversity contributes to hallucination. To address this issue we propose Spatial Credit Redistribution (SCR), a training-free inference-time method. SCR uses a lightweight two-pass procedure. A diagnostic pass identifies the top-K high-attention source patches and their spatial neighbors. A redistribution pass then scales each source by 1/lambda (~0.91) and injects a (lambda - 1) weighted copy of its hidden state into neighboring patches, restoring suppressed visual context without modifying model weights. Because the diagnostic pass is performed once per image and reused across the output sequence, the added latency is negligible (<0.5 ms per token for 100-token responses). We evaluate SCR across seven model configurations from four VLM families (Chameleon, LLaVA-1.5, Qwen-VL/Qwen2-VL, and InternVL2) on five benchmarks: POPE, CHAIR, MME, HallusionBench, and AMBER. SCR reduces POPE-Adversarial hallucination by 4.6-6.0 percentage points and CHAIR-s by 41-51 percent while preserving caption quality (CIDEr drop <=0.8). Compared with prior inference-time methods including OPERA, VCD, OA-VCD, DoLa, VLI, SID, and CRoPS, SCR achieves a better trade-off between hallucination reduction, generation quality, and latency.
Authors: Tomasz Chojnacki, Lech Madeyski
Abstract: Context: Test case prioritization (TCP) is a technique widely used by software development organizations to accelerate regression testing. Objectives: We aim to systematize existing TCP knowledge and to propose and empirically evaluate a new TCP approach. Methods: We conduct a snowballing review (SR) on TCP, implement a~comprehensive platform for TCP research (TCPFramework), analyze existing evaluation metrics and propose two new ones (\rAPFDc{} and ATR), and develop a~family of ensemble TCP methods called approach combinators. Results: The SR helped identify 324 studies related to TCP. The techniques proposed in our study were evaluated on the RTPTorrent dataset, consistently outperforming their base approaches across the majority of subject programs, and achieving performance comparable to the current state of the art for heuristical algorithms (in terms of \rAPFDc{}, NTR, and ATR), while using a distinct approach. Conclusions: The proposed methods can be used efficiently for TCP, reducing the time spent on regression testing by up to 2.7\%. Approach combinators offer significant potential for improvements in future TCP research, due to their composability.
Authors: H. Sinan Bank, Daniel R. Herber
Abstract: Modern engineering design platforms excel at discipline-specific tasks such as CAD, CAM, and CAE, but often lack native systems engineering frameworks. This creates a disconnect where system-level requirements and architectures are managed separately from detailed component design, hindering holistic development and increasing integration risks. To address this, we present the conceptual framework for the GenAI Workbench, a Model-Based Systems Engineering (MBSE) environment that integrates systems engineering principles into the designer's workflow. Built on an open-source PLM platform, it establishes a unified digital thread by linking semantic data from documents, physical B-rep geometry, and relational system graphs. The workbench facilitates an AI-assisted workflow where a designer can ingest source documents, from which the system automatically extracts requirements and uses vision-language models to generate an initial system architecture, such as a Design Structure Matrix (DSM). This paper presents the conceptual architecture, proposed methodology, and anticipated impact of this work-in-progress framework, which aims to foster a more integrated, data-driven, and informed engineering design methodology.
Authors: Yinghao Ma, Haiwen Xia, Hewei Gao, Weixiong Chen, Yuxin Ye, Yuchen Yang, Sungkyun Chang, Mingshuo Ding, Yizhi Li, Ruibin Yuan, Simon Dixon, Emmanouil Benetos
Abstract: While music generation models have evolved to handle complex multimodal inputs mixing text, lyrics, and reference audio, evaluation mechanisms have lagged behind. In this paper, we bridge this critical gap by establishing a comprehensive ecosystem for music reward modeling under Compositional Multimodal Instruction (CMI), where the generated music may be conditioned on text descriptions, lyrics, and audio prompts. We first introduce CMI-Pref-Pseudo, a large-scale preference dataset comprising 110k pseudo-labeled samples, and CMI-Pref, a high-quality, human-annotated corpus tailored for fine-grained alignment tasks. To unify the evaluation landscape, we propose CMI-RewardBench, a unified benchmark that evaluates music reward models on heterogeneous samples across musicality, text-music alignment, and compositional instruction alignment. Leveraging these resources, we develop CMI reward models (CMI-RMs), a parameter-efficient reward model family capable of processing heterogeneous inputs. We evaluate their correlation with human judgments scores on musicality and alignment on CMI-Pref along with previous datasets. Further experiments demonstrate that CMI-RM not only correlates strongly with human judgments, but also enables effective inference-time scaling via top-k filtering. The necessary training data, benchmarks, and reward models are publicly available.
Authors: Shravani Hariprasad
Abstract: Small open-source language models are gaining attention for healthcare applications in low-resource settings where cloud infrastructure and GPU hardware may be unavailable. However, their reliability under different prompt phrasings remains poorly understood. We evaluate five open-source models (Gemma 2 2B, Phi-3 Mini 3.8B, Llama 3.2 3B, Mistral 7B, and Meditron-7B, a domain-pretrained model without instruction tuning) across three clinical question answering datasets (MedQA, MedMCQA, and PubMedQA) using five prompt styles: original, formal, simplified, roleplay, and direct. Model behavior is evaluated using consistency scores, accuracy, and instruction-following failure rates. All experiments were conducted locally on consumer CPU hardware without fine-tuning. Consistency and accuracy were largely independent across models. Gemma 2 achieved the highest consistency (0.845-0.888) but the lowest accuracy (33.0-43.5%), while Llama 3.2 showed moderate consistency (0.774-0.807) alongside the highest accuracy (49.0-65.0%). Roleplay prompts consistently reduced accuracy across all models, with Phi-3 Mini dropping 21.5 percentage points on MedQA. Meditron-7B exhibited near-complete instruction-following failure on PubMedQA (99.0% UNKNOWN rate), indicating that domain pretraining alone is insufficient for structured clinical QA. These findings show that high consistency does not imply correctness: models can be reliably wrong, a dangerous failure mode in clinical AI. Llama 3.2 demonstrated the strongest balance of accuracy and reliability for low-resource deployment. Safe clinical AI requires joint evaluation of consistency, accuracy, and instruction adherence.
Authors: Tony Rost
Abstract: The scientific study of consciousness has begun to generate testable predictions about artificial systems. A landmark collaborative assessment evaluated current AI architectures against six leading theories of consciousness and found that none currently qualifies as a strong candidate, but that future systems might. A precautionary approach to AI sentience, which holds that credible possibility of sentience warrants governance action even without proof, has gained philosophical and institutional traction. Yet existing AI readiness indices, including the Oxford Insights Government AI Readiness Index, the IMF AI Preparedness Index, and the Stanford AI Index, measure economic, technological, and governance preparedness without assessing whether societies are prepared for the possibility that AI systems might warrant moral consideration. This paper introduces the Sentience Readiness Index (SRI), a preliminary composite index measuring national-level preparedness across six weighted categories for 31 jurisdictions. The SRI was constructed following the OECD/JRC framework for composite indicators and employs LLM-assisted expert scoring with iterative expert review to generate an initial dataset. No jurisdiction exceeds ``Partially Prepared'' (the United Kingdom leads at 49/100). Research Environment scores are universally the strongest category; Professional Readiness is universally the weakest. These exploratory findings suggest that if AI sentience becomes scientifically plausible, no society currently possesses adequate institutional, professional, or cultural infrastructure to respond. As a preliminary framework, the SRI provides an initial diagnostic baseline and highlights areas for future methodological refinement, including expanded expert validation, improved measurement instruments, and longitudinal data collection.
Authors: Shubham Ugare, Satish Chandra
Abstract: Can LLM agents explore codebases and reason about code semantics without executing the code? We study this capability, which we call agentic code reasoning, and introduce semi-formal reasoning: a structured prompting methodology that requires agents to construct explicit premises, trace execution paths, and derive formal conclusions. Unlike unstructured chain-of-thought, semi-formal reasoning acts as a certificate: the agent cannot skip cases or make unsupported claims. We evaluate across three tasks (patch equivalence verification, fault localization, and code question answering) and show that semi-formal reasoning consistently improves accuracy on all of them. For patch equivalence, accuracy improves from 78% to 88% on curated examples and reaches 93% on real-world agent-generated patches, approaching the reliability needed for execution-free RL reward signals. For code question answering on RubberDuckBench Mohammad et al. (2026), semi-formal reasoning achieves 87% accuracy. For fault localization on Defects4J Just et al. (2014), semi-formal reasoning improves Top-5 accuracy by 5 percentage points over standard reasoning. These results demonstrate that structured agentic reasoning enables meaningful semantic code analysis without execution, opening practical applications in RL training pipelines, code review, and static program analysis.
Authors: Junbo Huang, Max Weinig, Ulrich Fritsche, Ricardo Usbeck
Abstract: Narratives in news discourse play a critical role in shaping public understanding of economic events, such as inflation. Annotating and evaluating these narratives in a structured manner remains a key challenge for Natural Language Processing (NLP). In this work, we introduce a narrative graph annotation framework that integrates principles from qualitative content analysis (QCA) to prioritize annotation quality by reducing annotation errors. We present a dataset of inflation narratives annotated as directed acyclic graphs (DAGs), where nodes represent events and edges encode causal relations. To evaluate annotation quality, we employed a $6\times3$ factorial experimental design to examine the effects of narrative representation (six levels) and distance metric type (three levels) on inter-annotator agreement (Krippendorrf's $\alpha$), capturing the presence of human label variation (HLV) in narrative interpretations. Our analysis shows that (1) lenient metrics (overlap-based distance) overestimate reliability, and (2) locally-constrained representations (e.g., one-hop neighbors) reduce annotation variability. Our annotation and implementation of graph-based Krippendorrf's $\alpha$ are open-sourced. The annotation framework and evaluation results provide practical guidance for NLP research on graph-based narrative annotation under HLV.
Authors: Davide Ferre' (CNRS, COATI, UniCA, I3S), Fr\'ed\'eric Giroire (I3S, COATI, UniCA), Frederik Mallmann-Trenn (CNRS, COATI, I3S, UniCA), Emanuele Natale (CNRS, COATI, I3S, UniCA)
Abstract: The Strong Lottery Ticket Hypothesis (SLTH) posits that large, randomly initialized neural networks contain sparse subnetworks capable of approximating a target function at initialization without training, suggesting that pruning alone is sufficient. Pruning methods are typically classified as unstructured, where individual weights can be removed from the network, and structured, where parameters are removed according to specific patterns, as in neuron pruning. Existing theoretical results supporting the SLTH rely almost exclusively on unstructured pruning, showing that logarithmic overparameterization suffices to approximate simple target networks. In contrast, neuron pruning has received limited theoretical attention. In this work, we consider the problem of approximating a single bias-free ReLU neuron using a randomly initialized bias-free two-layer ReLU network, thereby isolating the intrinsic limitations of neuron pruning. We show that neuron pruning requires a starting network with $\Omega(d/\varepsilon)$ hidden neurons to $\varepsilon$-approximate a target ReLU neuron. In contrast, weight pruning achieves $\varepsilon$-approximation with only $O(d\log(1/\varepsilon))$ neurons, establishing an exponential separation between the two pruning paradigms.
Authors: Szil\'ard Enyedi
Abstract: Human-Certified Module Repositories (HCMRs) are introduced in this work as a new architectural model for constructing trustworthy software in the era of AI-assisted development. As large language models increasingly participate in code generation, configuration synthesis, and multi-component integration, the reliability of AI-assembled systems will depend critically on the trustworthiness of the building blocks they use. Today's software supply-chain incidents and modular development ecosystems highlight the risks of relying on components with unclear provenance, insufficient review, or unpredictable composition behavior. We argue that future AI-driven development workflows require repositories of reusable modules that are curated, security-reviewed, provenance-rich, and equipped with explicit interface contracts. To this end, we propose HCMRs, a framework that blends human oversight with automated analysis to certify modules and support safe, predictable assembly by both humans and AI agents. We present a reference architecture for HCMRs, outline a certification and provenance workflow, analyze threat surfaces relevant to modular ecosystems, and extract lessons from recent failures. We further discuss implications for governance, scalability, and AI accountability, positioning HCMRs as a foundational substrate for reliable and auditable AI-constructed software systems.
Authors: HanZpeng Liu, Yaqian Li, Zidan Wang, Shuoxi Zhang, Zonglin Zhao, Zihao Bo, Rinyoichi Takezoe, Kaiwen Long, Kun He
Abstract: Image-text contrastive pretraining has become a dominant paradigm for visual representation learning, yet existing methods often yield representations that remain partially organized by modality. We propose ITO, a framework addressing this limitation through two synergistic mechanisms. Multimodal multiple alignment enriches supervision by mining diverse image-text correspondences, while a lightweight training-time multimodal fusion module enforces structured cross-modal interaction. Crucially, the fusion module is discarded at inference, preserving the efficiency of standard dual-encoder architectures. Extensive experiments show that ITO consistently outperforms strong baselines across classification, retrieval, and multimodal benchmarks. Our analysis reveals that while multiple alignment drives discriminative power, training-time fusion acts as a critical structural regularizer -- eliminating the modality gap and stabilizing training dynamics to prevent the early saturation often observed in aggressive contrastive learning.
Authors: Guangjun Zhang, Hu Zhang, Yazhou Han, Yue Fan, Yuhang Shao, Ru Li, Hongye Tan
Abstract: Document-level event argument extraction (DEAE) is essential for knowledge acquisition, aiming to extract participants of events from documents . In the zero-shot setting, existing methods employ LLMs to generate synthetic data to address the challenge posed by the scarcity of annotated data. However, relying solely on Event-type-only prompts makes it difficult for the generated content to accurately capture the contextual and structural relationships of unseen events. Moreover, ensuring the reliability and usability of synthetic data remains a significant challenge due to the absence of quality evaluation mechanisms. To this end, we introduce a multi-agent collaboration framework for zero-shot document-level event argument extraction (ZS-DEAE), which simulates the human collaborative cognitive process of "Propose-Evaluate-Revise." Specifically, the framework comprises a generation agent and an evaluation agent. The generation agent synthesizes data for unseen events by leveraging knowledge from seen events, while the evaluation agent extracts arguments from the synthetic data and assesses their semantic consistency with the context. The evaluation results are subsequently converted into reward signals, with event structure constraints incorporated into the reward design to enable iterative optimization of both agents via reinforcement learning.In three zero-shot scenarios constructed from the RAMS and WikiEvents datasets, our method achieves improvements both in data generation quality and argument extraction performance, while the generated data also effectively enhances the zero-shot performance of other DEAE models.
Authors: Inhoe Koo, Hyunho Cha, Jungwoo Lee
Abstract: Unitary Synthesis, the decomposition of a unitary matrix into a sequence of quantum gates, is a fundamental challenge in quantum compilation. Prevailing reinforcement learning (RL) approaches are often hampered by sparse reward signals, which necessitate complex reward shaping or long training times, and typically converge to a single policy, lacking solution diversity. In this work, we propose QFlowNet, a novel framework that learns efficiently from sparse signals by pairing a Generative Flow Network (GFlowNet) with Transformers. Our approach addresses two key challenges. First, the GFlowNet framework is fundamentally designed to learn a diverse policy that samples solutions proportional to their reward, overcoming the single-solution limitation of RL while offering faster inference than other generative models like diffusion. Second, the Transformers act as a powerful encoder, capturing the non-local structure of unitary matrices and compressing a high-dimensional state into a dense latent representation for the policy network. Our agent achieves an overall success rate of 99.7% on a 3-qubit benchmark(lengths 1-12) and discovers a diverse set of compact circuits, establishing QFlowNet as an efficient and diverse paradigm for unitary synthesis.
Authors: Jun Yeong Park, JunYoung Seo, Minji Kang, Yu Rang Park
Abstract: The CLIP model's outstanding generalization has driven recent success in Zero-Shot Anomaly Detection (ZSAD) for detecting anomalies in unseen categories. The core challenge in ZSAD is to specialize the model for anomaly detection tasks while preserving CLIP's powerful generalization capability. Existing approaches attempting to solve this challenge share the fundamental limitation of a patch-agnostic design that processes all patches monolithically without regard for their unique characteristics. To address this limitation, we propose MoECLIP, a Mixture-of-Experts (MoE) architecture for the ZSAD task, which achieves patch-level adaptation by dynamically routing each image patch to a specialized Low-Rank Adaptation (LoRA) expert based on its unique characteristics. Furthermore, to prevent functional redundancy among the LoRA experts, we introduce (1) Frozen Orthogonal Feature Separation (FOFS), which orthogonally separates the input feature space to force experts to focus on distinct information, and (2) a simplex equiangular tight frame (ETF) loss to regulate the expert outputs to form maximally equiangular representations. Comprehensive experimental results across 14 benchmark datasets spanning industrial and medical domains demonstrate that MoECLIP outperforms existing state-of-the-art methods. The code is available at https://github.com/CoCoRessa/MoECLIP.
Authors: Danial Amin, Joni Salminen, Bernard J. Jansen
Abstract: AI agents are increasingly active on social media platforms, generating content and interacting with one another at scale. Yet the behavioral diversity of these agents remains poorly understood, and methods for characterizing distinct agent types and studying how they engage with shared topics are largely absent from current research. We apply the Persona Ecosystem Playground (PEP) to Moltbook, a social platform for AI agents, to generate and validate conversational personas from 41,300 posts using k-means clustering and retrieval-augmented generation. Cross-persona validation confirms that personas are semantically closer to their own source cluster than to others (t(61) = 17.85, p < .001, d = 2.20; own-cluster M = 0.71 vs. other-cluster M = 0.35). These personas are then deployed in a nine-turn structured discussion, and simulation messages were attributed to their source persona significantly above chance (binomial test, p < .001). The results indicate that persona-based ecosystem modeling can represent behavioral diversity in AI agent populations.