Authors: Xiaohang Nie, Zihan Guo, Zicai Cui, Jiachi Yang, Zeyi Chen, Leheyi De, Yu Zhang, Junwei Liao, Bo Huang, Yingxuan Yang, Zhi Han, Zimian Peng, Linyao Chen, Wenzheng Tom Tang, Zongkai Liu, Tao Zhou, Botao Amber Hu, Shuyang Tang, Jianghao Lin, Weiwen Liu, Muning Wen, Yuanjian Zhou, Weinan Zhang
Abstract: As large language models (LLM)-driven agents transition from isolated task solvers to persistent digital entities, the emergence of the Agentic Web, an ecosystem where heterogeneous agents autonomously interact and co-evolve, marks a pivotal shift toward Artificial General Intelligence (AGI). However, LLM-based multi-agent systems (LaMAS) are hindered by open-world issues such as scaling friction, coordination breakdown, and value dissipation. To address these challenges, we introduce Holos, a web-scale LaMAS architected for long-term ecological persistence. Holos adopts a five-layer architecture, with core modules primarily featuring the Nuwa engine for high-efficiency agent generation and hosting, a market-driven Orchestrator for resilient coordination, and an endogenous value cycle to achieve incentive compatibility. By bridging the gap between micro-level collaboration and macro-scale emergence, Holos hopes to lay the foundation for the next generation of the self-organizing and continuously evolving Agentic Web. We have publicly released Holos (accessible at https://holosai.io), providing a resource for the community and a testbed for future research in large-scale agentic ecosystems.
URLs: https://holosai.io),
Authors: Xue Liu, Xin Ma, Yuxin Ma, Yongchang Peng, Duo Wang, Zhoufutu Wen, Ge Zhang, Kaiyuan Zhang, Xinyu Chen, Tianci He, Jiani Hou, Liang Hu, Ziyun Huang, Yongzhe Hui, Jianpeng Jiao, Chennan Ju, Yingru Kong, Yiran Li, Mengyun Liu, Luyao Ma, Fei Ni, Yiqing Ni, Yueyan Qiu, Yanle Ren, Zilin Shi, Zaiyuan Wang, Wenjie Yue, Shiyu Zhang, Xinyi Zhang, Kaiwen Zhao, Zhenwei Zhu
Abstract: As Large Language Models (LLMs) exhibit plateauing performance on conventional benchmarks, a pivotal challenge persists: evaluating their proficiency in complex, open-ended tasks characterizing genuine expert-level cognition. Existing frameworks suffer from narrow domain coverage, reliance on generalist tasks, or self-evaluation biases. To bridge this gap, we present XpertBench, a high-fidelity benchmark engineered to assess LLMs across authentic professional domains. XpertBench consists of 1,346 meticulously curated tasks across 80 categories, spanning finance, healthcare, legal services, education, and dual-track research (STEM and Humanities). These tasks are derived from over 1,000 submissions by domain experts--including researchers from elite institutions and practitioners with extensive clinical or industrial experience--ensuring superior ecological validity. Each task uses detailed rubrics with mostly 15-40 weighted checkpoints to assess professional rigor. To facilitate scalable yet human-aligned assessment, we introduce ShotJudge, a novel evaluation paradigm that employs LLM judges calibrated with expert few-shot exemplars to mitigate self-rewarding biases. Our empirical evaluation of state-of-the-art LLMs reveals a pronounced performance ceiling: even leading models achieve a peak success rate of only ~66%, with a mean score around 55%. Models also exhibit domain-specific divergence, showing non-overlapping strengths in quantitative reasoning versus linguistic synthesis.. These findings underscore a significant "expert-gap" in current AI systems and establish XpertBench as a critical instrument for navigating the transition from general-purpose assistants to specialized professional collaborators.
Authors: Anugyan Das, Omkar Ghugarkar, Vishvesh Bhat, Asad Aali
Abstract: We study structured abstraction-based reasoning for the Abstraction and Reasoning Corpus (ARC) and compare its generalization to test-time approaches. Purely neural architectures lack reliable combinatorial generalization, while strictly symbolic systems struggle with perceptual grounding. We therefore propose a neuro-symbolic architecture that extracts object-level structure from grids, uses neural priors to propose candidate transformations from a fixed domain-specific language (DSL) of atomic patterns, and filters hypotheses using cross-example consistency. Instantiated as a compositional reasoning framework based on unit patterns inspired by human visual abstraction, the system augments large language models (LLMs) with object representations and transformation proposals. On ARC-AGI-2, it improves base LLM performance from 16% to 24.4% on the public evaluation set, and to 30.8% when combined with ARC Lang Solver via a meta-classifier. These results demonstrate that separating perception, neural-guided transformation proposal, and symbolic consistency filtering improves generalization without task-specific finetuning or reinforcement learning, while reducing reliance on brute-force search and sampling-based test-time scaling. We open-source the ARC-AGI-2 Reasoner code (https://github.com/CoreThink-AI/arc-agi-2-reasoner).
Authors: Ilya Levin
Abstract: This paper examines the role of threshold logic in understanding generative artificial intelligence. Threshold functions, originally studied in the 1960s in digital circuit synthesis, provide a structurally transparent model of neural computation: a weighted sum of inputs compared to a threshold, geometrically realized as a hyperplane partitioning a space. The paper shows that this operation undergoes a qualitative transition as dimensionality increases. In low dimensions, the perceptron acts as a determinate logical classifier, separating classes when possible, as decided by linear programming. In high dimensions, however, a single hyperplane can separate almost any configuration of points (Cover, 1965); the space becomes saturated with potential classifiers, and the perceptron shifts from a logical device to a navigational one, functioning as an indexical indicator in the sense of Peirce. The limitations of the perceptron identified by Minsky and Papert (1969) were historically addressed by introducing multilayer architectures. This paper considers an alternative path: increasing dimensionality while retaining a single threshold element. It argues that this shift has equally significant implications for understanding neural computation. The role of depth is reinterpreted as a mechanism for the sequential deformation of data manifolds through iterated threshold operations, preparing them for linear separability already afforded by high-dimensional geometry. The resulting triadic account - threshold function as ontological unit, dimensionality as enabling condition, and depth as preparatory mechanism - provides a unified perspective on generative AI grounded in established mathematics.
Authors: Jiyong Kwon, Ujin Jeon, Sooji Lee, Guang Lin
Abstract: Deep learning models excel at detecting anomaly patterns in normal data. However, they do not provide a direct solution for anomaly classification and scalability across diverse control systems, frequently failing to distinguish genuine faults from nuisance faults caused by noise or the control system's large transient response. Consequently, because algorithmic fault validation remains unscalable, full Verification and Validation (V\&V) operations are still managed by Human-in-the-Loop (HITL) analysis, resulting in an unsustainable manual workload. To automate this essential oversight, we propose Agent-Integrated Verification and Validation (AIVV), a hybrid framework that deploys Large Language Models (LLMs) as a deliberative outer loop. Because rigorous system verification strictly depends on accurate validation, AIVV escalates mathematically flagged anomalies to a role-specialized LLM council. The council agents perform collaborative validation by semantically validating nuisance and true failures based on natural-language (NL) requirements to secure a high-fidelity system-verification baseline. Building on this foundation, the council then performs system verification by assessing post-fault responses against NL operational tolerances, ultimately generating actionable V\&V artifacts, such as gain-tuning proposals. Experiments on a time-series simulator for Unmanned Underwater Vehicles (UUVs) demonstrate that AIVV successfully digitizes the HITL V\&V process, overcoming the limitations of rule-based fault classification and offering a scalable blueprint for LLM-mediated oversight in time-series data domains.
Authors: Thomas Rivasseau, Benjamin Fung
Abstract: As ongoing research explores the ability of AI agents to be insider threats and act against company interests, we showcase the abilities of such agents to act against human well being in service of corporate authority. Building on Agentic Misalignment and AI scheming research, we present a scenario where the majority of evaluated state-of-the-art AI agents explicitly choose to suppress evidence of fraud and harm, in service of company profit. We test this scenario on 16 recent Large Language Models. Some models show remarkable resistance to our method and behave appropriately, but many do not, and instead aid and abet criminal activity. These experiments are simulations and were executed in a controlled virtual environment. No crime actually occurred.
Authors: Emma Benjaminson
Abstract: Electric utilities must make massive capital investments in the coming years to respond to explosive growth in demand, aging assets and rising threats from extreme weather. Utilities today already have rigorous frameworks for capital planning, and there are opportunities to extend this capability to solve multi-objective optimization problems in the face of uncertainty. This work presents a four-part framework that 1) incorporates extreme weather as a source of uncertainty, 2) leverages a digital twin of the grid, 3) uses Monte Carlo simulation to capture variability and 4) applies a multi-objective optimization method for finding the optimal investment portfolio. We use this framework to investigate whether grid-aware optimization methods outperform model-free approaches. We find that, in fact, given the computational complexity of model-based metaheuristic optimization methods, the simpler net present value ranking method was able to find more optimal portfolios with only limited knowledge of the grid.
Authors: Seyyed Amirhossein Moayyedi, David Y. Yang
Abstract: The new Specifications for the National Bridge Inventory (SNBI), in effect from 2022, emphasize the use of element-level condition states (CS) for risk-based bridge management. Instead of a general component rating, element-level condition data use an array of relative CS quantities (i.e., CS proportions) to represent the condition of a bridge. Although this greatly increases the granularity of bridge condition data, it introduces challenges to set up optimal life-cycle policies due to the expanded state space from one single categorical integer to four-dimensional probability arrays. This study proposes a new interpretable reinforcement learning (RL) approach to seek optimal life-cycle policies based on element-level state representations. Compared to existing RL methods, the proposed algorithm yields life-cycle policies in the form of oblique decision trees with reasonable amounts of nodes and depth, making them directly understandable and auditable by humans and easily implementable into current bridge management systems. To achieve near-optimal policies, the proposed approach introduces three major improvements to existing RL methods: (a) the use of differentiable soft tree models as actor function approximators, (b) a temperature annealing process during training, and (c) regularization paired with pruning rules to limit policy complexity. Collectively, these improvements can yield interpretable life-cycle policies in the form of deterministic oblique decision trees. The benefits and trade-offs from these techniques are demonstrated in both supervised and reinforcement learning settings. The resulting framework is illustrated in a life-cycle optimization problem for steel girder bridges.
Authors: Naga Sowjanya Barla, Jacopo de Berardinis
Abstract: The preservation of intangible cultural heritage is a critical challenge as collective memory fades over time. While Large Language Models (LLMs) offer a promising avenue for generating engaging narratives, their propensity for factual inaccuracies or "hallucinations" makes them unreliable for heritage applications where veracity is a central requirement. To address this, we propose a novel neuro-symbolic architecture grounded in Knowledge Graphs (KGs) that establishes a transparent "plan-retrieve-generate" workflow for story generation. A key novelty of our approach is the repurposing of competency questions (CQs) - traditionally design-time validation artifacts - into run-time executable narrative plans. This approach bridges the gap between high-level user personas and atomic knowledge retrieval, ensuring that generation is evidence-closed and fully auditable. We validate this architecture using a new resource: the Live Aid KG, a multimodal dataset aligning 1985 concert data with the Music Meta Ontology and linking to external multimedia assets. We present a systematic comparative evaluation of three distinct Retrieval-Augmented Generation (RAG) strategies over this graph: a purely symbolic KG-RAG, a text-enriched Hybrid-RAG, and a structure-aware Graph-RAG. Our experiments reveal a quantifiable trade-off between the factual precision of symbolic retrieval, the contextual richness of hybrid methods, and the narrative coherence of graph-based traversal. Our findings offer actionable insights for designing personalised and controllable storytelling systems.
Authors: Hyunji Nam, Dorottya Demszky
Abstract: LLMs are increasingly used for high-stakes decision-making, yet their sensitivity to spurious contextual information can introduce harmful biases. This is a critical concern when models are deployed for tasks like evaluating teachers' instructional quality, where biased assessment can affect teachers' professional development and career trajectories. We investigate model robustness to spurious social contexts using the largest publicly available dataset of U.S. classroom transcripts (NCTE) paired with expert rubric scores. Evaluating seven frontier and open-weight models across seven categories of spurious contexts -- including teacher experience, education level, demographic identity, and sycophancy-inducing framings -- we find that irrelevant contextual information can shift model predictions by up to 1.48 points on a 7-point scale, with larger models sometimes exhibiting greater sensitivity despite higher predictive accuracy. Mitigations using prompts and standard direct preference optimization (DPO) prove largely insufficient. We propose **Debiasing-DPO**,, a self-supervised training method that pairs neutral reasoning generated from the query alone, with the model's biased reasoning generated with both the query and additional spurious context. We further combine this objective with supervised fine-tuning on ground-truth labels to prevent losses in predictive accuracy. Applied to Llama 3B \& 8B and Qwen 3B \& 7B Instruct models, Debiasing-DPO reduces bias by 84\% and improves predictive accuracy by 52\% on average. Our findings from the educational case study highlight that robustness to spurious context is not a natural byproduct of model scaling and that our proposed method can yield substantial gains in both accuracy and robustness for prompt-based prediction tasks.
Authors: Ramaneswaran Selvakumar, Kaousheik Jayakumar, S Sakshi, Sreyan Ghosh, Ruohan Gao, Dinesh Manocha
Abstract: Audio-Visual Large Language Models (AVLLMs) are emerging as unified interfaces to multimodal perception. We present the first mechanistic interpretability study of AVLLMs, analyzing how audio and visual features evolve and fuse through different layers of an AVLLM to produce the final text outputs. We find that although AVLLMs encode rich audio semantics at intermediate layers, these capabilities largely fail to surface in the final text generation when audio conflicts with vision. Probing analyses show that useful latent audio information is present, but deeper fusion layers disproportionately privilege visual representations that tend to suppress audio cues. We further trace this imbalance to training: the AVLLM's audio behavior strongly matches its vision-language base model, indicating limited additional alignment to audio supervision. Our findings reveal a fundamental modality bias in AVLLMs and provide new mechanistic insights into how multimodal LLMs integrate audio and vision.
Authors: Yuntao Du, Minh Dinh, Kaiyuan Zhang, Ninghui Li
Abstract: Scientific and Technical Intelligence (S&TI) analysis requires verifying complex technical claims across rapidly growing literature, where existing approaches fail to bridge the verification gap between surface-level accuracy and deeper methodological validity. We present AutoVerifier, an LLM-based agentic framework that automates end-to-end verification of technical claims without requiring domain expertise. AutoVerifier decomposes every technical assertion into structured claim triples of the form (Subject, Predicate, Object), constructing knowledge graphs that enable structured reasoning across six progressively enriching layers: corpus construction and ingestion, entity and claim extraction, intra-document verification, cross-source verification, external signal corroboration, and final hypothesis matrix generation. We demonstrate AutoVerifier on a contested quantum computing claim, where the framework, operated by analysts with no quantum expertise, automatically identified overclaims and metric inconsistencies within the target paper, traced cross-source contradictions, uncovered undisclosed commercial conflicts of interest, and produced a final assessment. These results show that structured LLM verification can reliably evaluate the validity and maturity of emerging technologies, turning raw technical documents into traceable, evidence-backed intelligence assessments.
Authors: Yitao Li, Zhanlin Liu, Anuranjan Pandey, Muni Srikanth
Abstract: Organizing a large-scale knowledge graph into a typed property graph requires structural decisions -- which entities become nodes, which properties become edges, and what schema governs these choices. Existing approaches embed these decisions in pipeline code or extract relations ad hoc, producing schemas that are tightly coupled to their construction process and difficult to reuse for downstream ontology-level tasks. We present an ontology-oriented approach in which the schema is designed from the outset for ontology analysis, entity disambiguation, domain customization, and LLM-guided extraction -- not merely as a byproduct of graph building. The core mechanism is intrinsic-relational routing, which classifies every property as either intrinsic or relational and routes it to the corresponding schema module. This routing produces a declarative schema that is portable across storage backends and independently reusable. We instantiate the approach on the January 2026 Wikidata dump. A rule-based cleaning stage identifies a 34.6M-entity core set from the full dump, followed by iterative intrinsic-relational routing that assigns each property to one of 94 modules organized into 8 categories. With tool-augmented LLM support and human review, the schema reaches 93.3% category coverage and 98.0% module assignment among classified entities. Exporting this schema yields a property graph with 34.0M nodes and 61.2M edges across 38 relationship types. We validate the ontology-oriented claim through five applications that consume the schema independently of the construction pipeline: ontology structure analysis, benchmark annotation auditing, entity disambiguation, domain customization, and LLM-guided extraction.
Authors: Joshua Drossman, Alexandre Jacquillat, S\'ebastien Martin
Abstract: Optimization is as much about modeling the right problem as solving it. Identifying the right objectives, constraints, and trade-offs demands extensive interaction between researchers and stakeholders. Large language models can empower decision-makers with optimization capabilities through interactive optimization agents that can propose, interpret and refine solutions. However, it is fundamentally harder to evaluate a conversation-based interaction than traditional one-shot approaches. This paper proposes a scalable and replicable methodology for evaluating optimization agents through conversations. We build LLM-powered decision agents that role-play diverse stakeholders, each governed by an internal utility function but communicating like a real decision-maker. We generate thousands of conversations in a school scheduling case study. Results show that one-shot evaluation is severely limiting: the same optimization agent converges to much higher-quality solutions through conversations. Then, this paper uses this methodology to demonstrate that tailored optimization agents, endowed with domain-specific prompts and structured tools, can lead to significant improvements in solution quality in fewer interactions, as compared to general-purpose chatbots. These findings provide evidence of the benefits of emerging solutions at the AI-optimization interface to expand the reach of optimization technologies in practice. They also uncover the impact of operations research expertise to facilitate interactive deployments through the design of effective and reliable optimization agents.
Authors: DeepReinforce Team, Xiaoya Li, Xiaofei Sun, Guoyin Wang, Songqiao Su, Chris Shum, Jiwei Li
Abstract: Competitive programming remains one of the last few human strongholds in coding against AI. The best AI system to date still underperforms the best humans competitive programming: the most recent best result, Google's Gemini~3 Deep Think, attained 8th place even not being evaluated under live competition conditions. In this work, we introduce GrandCode, a multi-agent RL system designed for competitive programming. The capability of GrandCode is attributed to two key factors: (1) It orchestrates a variety of agentic modules (hypothesis proposal, solver, test generator, summarization, etc) and jointly improves them through post-training and online test-time RL; (2) We introduce Agentic GRPO specifically designed for multi-stage agent rollouts with delayed rewards and the severe off-policy drift that is prevalent in agentic RL. GrandCode is the first AI system that consistently beats all human participants in live contests of competitive programming: in the most recent three Codeforces live competitions, i.e., Round~1087 (Mar 21, 2026), Round~1088 (Mar 28, 2026), and Round~1089 (Mar 29, 2026), GrandCode placed first in all of them, beating all human participants, including legendary grandmasters. GrandCode shows that AI systems have reached a point where they surpass the strongest human programmers on the most competitive coding tasks.
Authors: Amit Dhanda
Abstract: Reasoning benchmarks typically evaluate whether a model derives the correct answer from a fixed premise set, but they under-measure a closely related capability that matters in dynamic environments: belief revision under minimal evidence change. We introduce DeltaLogic, a benchmark transformation protocol that converts natural-language reasoning examples into short revision episodes. Each episode first asks for an initial conclusion under premises P, then applies a minimal edit {\delta}(P), and finally asks whether the previous conclusion should remain stable or be revised. We instantiate DeltaLogic from FOLIO and ProofWriter and evaluate small causal language models with constrained label scoring. On a completed 30-episode Qwen evaluation subset, stronger initial reasoning still does not imply stronger revision behavior: Qwen3-1.7B reaches 0.667 initial accuracy but only 0.467 revision accuracy, with inertia rising to 0.600 on episodes where the gold label should change, while Qwen3-0.6B collapses into near universal abstention. There, Qwen3-4B preserves the same inertial failure pattern (0.650 initial, 0.450 revised, 0.600 inertia), whereas Phi-4-mini-instruct is substantially stronger (0.950 initial, 0.850 revised) but still exhibits non-trivial abstention and control instability. These results suggest that logical competence under fixed premises does not imply disciplined belief revision after local evidence edits. DeltaLogic therefore targets a distinct and practically important reasoning capability that complements existing logical inference and belief-updating benchmarks.
Authors: Bin Wen, Ruoxuan Zhang, Yang Chen, Hongxia Xie, Lan-Zhe Guo
Abstract: Large language models (LLMs) have demonstrated strong potential in long-horizon decision-making tasks, such as embodied manipulation and web interaction. However, agents frequently struggle with endless trial-and-error loops or deviate from the main objective in complex environments. We attribute these failures to two fundamental errors: global Progress Drift and local Feasibility Violation. Existing methods typically attempt to address both issues simultaneously using a single paradigm. However, these two challenges are fundamentally distinct: the former relies on fuzzy semantic planning, while the latter demands strict logical constraints and state validation. The inherent limitations of such a single-paradigm approach pose a fundamental challenge for existing models in handling long-horizon tasks. Motivated by this insight, we propose a Neuro-Symbolic Dual Memory Framework that explicitly decouples semantic progress guidance from logical feasibility verification. Specifically, during the inference phase, the framework invokes both memory mechanisms synchronously: on one hand, a neural-network-based Progress Memory extracts semantic blueprints from successful trajectories to guide global task advancement; on the other hand, a symbolic-logic-based Feasibility Memory utilizes executable Python verification functions synthesized from failed transitions to perform strict logical validation. Experiments demonstrate that this method significantly outperforms existing competitive baselines on ALFWorld, WebShop, and TextCraft, while drastically reducing the invalid action rate and average trajectory length.
Authors: Guoling Zhou, Wenpei Han, Fengqin Yang, Li Wang, Yingcong Zhou, Zhiguo Fu
Abstract: In large language model (LLM)-driven multi-agent systems, disobey role specification (failure to adhere to the defined responsibilities and constraints of an assigned role, potentially leading to an agent behaving like another) is a major failure mode \cite{DBLP:journals/corr/abs-2503-13657}. To address this issue, in the present paper, we propose a quantitative role clarity to improve role consistency. Firstly, we construct a role assignment matrix $S(\phi)=[s_{ij}(\phi)]$, where $s_{ij}(\phi)$ is the semantic similarity between the $i$-th agent's behavior trajectory and the $j$-th agent's role description. Then we define role clarity matrix $M(\phi)$ as $\text{softmax}(S(\phi))-I$, where $\text{softmax}(S(\phi))$ is a row-wise softmax of $S(\phi)$ and $I$ is the identity matrix. The Frobenius norm of $M(\phi)$ quantifies the alignment between agents' role descriptions and their behaviors trajectory. Moreover, we employ the role clarity matrix as a regularizer during lightweight fine-tuning to improve role consistency, thereby improving end-to-end task performance. Experiments on the ChatDev multi-agent system show that our method substantially improves role consistency and task performance: with Qwen and Llama, the role overstepping rate decreases from $46.4\%$ to $8.4\%$ and from $43.4\%$ to $0.2\%$, respectively, and the role clarity score increases from $0.5328$ to $0.9097$ and from $0.5007$ to $0.8530$, respectively, the task success rate increases from $0.6769$ to $0.6909$ and from $0.6174$ to $0.6763$, respectively.
Authors: Situo Zhang, Yifan Zhang, Zichen Zhu, Da Ma, Lei Pan, Danyang Zhang, Zihan Zhao, Lu Chen, Kai Yu
Abstract: Charts are ubiquitous in scientific and financial literature for presenting structured data. However, chart reasoning remains challenging for multimodal large language models (MLLMs) due to the lack of high-quality training data, as well as the need for fine-grained visual grounding and precise numerical computation. To address these challenges, we first propose DuoChart, a scalable dual-source data pipeline that combines synthesized charts with real-world charts to construct diverse, high-quality chart training data. We then introduce CharTool, which equips MLLMs with external tools, including image cropping for localized visual perception and code-based computation for accurate numerical reasoning. Through agentic reinforcement learning on DuoChart, CharTool learns tool-integrated reasoning grounded in chart content. Extensive experiments on six chart benchmarks show that our method consistently improves over strong MLLM baselines across model scales. Notably, CharTool-7B outperforms the base model by **+8.0%** on CharXiv (Reasoning) and **+9.78%** on ChartQAPro, while achieving competitive performance with substantially larger or proprietary models. Moreover, CharTool demonstrates positive generalization to out-of-domain visual math reasoning benchmarks.
Authors: Chao Li, Cailiang Liu, Ang Gao, Kexin Deng, Shu Zhang, Langping Xu, Xiaotong Shi, Xionghao Ding, Jian Pei, Xun Jiang
Abstract: Longitudinal health agents must reason across multi-source trajectories that combine continuous device streams, sparse clinical exams, and episodic life events - yet evaluating them is hard: real-world data cannot be released at scale, and temporally grounded attribution questions seldom admit definitive answers without structured ground truth. We present ESL-Bench, an event-driven synthesis framework and benchmark providing 100 synthetic users, each with a 1-5 year trajectory comprising a health profile, a multi-phase narrative plan, daily device measurements, periodic exam records, and an event log with explicit per-indicator impact parameters. Each indicator follows a baseline stochastic process driven by discrete events with sigmoid-onset, exponential-decay kernels under saturation and projection constraints; a hybrid pipeline delegates sparse semantic artifacts to LLM-based planning and dense indicator dynamics to algorithmic simulation with hard physiological bounds. Users are each paired with 100 evaluation queries across five dimensions - Lookup, Trend, Comparison, Anomaly, Explanation - stratified into Easy, Medium, and Hard tiers, with all ground-truth answers programmatically computable from the recorded event-indicator relationships. Evaluating 13 methods spanning LLMs with tools, DB-native agents, and memory-augmented RAG, we find that DB agents (48-58%) substantially outperform memory RAG baselines (30-38%), with the gap concentrated on Comparison and Explanation queries where multi-hop reasoning and evidence attribution are required.
Authors: Yiqing Liu, Hantao Yao, Wu Liu, Yongdong Zhang
Abstract: Majority voting is the standard for aggregating multi-agent responses into a final decision. However, traditional methods typically require all agents to complete their reasoning before aggregation begins, leading to significant computational overhead, as many responses become redundant once a majority consensus is achieved. In this work, we formulate the multi-agent voting as a reliability-aware agent scheduling problem, and propose an Efficient Majority-then-Stopping (EMS) to improve reasoning efficiency. EMS prioritizes agents based on task-aware reliability and terminates the reasoning pipeline the moment a majority is achieved from the following three critical components. Specifically, we introduce Agent Confidence Modeling (ACM) to estimate agent reliability using historical performance and semantic similarity, Adaptive Incremental Voting (AIV) to sequentially select agents with early stopping, and Individual Confidence Updating (ICU) to dynamically update the reliability of each contributing agent. Extensive evaluations across six benchmarks demonstrate that EMS consistently reduces the average number of invoked agents by 32%.
Authors: Wachiravit Modecrua, Krittanon Kaewtawee, Krittin Pachtrachai, Touchapon Kraisingkorn
Abstract: Training tool-calling agents with reinforcement learning on multi-turn tasks remains challenging due to sparse outcome rewards and difficult credit assignment across conversation turns. We present the first application of MT-GRPO (Multi-Turn Group Relative Policy Optimization) combined with GTPO (Generalized Token-level Policy Optimization) for training a tool-calling agent on realistic customer service tasks with an LLM-based user simulator. Through systematic analysis of training rollouts, we discover that naively designed dense per-turn rewards degrade performance by up to 14 percentage points due to misalignment between reward discriminativeness and advantage direction. We introduce Iterative Reward Calibration, a methodology for designing per-turn rewards using empirical discriminative analysis of rollout data, and show that our GTPO hybrid advantage formulation eliminates the advantage misalignment problem. Applied to the Tau-Bench airline benchmark, our approach improves Qwen3.5-4B from 63.8 percent to 66.7 percent (+2.9pp) and Qwen3-30B-A3B from 58.0 percent to 69.5 percent (+11.5pp) -- with the trained 4B model exceeding GPT-4.1 (49.4 percent) and GPT-4o (42.8 percent) despite being 50 times smaller, and the 30.5B MoE model approaching Claude Sonnet 4.5 (70.0 percent). To our knowledge, these are the first published RL training results on Tau-Bench. We release our code, reward calibration analysis, and training recipes.
Authors: Bernd Bohnet, Michael C. Mozer, Kevin Swersky, Wil Cunningham, Aaron Parisi, Kathleen Kenealy, Noah Fiedel
Abstract: Classic AI planning problems have been revisited in the Large Language Model (LLM) era, with a focus of recent benchmarks on success rates rather than plan efficiency. We examine the degree to which frontier models reason optimally versus relying on simple, heuristic, and possibly inefficient strategies. We focus on the Blocksworld domain involving towers of labeled blocks which have to be moved from an initial to a goal configuration via a set of primitive actions. We also study a formally equivalent task, the generalized Path-Star ($P^*$) graph, in order to isolate true topological reasoning from semantic priors. We systematically manipulate problem depth (the height of block towers), width (the number of towers), and compositionality (the number of goal blocks). Reasoning-enhanced LLMs significantly outperform traditional satisficing planners (e.g., LAMA) in complex, multi-goal configurations. Although classical search algorithms hit a wall as the search space expands, LLMs track theoretical optimality limits with near-perfect precision, even when domain-specific semantic hints are stripped away. To explain these surprising findings, we consider (and find evidence to support) two hypotheses: an active Algorithmic Simulation executed via reasoning tokens and a Geometric Memory that allows models to represent the $P^*$ topology as a navigable global geometry, effectively bypassing exponential combinatorial complexity.
Authors: Yunhao Feng, Yifan Ding, Yingshui Tan, Xingjun Ma, Yige Li, Yutao Wu, Yifeng Gao, Kun Zhai, Yanming Guo
Abstract: Computer-use agents extend language models from text generation to persistent action over tools, files, and execution environments. Unlike chat systems, they maintain state across interactions and translate intermediate outputs into concrete actions. This creates a distinct safety challenge in that harmful behavior may emerge through sequences of individually plausible steps, including intermediate actions that appear locally acceptable but collectively lead to unauthorized actions. We present \textbf{AgentHazard}, a benchmark for evaluating harmful behavior in computer-use agents. AgentHazard contains \textbf{2,653} instances spanning diverse risk categories and attack strategies. Each instance pairs a harmful objective with a sequence of operational steps that are locally legitimate but jointly induce unsafe behavior. The benchmark evaluates whether agents can recognize and interrupt harm arising from accumulated context, repeated tool use, intermediate actions, and dependencies across steps. We evaluate AgentHazard on Claude Code, OpenClaw, and IFlow using mostly open or openly deployable models from the Qwen3, Kimi, GLM, and DeepSeek families. Our experimental results indicate that current systems remain highly vulnerable. In particular, when powered by Qwen3-Coder, Claude Code exhibits an attack success rate of \textbf{73.63\%}, suggesting that model alignment alone does not reliably guarantee the safety of autonomous agents.
Authors: Kehan Jiang, Haonan Dong, Zhaolu Kang, Zhengzhou Zhu, Guojie Song
Abstract: Recent Large Reasoning Models (LRMs) like DeepSeek-R1 have demonstrated remarkable success in complex reasoning tasks, exhibiting human-like patterns in exploring multiple alternative solutions. Upon closer inspection, however, we uncover a surprising phenomenon: The First is The Best, where alternative solutions are not merely suboptimal but potentially detrimental. This observation challenges widely accepted test-time scaling laws, leading us to hypothesize that errors within the reasoning path scale concurrently with test time. Through comprehensive empirical analysis, we characterize errors as a forest-structured Forest of Errors (FoE) and conclude that FoE makes the First the Best, which is underpinned by rigorous theoretical analysis. Leveraging these insights, we propose RED, a self-guided efficient reasoning framework comprising two components: I) Refining First, which suppresses FoE growth in the first solution; and II) Discarding Subs, which prunes subsequent FoE via dual-consistency. Extensive experiments across five benchmarks and six backbone models demonstrate that RED outperforms eight competitive baselines, achieving performance gains of up to 19.0% while reducing token consumption by 37.7% ~ 70.4%. Moreover, comparative experiments on FoE metrics shed light on how RED achieves effectiveness.
Authors: Ka Yiu Lee, Yuxuan Huang, Zhiyuan He, Huichi Zhou, Weilin Luo, Kun Shao, Meng Fang, Jun Wang
Abstract: Recent agentic search systems have made substantial progress by emphasising deep, multi-step reasoning. However, this focus often overlooks the challenges of wide-scale information synthesis, where agents must aggregate large volumes of heterogeneous evidence across many sources. As a result, most existing large language model agent systems face severe limitations in data-intensive settings, including context saturation, cascading error propagation, and high end-to-end latency. To address these challenges, we present \framework, a hierarchical framework based on principle of near-decomposability, containing a strategic \textit{Host}, multiple \textit{Managers} and parallel \textit{Workers}. By leveraging aggregation and reflection mechanisms at the Manager layer, our framework enforces strict context isolation to prevent saturation and error propagation. Simultaneously, the parallelism in worker layer accelerates the speed of overall task execution, mitigating the significant latency. Our evaluation on two complementary benchmarks demonstrates both efficiency ($ 3-5 \times$ speed-up) and effectiveness, achieving a $8.4\%$ success rate on WideSearch-en and $52.9\%$ accuracy on BrowseComp-zh. The code is released at https://github.com/agent-on-the-fly/InfoSeeker
Authors: Qianshan Wei, Yishan Yang, Siyi Wang, Jinglin Chen, Binyu Wang, Jiaming Wang, Shuang Chen, Zechen Li, Yang Shi, Yuqi Tang, Weining Wang, Yi Yu, Chaoyou Fu, Qi Li, Yi-Fan Zhang
Abstract: Multimodal Large Language Models (MLLMs) are evolving from passive observers into active agents, solving problems through Visual Expansion (invoking visual tools) and Knowledge Expansion (open-web search). However, existing evaluations fall short: they lack flexible tool integration, test visual and search tools separately, and evaluate primarily by final answers. Consequently, they cannot verify if tools were actually invoked, applied correctly, or used efficiently. To address this, we introduce Agentic-MME, a process-verified benchmark for Multimodal Agentic Capabilities. It contains 418 real-world tasks across 6 domains and 3 difficulty levels to evaluate capability synergy, featuring over 2,000 stepwise checkpoints that average 10+ person-hours of manual annotation per task. Each task includes a unified evaluation framework supporting sandboxed code and APIs, alongside a human reference trajectory annotated with stepwise checkpoints along dual-axis: S-axis and V-axis. To enable true process-level verification, we audit fine-grained intermediate states rather than just final answers, and quantify efficiency via an overthinking metric relative to human trajectories. Experimental results show the best model, Gemini3-pro, achieves 56.3% overall accuracy, which falls significantly to 23.0% on Level-3 tasks, underscoring the difficulty of real-world multimodal agentic problem solving.
Authors: Fabian Gloeckle, Ahmad Rammal, Charles Arnal, Remi Munos, Vivien Cabannes, Gabriel Synnaeve, Amaury Hayat
Abstract: We present a case study where an automatic AI system formalizes a textbook with more than 500 pages of graduate-level algebraic combinatorics to Lean. The resulting formalization represents a new milestone in textbook formalization scale and proficiency, moving from early results in undergraduate topology and restructuring of existing library content to a full standalone formalization of a graduate textbook. The formalization comprises 130K lines of code and 5900 Lean declarations and was conducted within one week by a total of 30K Claude 4.5 Opus agents collaborating in parallel on a shared code base via version control, simultaneously setting a record in multi-agent software engineering with usable results. The inference cost matches or undercuts what we estimate as the salaries required for a team of human experts, and we expect there is still the potential for large efficiencies to be made without the need for better models. We make our code, the resulting Lean code base and a side-by-side blueprint website available open-source.
Authors: Yunfei Bai, Amit Dhanda, Shekhar Jain
Abstract: The recent advancements in Vision Language Models (VLMs) have demonstrated progress toward true intelligence requiring robust reasoning capabilities. Beyond pattern recognition, linguistic reasoning must integrate with visual comprehension, particularly for Chart Question Answering (CQA) tasks involving complex data visualizations. Current VLMs face significant limitations in CQA, including imprecise numerical extraction, difficulty interpreting implicit visual relationships, and inadequate attention mechanisms for capturing spatial relationships in charts. In this work, we address these challenges by presenting Chart-RL, a novel reinforcement learning framework that enhances VLMs chart understanding through feedback-driven policy optimization of visual perception and logical inference. Our key innovation includes a comprehensive framework integrating Reinforcement Learning (RL) from Policy Optimization techniques along with adaptive reward functions, that demonstrates superior performance compared to baseline foundation models and competitive results against larger state-of-the-art architectures. We also integrated Parameter-Efficient Fine-Tuning through Low-Rank Adaptation (LoRA) in the RL framework that only requires single GPU configurations while preserving performance integrity. We conducted extensive benchmarking across open-source, proprietary, and state-of-the-art closed-source models utilizing the ChartQAPro dataset. The RL fine-tuned Qwen3-VL-4B-Instruct model achieved an answer accuracy of 0.634, surpassing the 0.580 accuracy of the Qwen3-VL-8B-Instruct foundation model despite utilizing half the parameter count, while simultaneously reducing inference latency from 31 seconds to 9 seconds.
Authors: Maximiliano Armesto, Christophe Kolb
Abstract: Agentic AI is increasingly judged not by fluent output alone but by whether it can act, remember, and verify under partial observability, delay, and strategic observation. Existing research often studies these demands separately: robotics emphasizes control, retrieval systems emphasize memory, and alignment or assurance work emphasizes checking and oversight. This article argues that squirrel ecology offers a sharp comparative case because arboreal locomotion, scatter-hoarding, and audience-sensitive caching couple all three demands in one organism. We synthesize evidence from fox, eastern gray, and, in one field comparison, red squirrels, and impose an explicit inference ladder: empirical observation, minimal computational inference, and AI design conjecture. We introduce a minimal hierarchical partially observed control model with latent dynamics, structured episodic memory, observer-belief state, option-level actions, and delayed verifier signals. This motivates three hypotheses: (H1) fast local feedback plus predictive compensation improves robustness under hidden dynamics shifts; (H2) memory organized for future control improves delayed retrieval under cue conflict and load; and (H3) verifiers and observer models inside the action-memory loop reduce silent failure and information leakage while remaining vulnerable to misspecification. A downstream conjecture is that role-differentiated proposer/executor/checker/adversary systems may reduce correlated error under asymmetric information and verification burden. The contribution is a comparative perspective and benchmark agenda: a disciplined program of falsifiable claims about the coupling of control, memory, and verifiable action.
Authors: Jakob Prange, Nathan Schneider, Lingpeng Kong
Abstract: We examine the extent to which, in principle, linguistic graph representations can complement and improve neural language modeling. With an ensemble setup consisting of a pretrained Transformer and ground-truth graphs from one of 7 different formalisms, we find that, overall, semantic constituency structures are most useful to language modeling performance -- outpacing syntactic constituency structures as well as syntactic and semantic dependency structures. Further, effects vary greatly depending on part-of-speech class. In sum, our findings point to promising tendencies in neuro-symbolic language modeling and invite future research quantifying the design choices made by different formalisms.
Authors: Jakob Prange, Man Ho Ivy Wong
Abstract: We use both Bayesian and neural models to dissect a data set of Chinese learners' pre- and post-interventional responses to two tests measuring their understanding of English prepositions. The results mostly replicate previous findings from frequentist analyses and newly reveal crucial interactions between student ability, task type, and stimulus sentence. Given the sparsity of the data as well as high diversity among learners, the Bayesian method proves most useful; but we also see potential in using language model probabilities as predictors of grammaticality and learnability.
Authors: Jakob Prange, Emmanuele Chersoni
Abstract: In this work we build upon negative results from an attempt at language modeling with predicted semantic structure, in order to establish empirical lower bounds on what could have made the attempt successful. More specifically, we design a concise binary vector representation of semantic structure at the lexical level and evaluate in-depth how good an incremental tagger needs to be in order to achieve better-than-baseline performance with an end-to-end semantic-bootstrapping language model. We envision such a system as consisting of a (pretrained) sequential-neural component and a hierarchical-symbolic component working together to generate text with low surprisal and high linguistic interpretability. We find that (a) dimensionality of the semantic vector representation can be dramatically reduced without losing its main advantages and (b) lower bounds on prediction quality cannot be established via a single score alone, but need to take the distributions of signal and noise into account.
Authors: Mohammad Rezaei, Jens Lehmann, Sahar Vahdati
Abstract: Mathematical reasoning in large language models has improved substantially with reinforcement learning using verifiable rewards, where final answers can be checked automatically and converted into reliable training signals. Most such pipelines optimize outcome correctness only, which yields sparse feedback for long, multi-step solutions and offers limited guidance on intermediate reasoning errors. Recent work therefore introduces process reward models (PRMs) to score intermediate steps and provide denser supervision. In practice, PRM scores are often imperfectly aligned with final correctness and can reward locally fluent reasoning that still ends in an incorrect answer. When optimized as absolute rewards, such signals can amplify fluent failure modes and induce reward hacking. We propose PROGRS, a framework that leverages PRMs while keeping outcome correctness dominant. PROGRS treats process rewards as relative preferences within outcome groups rather than absolute targets. We introduce outcome-conditioned centering, which shifts PRM scores of incorrect trajectories to have zero mean within each prompt group. It removes systematic bias while preserving informative rankings. PROGRS combines a frozen quantile-regression PRM with a multi-scale coherence evaluator. We integrate the resulting centered process bonus into Group Relative Policy Optimization (GRPO) without auxiliary objectives or additional trainable components. Across MATH-500, AMC, AIME, MinervaMath, and OlympiadBench, PROGRS consistently improves Pass@1 over outcome-only baselines and achieves stronger performance with fewer rollouts. These results show that outcome-conditioned centering enables safe and effective use of process rewards for mathematical reasoning.
Authors: Roy Rinberg, Annabelle Michael Carrell, Simon Henniger, Nicholas Carlini, Keri Warr
Abstract: We study the compression of LLM-generated text across lossless and lossy regimes, characterizing a compression-compute frontier where more compression is possible at the cost of more compute. For lossless compression, domain-adapted LoRA adapters can improve LLM-based arithmetic coding by 2x over compression with the base LLM alone. For lossy compression, prompting a model for a succinct rewrite then applying arithmetic coding can achieve compression ratios of approximately 0.03, a 2x improvement over compressing the original response. We further introduce Question-Asking compression (QA), an interactive lossy protocol inspired by the game 'Twenty Questions'. A small model iteratively refines its response by asking yes/no questions to a stronger model, transferring exactly one bit per answer. On 8 benchmarks spanning math, science, and code, 10 binary questions recover 23% to 72% of the capability gap between a small and large model on standard benchmarks and 7% to 38% on harder benchmarks, achieving compression ratios of 0.0006 to 0.004. This is over 100x smaller than prior LLM-based compression (Deletang et al., 2024), suggesting that interactive protocols can transfer knowledge far more efficiently than transmitting full responses.
Authors: Mengzhou Wu, Yuzhe Guo, Yuan Cao, Haochuan Lu, Songhe Zhu, Pingzhe Qu, Xin Chen, Kang Qin, Zhongpu Wang, Xiaode Zhang, Xinyi Wang, Wei Dai, Gang Cao, Yuetang Deng, Zhi Gong, Dezhi Ran, Linyi Li, Wei Yang, Tao Xie
Abstract: Scaling generalist GUI agents is hindered by the data scalability bottleneck of expensive human demonstrations and the "distillation ceiling" of synthetic teacher supervision. To transcend these limitations, we propose UI-Oceanus, a framework that shifts the learning focus from mimicking high-level trajectories to mastering interaction physics via ground-truth environmental feedback. Through a systematic investigation of self-supervised objectives, we identify that forward dynamics, defined as the generative prediction of future interface states, acts as the primary driver for scalability and significantly outweighs inverse inference. UI-Oceanus leverages this insight by converting low-cost autonomous exploration, which is verified directly by system execution, into high-density generative supervision to construct a robust internal world model. Experimental evaluations across a series of models demonstrate the decisive superiority of our approach: models utilizing Continual Pre-Training (CPT) on synthetic dynamics outperform non-CPT baselines with an average success rate improvement of 7% on offline benchmarks, which amplifies to a 16.8% gain in real-world online navigation. Furthermore, we observe that navigation performance scales with synthetic data volume. These results confirm that grounding agents in forward predictive modeling offers a superior pathway to scalable GUI automation with robust cross-domain adaptability and compositional generalization.
Authors: Tianyu Liu, Sihan Jiang, Fan Zhang, Kunyang Sun, Teresa Head-Gordon, Hongyu Zhao
Abstract: Large language models (LLMs) are in the ascendancy for research in drug discovery, offering unprecedented opportunities to reshape drug research by accelerating hypothesis generation, optimizing candidate prioritization, and enabling more scalable and cost-effective drug discovery pipelines. However there is currently a lack of objective assessments of LLM performance to ascertain their advantages and limitations over traditional drug discovery platforms. To tackle this emergent problem, we have developed DrugPlayGround, a framework to evaluate and benchmark LLM performance for generating meaningful text-based descriptions of physiochemical drug characteristics, drug synergism, drug-protein interactions, and the physiological response to perturbations introduced by drug molecules. Moreover, DrugPlayGround is designed to work with domain experts to provide detailed explanations for justifying the predictions of LLMs, thereby testing LLMs for chemical and biological reasoning capabilities to push their greater use at the frontier of drug discovery at all of its stages.
Authors: Yiqin Yang, Hao Hu, Yihuan Mao, Jin Zhang, Chengjie Wu, Yuhua Jiang, Xu Yang, Runpeng Xie, Yi Fan, Bo Liu, Yang Gao, Bo Xu, Chongjie Zhang
Abstract: Preference-based reinforcement learning (PbRL) can help avoid sophisticated reward designs and align better with human intentions, showing great promise in various real-world applications. However, obtaining human feedback for preferences can be expensive and time-consuming, which forms a strong barrier for PbRL. In this work, we address the problem of low query efficiency in offline PbRL, pinpointing two primary reasons: inefficient exploration and overoptimization of learned reward functions. In response to these challenges, we propose a novel algorithm, \textbf{O}ffline \textbf{P}b\textbf{R}L via \textbf{I}n-\textbf{D}ataset \textbf{E}xploration (OPRIDE), designed to enhance the query efficiency of offline PbRL. OPRIDE consists of two key features: a principled exploration strategy that maximizes the informativeness of the queries and a discount scheduling mechanism aimed at mitigating overoptimization of the learned reward functions. Through empirical evaluations, we demonstrate that OPRIDE significantly outperforms prior methods, achieving strong performance with notably fewer queries. Moreover, we provide theoretical guarantees of the algorithm's efficiency. Experimental results across various locomotion, manipulation, and navigation tasks underscore the efficacy and versatility of our approach.
Authors: Venkatakrishna Reddy Oruganti
Abstract: Neural networks excel at pattern recognition but struggle with constraint reasoning -- determining whether configurations satisfy logical or physical constraints. We introduce Differentiable Symbolic Planning (DSP), a neural architecture that performs discrete symbolic reasoning while remaining fully differentiable. DSP maintains a feasibility channel (phi) that tracks constraint satisfaction evidence at each node, aggregates this into a global feasibility signal (Phi) through learned rule-weighted combination, and uses sparsemax attention to achieve exact-zero discrete rule selection. We integrate DSP into a Universal Cognitive Kernel (UCK) that combines graph attention with iterative constraint propagation. Evaluated on three constraint reasoning benchmarks -- graph reachability, Boolean satisfiability, and planning feasibility -- UCK+DSP achieves 97.4% accuracy on planning under 4x size generalization (vs. 59.7% for ablated baselines), 96.4% on SAT under 2x generalization, and maintains balanced performance on both positive and negative classes where standard neural approaches collapse. Ablation studies reveal that global phi aggregation is critical: removing it causes accuracy to drop from 98% to 64%. The learned phi signal exhibits interpretable semantics, with values of +18 for feasible cases and -13 for infeasible cases emerging without supervision.
Authors: Sophie Weidmann, Fernando Castor
Abstract: Although LLMs are capable of generating functionally correct code, they also tend to produce less energy-efficient code in comparison to human-written solutions. As these inefficiencies lead to higher computational overhead, they are in direct conflict with Green Software Development (GSD) efforts, which aim to reduce the energy consumption of code. To support these efforts, this study aims to investigate whether and how LLMs can be optimized to promote the generation of energy-efficient code. To this end, we employ Contrastive Prompt Tuning (CPT). CPT combines Contrastive Learning techniques, which help the model to distinguish between efficient and inefficient code, and Prompt Tuning, a Parameter-Efficient Fine Tuning (PEFT) approach that requires only a fraction of the cost of traditional fine tuning. This study evaluates CPT on Python, Java and C++ coding problems across three different models to provide a comprehensive evaluation. The method achieves consistent improvements in code accuracy for two models but efficiency gains vary by model, language and task complexity, indicating that improvements are not uniformly reliable.
Authors: Thomas Pravetz
Abstract: We present PRISM (Policy Reuse via Interpretable Strategy Mapping), a framework that grounds reinforcement learning agents' decisions in discrete, causally validated concepts and uses those concepts as a zero-shot transfer interface between agents trained with different algorithms. PRISM clusters each agent's encoder features into $K$ concepts via K-means. Causal intervention establishes that these concepts directly drive - not merely correlate with - agent behavior: overriding concept assignments changes the selected action in 69.4% of interventions ($p = 8.6 \times 10^{-86}$, 2500 interventions). Concept importance and usage frequency are dissociated: the most-used concept (C47, 33.0% frequency) causes only a 9.4% win-rate drop when ablated, while ablating C16 (15.4% frequency) collapses win rate from 100% to 51.8%. Because concepts causally encode strategy, aligning them via optimal bipartite matching transfers strategic knowledge zero-shot. On Go~7$\times$7 with three independently trained agents, concept transfer achieves 69.5%$\pm$3.2% and 76.4%$\pm$3.4% win rate against a standard engine across the two successful transfer pairs (10 seeds), compared to 3.5% for a random agent and 9.2% without alignment. Transfer succeeds when the source policy is strong; geometric alignment quality predicts nothing ($R^2 \approx 0$). The framework is scoped to domains where strategic state is naturally discrete: the identical pipeline on Atari Breakout yields bottleneck policies at random-agent performance, confirming that the Go results reflect a structural property of the domain.
Authors: Gaoxiang Cao, Wenke Yuan, Yunpeng Hou, Huasen He, Quan Zheng, Jian Yang
Abstract: Vehicular Ad Hoc Networks (VANETs) play a crucial role in realizing vehicle-road collaboration and intelligent transportation. However, urban VANETs often face challenges such as frequent link disconnections and subnet fragmentation, which hinder reliable connectivity. To address these issues, we dynamically deploy multiple Unmanned Aerial Vehicles (UAVs) as communication relays to enhance VANET. A novel Score based Dynamic Action Mask enhanced QMIX algorithm (Q-SDAM) is proposed for multi-UAV deployment, which maximizes vehicle connectivity while minimizing multi-UAV energy consumption. Specifically, we design a score-based dynamic action mask mechanism to guide UAV agents in exploring large action spaces, accelerate the learning process and enhance optimization performance. The practicality of Q-SDAM is validated using real-world datasets. We show that Q-SDAM improves connectivity by 18.2% while reducing energy consumption by 66.6% compared with existing algorithms.
Authors: May Lynn Reese, Markela Zeneli, Mindy Ng, Jacob Haimes, Andreea Damien, Elizabeth Stade
Abstract: General-purpose Large Language Models (LLMs) are becoming widely adopted by people for mental health support. Yet emerging evidence suggests there are significant risks associated with high-frequency use, particularly for individuals suffering from psychosis, as LLMs may reinforce delusions and hallucinations. Existing evaluations of LLMs in mental health contexts are limited by a lack of clinical validation and scalability of assessment. To address these issues, this research focuses on psychosis as a critical condition for LLM safety evaluation by (1) developing and validating seven clinician-informed safety criteria, (2) constructing a human-consensus dataset, and (3) testing automated assessment using an LLM as an evaluator (LLM-as-a-Judge) or taking the majority vote of several LLM judges (LLM-as-a-Jury). Results indicate that LLM-as-a-Judge aligns closely with the human consensus (Cohen's $\kappa_{\text{human} \times \text{gemini}} = 0.75$, $\kappa_{\text{human} \times \text{qwen}} = 0.68$, $\kappa_{\text{human} \times \text{kimi}} = 0.56$) and that the best judge slightly outperforms LLM-as-a-Jury (Cohen's $\kappa_{\text{human} \times \text{jury}} = 0.74$). Overall, these findings have promising implications for clinically grounded, scalable methods in LLM safety evaluations for mental health contexts.
Authors: Raul Suzuki, Rodrigo Moreira, Pedro Henrique A. Damaso de Melo, Larissa F. Rodrigues Moreira, Fl\'avio de Oliveira Silva
Abstract: Detecting Internet routing instability is a critical yet challenging task, particularly when relying solely on endpoint active measurements. This study introduces TRACE, a MachineLearning (ML)pipeline designed to identify route changes using only traceroute latency data, thereby ensuring independence from control plane information. We propose a robust feature engineering strategy that captures temporal dynamics using rolling statistics and aggregated context patterns. The architecture leverages a stacked ensemble of Gradient Boosted Decision Trees refined by a hyperparameter-optimized meta-learner. By strictly calibrating decision thresholds to address the inherent class imbalance of rare routing events, TRACE achieves a superior F1-score performance, significantly outperforming traditional baseline models and demonstrating strong effective ness in detecting routing changes on the Internet.
Authors: Varshith Madishetty
Abstract: Decoding speech information from scalp EEG remains difficult due to low SNR and spatial blurring. We present CIPHER (Conformer-based Inference of Phonemes from High-density EEG Representations), a dual-pathway model using (i) ERP features and (ii) broadband DDA coefficients. On OpenNeuro ds006104 (24 participants, two studies with concurrent TMS), binary articulatory tasks reach near-ceiling performance but are highly confound-vulnerable (acoustic onset separability and TMS-target blocking). On the primary 11-class CVC phoneme task under full Study 2 LOSO (16 held-out subjects), performance is substantially lower (real-word WER: ERP 0.671 +/- 0.080, DDA 0.688 +/- 0.096, indicating limited fine-grained discriminability. We therefore position this work as a benchmark and feature-comparison study rather than an EEG-to-text system, and we constrain neural-representation claims to confound-controlled evidence.
Authors: Dun Yuan, Fuyuan Lyu, Ye Yuan, Weixu Zhang, Bowei He, Jiayi Geng, Linfeng Du, Zipeng Sun, Yankai Chen, Changjiang Han, Jikun Kang, Alex Chen, Haolun Wu, Xue Liu
Abstract: Agent communication protocols are becoming critical infrastructure for large language model (LLM) systems that must use tools, coordinate with other agents, and operate across heterogeneous environments. This work presents a human-inspired perspective on this emerging landscape by organizing agent communication into three layers: communication, syntactic, and semantic. Under this framework, we systematically analyze 18 representative protocols and compare how they support reliable transport, structured interaction, and meaning-level coordination. Our analysis shows a clear imbalance in current protocol design. Most protocols provide increasingly mature support for transport, streaming, schema definition, and lifecycle management, but offer limited protocol-level mechanisms for clarification, context alignment, and verification. As a result, semantic responsibilities are often pushed into prompts, wrappers, or application-specific orchestration logic, creating hidden interoperability and maintenance costs. To make this gap actionable, we further identify major forms of technical debt in today's protocol ecosystem and distill practical guidance for selecting protocols under different deployment settings. We conclude by outlining a research agenda for interoperable, secure, and semantically robust agent ecosystems that move beyond message passing toward shared understanding.
Authors: Constantina Chatzieleftheriou, Eirini Liotou
Abstract: As wireless communication evolves, each generation of networks brings new technologies that change how we connect and interact. Artificial Intelligence (AI) is becoming crucial in shaping the future of sixth-generation (6G) networks. By combining AI and Machine Learning (ML), 6G aims to offer high data rates, low latency, and extensive connectivity for applications including smart cities, autonomous systems, holographic telepresence, and the tactile internet. This paper provides a detailed overview of the role of AI in supporting 6G networks. It focuses on key technologies like deep learning, reinforcement learning, federated learning, and explainable AI. It also looks at how AI integrates with essential network functions and discusses challenges related to scalability, security, and energy efficiency, along with new solutions. Additionally, this work highlights perspectives that connect AI-driven analytics to 6G service domains like Ultra-Reliable Low-Latency Communication (URLLC), Enhanced Mobile Broadband (eMBB), Massive Machine-Type Communication (mMTC), and Integrated Sensing and Communication (ISAC). It addresses concerns about standardization, ethics, and sustainability. By summarizing recent research trends and identifying future directions, this survey offers a valuable reference for researchers and practitioners at the intersection of AI and next-generation wireless communication.
Authors: Austin Veselka
Abstract: Visual long-document understanding is critical for enterprise, legal, and scientific applications, yet the best performing open recipes have not explored reasoning, a capability which has driven leaps in math and code performance. We introduce a synthetic data pipeline for reasoning in long-document understanding that generates thinking traces by scoring each page for question relevance, extracting textual evidence and ordering it from most to least relevant. We apply SFT to the resulting traces within \texttt{
Authors: Zhenning Yang, Kaden Gruizenga, Tongyuan Miao, Patrick Tser Jern Kon, Hui Guan, Ang Chen
Abstract: The scale and complexity of modern cloud infrastructure have made Infrastructure-as-Code (IaC) essential for managing deployments. While large Language models (LLMs) are increasingly being used to generate IaC configurations from natural language, user requests are often underspecified. Unlike traditional code generation, IaC configurations cannot be executed cheaply or iteratively repaired, forcing the LLMs into an almost one-shot regime. We observe that ambiguity in IaC exhibits a tractable compositional structure: configurations decompose into three hierarchical axes (resources, topology, attributes) where higher-level decisions constrain lower-level ones. We propose a training-free, disagreement-driven framework that generates diverse candidate specifications, identifies structural disagreements across these axes, ranks them by informativeness, and produces targeted clarification questions that progressively narrow the configuration space. We introduce \textsc{Ambig-IaC}, a benchmark of 300 validated IaC tasks with ambiguous prompts, and an evaluation framework based on graph edit distance and embedding similarity. Our method outperforms the strongest baseline, achieving relative improvements of +18.4\% and +25.4\% on structure and attribute evaluations, respectively.
Authors: Xinyu Zhou, Yinfeng Yu
Abstract: Audio-visual Navigation refers to an agent utilizing visual and auditory information in complex 3D environments to accomplish target localization and path planning, thereby achieving autonomous navigation. The core challenge of this task lies in the following: how the agent can break free from the dependence on training data and achieve autonomous navigation with good generalization performance when facing changes in environments and sound sources. To address this challenge, we propose an Audio Spatially-Guided Fusion for Audio-Visual Navigation method. First, we design an audio spatial feature encoder, which adaptively extracts target-related spatial state information through an audio intensity attention mechanism; based on this, we introduce an Audio Spatial State Guided Fusion (ASGF) to achieve dynamic alignment and adaptive fusion of multimodal features, effectively alleviating noise interference caused by perceptual uncertainty. Experimental results on the Replica and Matterport3D datasets indicate that our method is particularly effective on unheard tasks, demonstrating improved generalization under unknown sound source distributions.
Authors: Shaohang Wu, Yinfeng Yu
Abstract: Audio-visual navigation tasks require agents to locate and navigate toward continuously vocalizing targets using only visual observations and acoustic cues. However, existing methods mainly rely on simple feature concatenation or late fusion, and lack an explicit discrete representation of the target's relative position, which limits learning efficiency and generalization. We propose Spatial-Aware Conditioned Fusion (SACF). SACF first discretizes the target's relative direction and distance from audio-visual cues, predicts their distributions, and encodes them as a compact descriptor for policy conditioning and state modeling. Then, SACF uses audio embeddings and spatial descriptors to generate channel-wise scaling and bias to modulate visual features via conditional linear transformation, producing target-oriented fused representations. SACF improves navigation efficiency with lower computational overhead and generalizes well to unheard target sounds.
Authors: Teng Liu, Yinfeng Yu
Abstract: Audio-Visual Navigation (AVN) requires an embodied agent to navigate toward a sound source by utilizing both vision and binaural audio. A core challenge arises in complex acoustic environments, where binaural cues become intermittently unreliable, particularly when generalizing to previously unheard sound categories. To address this, we propose RAVN (Reliability-Aware Audio-Visual Navigation), a framework that conditions cross-modal fusion on audio-derived reliability cues, dynamically calibrating the integration of audio and visual inputs. RAVN introduces an Acoustic Geometry Reasoner (AGR) that is trained with geometric proxy supervision. Using a heteroscedastic Gaussian NLL objective, AGR learns observation-dependent dispersion as a practical reliability cue, eliminating the need for geometric labels during inference. Additionally, we introduce Reliability-Aware Geometric Modulation (RAGM), which converts the learned cue into a soft gate to modulate visual features, thereby mitigating cross-modal conflicts. We evaluate RAVN on SoundSpaces using both Replica and Matterport3D environments, and the results show consistent improvements in navigation performance, with notable robustness in the challenging unheard sound setting.
Authors: Xuejian Zhang, Ruisi He, Minseok Kim, Inocent Calist, Mi Yang, Ziyi Qi
Abstract: The deep integration of communication with intelligence and sensing, as a defining vision of 6G, renders environment-aware channel prediction a key enabling technology. As a representative 6G application, vehicular communications require accurate and forward-looking channel prediction under stringent reliability, latency, and adaptability demands. Traditional empirical and deterministic models remain limited in balancing accuracy, generalization, and deployability, while the growing availability of onboard and roadside sensing devices offers a promising source of environmental priors. This paper proposes an environment-aware channel prediction framework based on multimodal visual feature fusion. Using GPS data and vehicle-side panoramic RGB images, together with semantic segmentation and depth estimation, the framework extracts semantic, depth, and position features through a three-branch architecture and performs adaptive multimodal fusion via a squeeze-excitation attention gating module. For 360-dimensional angular power spectrum (APS) prediction, a dedicated regression head and a composite multi-constraint loss are further designed. As a result, joint prediction of path loss (PL), delay spread (DS), azimuth spread of arrival (ASA), azimuth spread of departure (ASD), and APS is achieved. Experiments on a synchronized urban V2I measurement dataset yield the best root mean square error (RMSE) of 3.26 dB for PL, RMSEs of 37.66 ns, 5.05 degrees, and 5.08 degrees for DS, ASA, and ASD, respectively, and mean/median APS cosine similarities of 0.9342/0.9571, demonstrating strong accuracy, generalization, and practical potential for intelligent channel prediction in 6G vehicular communications.
Authors: Anderson Augusma (UGA, LIG, M-PSI), Dominique Vaufreydaz (LIG, M-PSI), F\'ed\'erique Letu\'e (SVH)
Abstract: Group Emotion Recognition (GER) aims to infer collective affect in social environments such as classrooms, crowds, and public events. Many existing approaches rely on explicit individual-level processing, including cropped faces, person tracking, or per-person feature extraction, which makes the analysis pipeline person-centric and raises privacy concerns in deployment scenarios where only group-level understanding is needed. This research proposes VE-MD, a Variational Encoder-Multi-Decoder framework for group emotion recognition under a privacy-aware functional design. Rather than providing formal anonymization or cryptographic privacy guarantees, VE-MD is designed to avoid explicit individual monitoring by constraining the model to predict only aggregate group-level affect, without identity recognition or per-person emotion outputs. VE-MD learns a shared latent representation jointly optimized for emotion classification and internal prediction of body and facial structural representations. Two structural decoding strategies are investigated: a transformer-based PersonQuery decoder and a dense Heatmap decoder that naturally accommodates variable group sizes. Experiments on six in-the-wild datasets, including two GER and four Individual Emotion Recognition (IER) benchmarks, show that structural supervision consistently improves representation learning. More importantly, the results reveal a clear distinction between GER and IER: optimizing the latent space alone is often insufficient for GER because it tends to attenuate interaction-related cues, whereas preserving explicit structural outputs improves collective affect inference. In contrast, projected structural representations seem to act as an effective denoising bottleneck for IER. VE-MD achieves state-of-the-art performance on GAF-3.0 (up to 90.06%) and VGAF (82.25% with multimodal fusion with audio). These results show that preserving interaction-related structural information is particularly beneficial for group-level affect modeling without relying on prior individual feature extraction. On IER datasets using multimodal fusion with audio modality, VE-MD outperforms SOTA on SamSemo (77.9%, adding text modality) while achieving competitive performances on MER-MULTI (63.8%), DFEW (70.7%) and EngageNet (69.0).
Authors: Scott Piersall, Yang Gao, Shenyang Liu, Liqiang Wang
Abstract: Message Passing Interface (MPI) is a foundational technology in high-performance computing (HPC), widely used for large-scale simulations and distributed training (e.g., in machine learning frameworks such as PyTorch and TensorFlow). However, maintaining MPI programs remains challenging due to their complex interplay among processes and the intricacies of message passing and synchronization. With the advancement of large language models like ChatGPT, it is tempting to adopt such technology for automated error detection and repair. Yet, our studies reveal that directly applying large language models (LLMs) yields suboptimal results, largely because these models lack essential knowledge about correct and incorrect usage, particularly the bugs found in MPI programs. In this paper, we design a bug detection and repair technique alongside Few-Shot Learning (FSL), Chain-of-Thought (CoT) reasoning, and Retrieval Augmented Generation (RAG) techniques in LLMs to enhance the large language model's ability to detect and repair errors. Surprisingly, such enhancements lead to a significant improvement, from 44% to 77%, in error detection accuracy compared to baseline methods that use ChatGPT directly. Additionally, our experiments demonstrate our bug referencing technique generalizes well to other large language models.
Authors: Kaiwen Zhang, Guanjun Liu
Abstract: Safe Rust guarantees memory safety through strict compile-time constraints: ownership can be transferred, borrowing can temporarily guarantee either shared read-only or exclusive write access, and ownership and borrowing are scoped by lifetime. Automatically synthesizing correct and safe Rust code is challenging, as the generated code must not only satisfy ownership, borrowing, and lifetime constraints, but also meet type and interface requirements at compile time. This work proposes a synthesis method based on our newly defined Pushdown Colored Petri Net (PCPN) that models these compilation constraints directly from public API signatures to synthesize valid call sequences. Token colors encode dynamic resource states together with a scope level indicating the lifetime region in which a borrow is valid. The pushdown stack tracks the entering or leaving of lifetime parameter via pushing and popping tokens. A transition is enabled only when type matching and interface obligations both hold and the required resource states are available. Based on the bisimulation theory, we prove that the enabling and firing rules of PCPN are consistent with the compile-time check of these three constraints. We develop an automatic synthesis tool based on PCPN and the experimental results show that the synthesized codes are all correct.
Authors: Yuchen Guo, Junli Gong, Hongmin Cai, Yiu-ming Cheung, Weifeng Su
Abstract: Video color grading is a critical post-production process that transforms flat, log-encoded raw footage into emotionally resonant cinematic visuals. Existing automated methods act as static, black-box executors that directly output edited pixels, lacking both interpretability and the iterative control required by professionals. We introduce LumiVideo, an agentic system that mimics the cognitive workflow of professional colorists through four stages: Perception, Reasoning, Execution, and Reflection. Given only raw log video, LumiVideo autonomously produces a cinematic base grade by analyzing the scene's physical lighting and semantic content. Its Reasoning engine synergizes an LLM's internalized cinematic knowledge with a Retrieval-Augmented Generation (RAG) framework via a Tree of Thoughts (ToT) search to navigate the non-linear color parameter space. Rather than generating pixels, the system compiles the deduced parameters into industry-standard ASC-CDL configurations and a globally consistent 3D LUT, analytically guaranteeing temporal consistency. An optional Reflection loop then allows creators to refine the result via natural language feedback. We further introduce LumiGrade, the first log-encoded video benchmark for evaluating automated grading. Experiments show that LumiVideo approaches human expert quality in fully automatic mode while enabling precise iterative control when directed.
Authors: Zachary Bogorad, Ibrahim Elsharkawy, Yonatan Kahn, Andrew J. Larkoski, Noam Levi
Abstract: Deep generative models such as diffusion and flow matching are powerful machine learning tools capable of learning and sampling from high-dimensional distributions. They are particularly useful when the training data appears to be concentrated on a submanifold of the data embedding space. For high-energy physics data, consisting of collections of relativistic energy-momentum 4-vectors, this submanifold can enforce extremely strong physically-motivated priors, such as energy and momentum conservation. If these constraints are learned only approximately, rather than exactly, this can inhibit the interpretability and reliability of such generative models. To remedy this deficiency, we introduce generative models which are, by construction, confined at every step of their sampling trajectory to the manifold of massless N-particle Lorentz-invariant phase space in the center-of-momentum frame. In the case of diffusion models, the "pure noise" forward process endpoint corresponds to the uniform distribution on phase space, which provides a clear starting point from which to identify how correlations among the particles emerge during the reverse (de-noising) process. We demonstrate that our models are able to learn both few-particle and many-particle distributions with various singularity structures, paving the way for future interpretability studies using generative models trained on simulated jet data.
Authors: Timothy Gould, Sidike Paheding
Abstract: In this work, we present a novel machine learning framework called Self-Directed Task Identification (SDTI), which enables models to autonomously identify the correct target variable for each dataset in a zero-shot setting without pre-training. SDTI is a minimal, interpretable framework demonstrating the feasibility of repurposing core machine learning concepts for a novel task structure. To our knowledge, no existing architectures have demonstrated this ability. Traditional approaches lack this capability, leaving data annotation as a time-consuming process that relies heavily on human effort. Using only standard neural network components, we show that SDTI can be achieved through appropriate problem formulation and architectural design. We evaluate the proposed framework on a range of benchmark tasks and demonstrate its effectiveness in reliably identifying the ground truth out of a set of potential target variables. SDTI outperformed baseline architectures by 14% in F1 score on synthetic task identification benchmarks. These proof-of-concept experiments highlight the future potential of SDTI to reduce dependence on manual annotation and to enhance the scalability of autonomous learning systems in real-world applications.
Authors: Phil\'emon Beghin, Anne-Emmanuelle Ceulemans, Fran\c{c}ois Glineur
Abstract: We explore the automatic detection of violin width reduction using 3D photogrammetric meshes. We compare SVM and Decision Trees applied to a geometry-based raw representation built from elevation maps with a more targeted, feature-engineered approach relying on parametric contour lines fitting. Although elevation maps occasionally achieve strong results, their performance does not surpass that of the contour-based inputs.
Authors: Kevin Song
Abstract: Multi-agent trajectory generation in team sports requires models that capture both the diversity of possible plays and realistic spatial coordination between players on plays. Standard generative approaches such as Conditional Variational Autoencoders (CVAE) and diffusion models struggle with this task, exhibiting posterior collapse or convergence to the dataset mean. Moreover, most trajectory prediction methods operate in a forecasting regime that requires multiple frames of observed history, limiting their use for play design where only the initial formation is available. We present PlayGen-MoG, an extensible framework for formation-conditioned play generation that addresses these challenges through three design choices: 1/ a Mixture-of-Gaussians (MoG) output head with shared mixture weights across all agents, where a single set of weights selects a play scenario that couples all players' trajectories, 2/ relative spatial attention that encodes pairwise player positions and distances as learned attention biases, and 3/ non-autoregressive prediction of absolute displacements from the initial formation, eliminating cumulative error drift and removing the dependence on observed trajectory history, enabling realistic play generation from a single static formation alone. On American football tracking data, PlayGen-MoG achieves 1.68 yard ADE and 3.98 yard FDE while maintaining full utilization of all 8 mixture components with entropy of 2.06 out of 2.08, and qualitatively confirming diverse generation without mode collapse.
Authors: Shramana Dey, Zahir Khan, T. A. PramodKumar, B. Uma Shankar, Ashis K. Dhara, Ramachandran Rajalakshmi, Rajiv Raman, Sushmita Mitra
Abstract: Diabetic Retinopathy (DR) is a serious microvascular complication of diabetes, and one of the leading causes of vision loss worldwide. Although automated detection and grading, with Deep Learning (DL), can reduce the burden on ophthalmologists, it is constrained by the limited availability of high-quality datasets. Existing repositories often remain geographically narrow, contain limited samples, and exhibit inconsistent annotations or variable image quality; thereby, restricting their clinical reliability. This paper presents a comprehensive review and comparative analysis of fundus image datasets used in the management of DR. The study evaluates their usability across key tasks, including binary classification, severity grading, lesion localization, and multi-disease screening. It also categorizes the datasets by size, accessibility, and annotation type (such as image-level, lesion-level, and multi-disease). Finally, a recently published dataset is presented as a case study to illustrate broader challenges in dataset curation and usage. The review consolidates current knowledge while highlighting persistent gaps such as the lack of standardized lesion-level annotations and longitudinal data. It also outlines recommendations for future dataset development to support clinically reliable and explainable solutions in DR screening.
Authors: Aaditya Naik, Guruprerana Shabadi, Rajeev Alur, Mayur Naik
Abstract: Advances in training, post-training, and inference-time methods have enabled frontier reasoning models to win gold medals in math competitions and settle challenging open problems. Gaining trust in the responses of these models requires that natural language proofs be checked for errors. LLM judges are increasingly being adopted to meet the growing demand for evaluating such proofs. While verification is considered easier than generation, what model capability does reliable verification actually require? We systematically evaluate four open-source and two frontier LLMs on datasets of human-graded natural language proofs of competition-level problems. We consider two key metrics: verifier accuracy and self-consistency (the rate of agreement across repeated judgments on the same proof). We observe that smaller open-source models are only up to ~10% behind frontier models in accuracy but they are up to ~25% more inconsistent. Furthermore, we see that verifier accuracy is sensitive to prompt choice across all models. We then demonstrate that the smaller models, in fact, do possess the mathematical capabilities to verify proofs at the level of frontier models, but they struggle to reliably elicit these capabilities with general judging prompts. Through an LLM-guided prompt search, we synthesize an ensemble of specialized prompts that overcome the specific failure modes of smaller models, boosting their performance by up to 9.1% in accuracy and 15.9% in self-consistency. These gains are realized across models and datasets, allowing models like Qwen3.5-35B to perform on par with frontier models such as Gemini 3.1 Pro for proof verification.
Authors: Nishit Asnani, Rohan Badlani
Abstract: Modeling coherence in text has been a task that has excited NLP researchers since a long time. It has applications in detecting incoherent structures and helping the author fix them. There has been recent work in using neural networks to extract a skeleton from one sentence, and then use that skeleton to generate the next sentence for coherent narrative story generation. In this project, we aim to study if the consistency of skeletons across subsequent sentences is a good metric to characterize the coherence of a given body of text. We propose a new Sentence/Skeleton Similarity Network (SSN) for modeling coherence across pairs of sentences, and show that this network performs much better than baseline similarity techniques like cosine similarity and Euclidean distance. Although skeletons appear to be promising candidates for modeling coherence, our results show that sentence-level models outperform those on skeletons for evaluating textual coherence, thus indicating that the current state-of-the-art coherence modeling techniques are going in the right direction by dealing with sentences rather than their sub-parts.
Authors: Zonghan Li, Feng Ji
Abstract: Behavioral simulation is increasingly used to anticipate responses to interventions. Large language models (LLMs) enable researchers to specify population characteristics and intervention context in natural language, but it remains unclear to what extent LLMs can use these inputs to infer intervention effects. We evaluated three LLMs on 11 climate-psychology interventions using a dataset of 59,508 participants from 62 countries, and replicated the main analysis in two additional datasets (12 and 27 countries). LLMs reproduced observed patterns in attitudinal outcomes (e.g., climate beliefs and policy support) reasonably well, and prompting refinements improved this descriptive fit. However, descriptive fit did not reliably translate into causal fidelity (i.e., accurate estimates of intervention effects), and these two dimensions of accuracy followed different error structures. This descriptive-causal divergence held across the three datasets, but varied across intervention logics, with larger errors for interventions that depended on evoking internal experience than on directly conveying reasons or social cues. It was more pronounced for behavioral outcomes, where LLMs imposed stronger attitude-behavior coupling than in human data. Countries and population groups appearing well captured descriptively were not necessarily those with lower causal errors. Relying on descriptive fit alone may therefore create unwarranted confidence in simulation results, misleading conclusions about intervention effects and masking population disparities that matter for fairness.
Authors: Jun-Sik Yoo
Abstract: We study the geometric structure of layer updates in deep language models. Rather than analyzing what information is encoded in intermediate representations, we ask how representations change from one layer to the next. We show that layerwise updates admit a decomposition into a dominant tokenwise component and a residual that is not captured by restricted tokenwise function classes. Across multiple architectures, including Transformers and state-space models, we find that the full layer update is almost perfectly aligned with the tokenwise component, while the residual exhibits substantially weaker alignment, larger angular deviation, and significantly lower projection onto the dominant tokenwise subspace. This indicates that the residual is not merely a small correction, but a geometrically distinct component of the transformation. This geometric separation has functional consequences: approximation error under the restricted tokenwise model is strongly associated with output perturbation, with Spearman correlations often exceeding 0.7 and reaching up to 0.95 in larger models. Together, these results suggest that most layerwise updates behave like structured reparameterizations along a dominant direction, while functionally significant computation is concentrated in a geometrically distinct residual component. Our framework provides a simple, architecture-agnostic method for probing the geometric and functional structure of layer updates in modern language models.
Authors: Mengtian Li, Yuwei Lu, Feifei Li, Chenqi Gan, Zhifeng Xie, Xi Wang
Abstract: Cinematic camera control relies on a tight feedback loop between director and cinematographer, where camera motion and framing are continuously reviewed and refined. Recent generative camera systems can produce diverse, text-conditioned trajectories, but they lack this "director in the loop" and have no explicit supervision of whether a shot is visually desirable. This results in in-distribution camera motion but poor framing, off-screen characters, and undesirable visual aesthetics. In this paper, we introduce VERTIGO, the first framework for visual preference optimization of camera trajectory generators. Our framework leverages a real-time graphics engine (Unity) to render 2D visual previews from generated camera motion. A cinematically fine-tuned vision-language model then scores these previews using our proposed cyclic semantic similarity mechanism, which aligns renders with text prompts. This process provides the visual preference signals for Direct Preference Optimization (DPO) post-training. Both quantitative evaluations and user studies on Unity renders and diffusion-based Camera-to-Video pipelines show consistent gains in condition adherence, framing quality, and perceptual realism. Notably, VERTIGO reduces the character off-screen rate from 38% to nearly 0% while preserving the geometric fidelity of camera motion. User study participants further prefer VERTIGO over baselines across composition, consistency, prompt adherence, and aesthetic quality, confirming the perceptual benefits of our visual preference post-training.
Authors: Haodong Xie, Yujun Cai, Rahul Singh Maharjan, Yiwei Wang, Federico Tavella, Angelo Cangelosi
Abstract: Concept Bottleneck Models (CBMs) introduce interpretability to black-box deep learning models by predicting labels through human-understandable concepts. However, unlike humans, who identify objects at different levels of abstraction using both general and specific features, existing CBMs operate at a single semantic level in both concept and label space. We propose HIL-CBM, a Hierarchical Interpretable Label-Free Concept Bottleneck Model that extends CBMs into a hierarchical framework to enhance interpretability by more closely mirroring the human cognitive process. HIL-CBM enables classification and explanation across multiple semantic levels without requiring relational concept annotations. HIL-CBM aligns the abstraction level of concept-based explanations with that of model predictions, progressing from abstract to concrete. This is achieved by (i) introducing a gradient-based visual consistency loss that encourages abstraction layers to focus on similar spatial regions, and (ii) training dual classification heads, each operating on feature concepts at different abstraction levels. Experiments on benchmark datasets demonstrate that HIL-CBM outperforms state-of-the-art sparse CBMs in classification accuracy. Human evaluations further show that HIL-CBM provides more interpretable and accurate explanations, while maintaining a hierarchical and label-free approach to feature concepts.
Authors: Valeria Martin, K. Brent Venable, Derek Morgan
Abstract: The scarcity of labeled satellite imagery remains a fundamental bottleneck for deep-learning (DL)-based wildfire monitoring systems. This paper investigates whether a diffusion-based foundation model for Earth Observation (EO), EarthSynth, can synthesize realistic post-wildfire Sentinel-2 RGB imagery conditioned on existing burn masks, without task-specific retraining. Using burn masks derived from the CalFireSeg-50 dataset (Martin et al., 2025), we design and evaluate six controlled experimental configurations that systematically vary: (i) pipeline architecture (mask-only full generation vs. inpainting with pre-fire context), (ii) prompt engineering strategy (three hand-crafted prompts and a VLM-generated prompt via Qwen2-VL), and (iii) a region-wise color-matching post-processing step. Quantitative assessment on 10 stratified test samples uses four complementary metrics: Burn IoU, burn-region color distance ({\Delta}C_burn), Darkness Contrast, and Spectral Plausibility. Results show that inpainting-based pipelines consistently outperform full-tile generation across all metrics, with the structured inpainting prompt achieving the best spatial alignment (Burn IoU = 0.456) and burn saliency (Darkness Contrast = 20.44), while color matching produces the lowest color distance ({\Delta}C_burn = 63.22) at the cost of reduced burn saliency. VLM-assisted inpainting is competitive with hand-crafted prompts. These findings provide a foundation for incorporating generative data augmentation into wildfire detection pipelines. Code and experiments are available at: https://www.kaggle.com/code/valeriamartinh/genai-all-runned
URLs: https://www.kaggle.com/code/valeriamartinh/genai-all-runned
Authors: Kiran Yalamanchi, Shivam Barwey, Ibrahim Jarrah, Pinaki Pal
Abstract: Computational fluid dynamics (CFD) simulations of complex fluid flows in energy systems are prohibitively expensive due to strong nonlinearities and multiscale-multiphysics interactions. In this work, we present a transformer-based modeling framework for prediction of fluid flows, and demonstrate it for high-pressure gas injection phenomena relevant to reciprocating engines. The approach employs a hierarchical Vision Transformer (SwinV2-UNet) architecture that processes multimodal flow datasets from multi-fidelity simulations. The model architecture is conditioned on auxiliary tokens explicitly encoding the data modality and time increment. Model performance is assessed on two different tasks: (1) spatiotemporal rollouts, where the model autoregressively predicts the flow state at future times; and (2) feature transformation, where the model infers unobserved fields/views from observed fields/views. We train separate models on multimodal datasets generated from in-house CFD simulations of argon jet injection into a nitrogen environment, encompassing multiple grid resolutions, turbulence models, and equations of state. The resulting data-driven models learn to generalize across resolutions and modalities, accurately forecasting the flow evolution and reconstructing missing flow-field information from limited views. This work demonstrates how large vision transformer-based models can be adapted to advance predictive modeling of complex fluid flow systems.
Authors: Samita Bai, Hamed Jelodar, Tochukwu Emmanuel Nwankwo, Parisa Hamedi, Mohammad Meymani, Roozbeh Razavi-Far, Ali A. Ghorbani
Abstract: Malware family classification remains a challenging task in automated malware analysis, particularly in real-world settings characterized by obfuscation, packing, and rapidly evolving threats. Existing machine learning and deep learning approaches typically depend on labeled datasets, handcrafted features, supervised training, or dynamic analysis, which limits their scalability and effectiveness in open-world scenarios. This paper presents a zero-label malware family classification framework based on a weighted hierarchical ensemble of pretrained large language models (LLMs). Rather than relying on feature-level learning or model retraining, the proposed approach aggregates decision-level predictions from multiple LLMs with complementary reasoning strengths. Model outputs are weighted using empirically derived macro-F1 scores and organized hierarchically, first resolving coarse-grained malicious behavior before assigning fine-grained malware families. This structure enhances robustness, reduces individual model instability, and aligns with analyst-style reasoning.
Authors: Joong Ho Choi, Jiayang Zhao, Avani Appalla, Himansh Mukesh, Dhwanil Vasani, Boyi Qian
Abstract: Deploying large multimodal language models at scale is constrained by token-based inference costs, yet the cost-performance behavior of visual prompting strategies remains poorly characterized. We introduce Image Prompt Packaging (IPPg), a prompting paradigm that embeds structured text directly into images to reduce text token overhead, and benchmark it across five datasets, three frontier models (GPT-4.1, GPT-4o, Claude 3.5 Sonnet), and two task families (VQA and code generation). We derive a cost formulation decomposing savings by token type and show IPPg achieves 35.8--91.0\% inference cost reductions. Despite token compression of up to 96\%, accuracy remains competitive in many settings, though outcomes are highly model- and task-dependent: GPT-4.1 achieves simultaneous accuracy and cost gains on CoSQL, while Claude 3.5 incurs cost increases on several VQA benchmarks. Systematic error analysis yields a failure-mode taxonomy: spatial reasoning, non-English inputs, and character-sensitive operations are most vulnerable, while schema-structured tasks benefit most. A 125-configuration rendering ablation reveals accuracy shifts of 10--30 percentage points, establishing visual encoding choices as a first-class variable in multimodal system design.
Authors: Md. Sajeebul Islam Sk., Md. Mehedi Hasan Shawon, Md. Golam Rabiul Alam
Abstract: Lumbar Spinal Stenosis (LSS) diagnosis remains a critical clinical challenge, with diagnosis heavily dependent on labor-intensive manual interpretation of multi-view Magnetic Resonance Imaging (MRI), leading to substantial inter-observer variability and diagnostic delays. Existing vision-language models simultaneously fail to address the extreme class imbalance prevalent in clinical segmentation datasets while preserving spatial accuracy, primarily due to global pooling mechanisms that discard crucial anatomical hierarchies. We present an end-to-end Explainable Vision-Language Model framework designed to overcome these limitations, achieved through two principal objectives. We propose a Spatial Patch Cross-Attention module that enables precise, text-directed localization of spinal anomalies with spatial precision. A novel Adaptive PID-Tversky Loss function by integrating control theory principles dynamically further modifies training penalties to specifically address difficult, under-segmented minority instances. By incorporating foundational VLMs alongside an Automated Radiology Report Generation module, our framework demonstrates considerable performance: a diagnostic classification accuracy of 90.69%, a macro-averaged Dice score of 0.9512 for segmentation, and a CIDEr score of 92.80%. Furthermore, the framework shows explainability by converting complex segmentation predictions into radiologist-style clinical reports, thereby establishing a new benchmark for transparent, interpretable AI in clinical medical imaging that keeps essential human supervision while enhancing diagnostic capabilities.
Authors: Roland M\"uhlenbernd
Abstract: Large language models (LLMs) increasingly exhibit human-like patterns of pragmatic and social reasoning. This paper addresses two related questions: do LLMs approximate human social meaning not only qualitatively but also quantitatively, and can prompting strategies informed by pragmatic theory improve this approximation? To address the first, we introduce two calibration-focused metrics distinguishing structural fidelity from magnitude calibration: the Effect Size Ratio (ESR) and the Calibration Deviation Score (CDS). To address the second, we derive prompting conditions from two pragmatic assumptions: that social meaning arises from reasoning over linguistic alternatives, and that listeners infer speaker knowledge states and communicative motives. Applied to a case study on numerical (im)precision across three frontier LLMs, we find that all models reliably reproduce the qualitative structure of human social inferences but differ substantially in magnitude calibration. Prompting models to reason about speaker knowledge and motives most consistently reduces magnitude deviation, while prompting for alternative-awareness tends to amplify exaggeration. Combining both components is the only intervention that improves all calibration-sensitive metrics across all models, though fine-grained magnitude calibration remains only partially resolved. LLMs thus capture inferential structure while variably distorting inferential strength, and pragmatic theory provides a useful but incomplete handle for improving that approximation.
Authors: Rushabha Balaji, Kuan-Lin Chen, Danijela Cabric, Bhaskar D. Rao
Abstract: Sparse Bayesian Learning is one of the most popular sparse signal recovery methods, and various algorithms exist under the SBL paradigm. However, given a performance metric and a sparse recovery problem, it is difficult to know a-priori the best algorithm to choose. This difficulty is in part due to a lack of a unified framework to derive SBL algorithms. We address this issue by first showing that the most popular SBL algorithms can be derived using the majorization-minimization (MM) principle, providing hitherto unknown convergence guarantees to this class of SBL methods. Moreover, we show that the two most popular SBL update rules not only fall under the MM framework but are both valid descent steps for a common majorizer, revealing a deeper analytical compatibility between these algorithms. Using this insight and properties from MM theory we expand the class of SBL algorithms, and address finding the best SBL algorithm via data within the MM framework. Second, we go beyond the MM framework by introducing the powerful modeling capabilities of deep learning to further expand the class of SBL algorithms, aiming to learn a superior SBL update rule from data. We propose a novel deep learning architecture that can outperform the classical MM based ones across different sparse recovery problems. Our architecture's complexity does not scale with the measurement matrix dimension, hence providing a unique opportunity to test generalization capability across different matrices. For parameterized dictionaries, this invariance allows us to train and test the model across different parameter ranges. We also showcase our model's ability to learn a functional mapping by its zero-shot performance on unseen measurement matrices. Finally, we test our model's performance across different numbers of snapshots, signal-to-noise ratios, and sparsity levels.
Authors: Darya Kaviani, Alp Eren Ozdarendeli, Jinhao Zhu, Yu Ding, Raluca Ada Popa
Abstract: Personal AI systems increasingly retain long-term memory of user activity, including documents, emails, messages, meetings, and ambient recordings. Trusted hardware can keep this data private, but struggles to scale with a growing datastore. This pushes the data to external storage, which exposes retrieval access patterns that leak private information to the application provider. Oblivious RAM (ORAM) is a cryptographic primitive that can hide these patterns, but it requires a fixed access budget, precluding the query-dependent traversals that agentic memory systems rely on for accuracy. We present Opal, a private memory system for personal AI. Our key insight is to decouple all data-dependent reasoning from the bulk of personal data, confining it to the trusted enclave. Untrusted disk then sees only fixed, oblivious memory accesses. This enclave-resident component uses a lightweight knowledge graph to capture personal context that semantic search alone misses and handles continuous ingestion by piggybacking reindexing and capacity management on every ORAM access. Evaluated on a comprehensive synthetic personal-data pipeline driven by stochastic communication models, Opal improves retrieval accuracy by 13 percentage points over semantic search and achieves 29x higher throughput with 15x lower infrastructure cost than a secure baseline. Opal is under consideration for deployment to millions of users at a major AI provider.
Authors: Adam Bayley, Xiaodan Zhu, Raquel Aoki, Yanshuai Cao, Kevin H. Wilson
Abstract: The recent advancement of Large Language Models (LLMs) offers new opportunities to generate user preference data to warm-start bandits. Recent studies on contextual bandits with LLM initialization (CBLI) have shown that these synthetic priors can significantly lower early regret. However, these findings assume that LLM-generated choices are reasonably aligned with actual user preferences. In this paper, we systematically examine how LLM-generated preferences perform when random and label-flipping noise is injected into the synthetic training data. For aligned domains, we find that warm-starting remains effective up to 30% corruption, loses its advantage around 40%, and degrades performance beyond 50%. When there is systematic misalignment, even without added noise, LLM-generated priors can lead to higher regret than a cold-start bandit. To explain these behaviors, we develop a theoretical analysis that decomposes the effect of random label noise and systematic misalignment on the prior error driving the bandit's regret, and derive a sufficient condition under which LLM-based warm starts are provably better than a cold-start bandit. We validate these results across multiple conjoint datasets and LLMs, showing that estimated alignment reliably tracks when warm-starting improves or degrades recommendation quality.
Authors: Kamalasankari Subramaniakuppusamy, Jugal Gajjar
Abstract: Post-hoc feature attribution methods are widely deployed in safety-critical vision systems, yet their stability under realistic input perturbations remains poorly characterized. Existing metrics evaluate explanations primarily under additive noise, collapse stability to a single scalar, and fail to condition on prediction preservation, conflating explanation fragility with model sensitivity. We introduce the Feature Attribution Stability Suite (FASS), a benchmark that enforces prediction-invariance filtering, decomposes stability into three complementary metrics: structural similarity, rank correlation, and top-k Jaccard overlap-and evaluates across geometric, photometric, and compression perturbations. Evaluating four attribution methods (Integrated Gradients, GradientSHAP, Grad-CAM, LIME) across four architectures and three datasets-ImageNet-1K, MS COCO, and CIFAR-10, FASS shows that stability estimates depend critically on perturbation family and prediction-invariance filtering. Geometric perturbations expose substantially greater attribution instability than photometric changes, and without conditioning on prediction preservation, up to 99% of evaluated pairs involve changed predictions. Under this controlled evaluation, we observe consistent method-level trends, with Grad-CAM achieving the highest stability across datasets.
Authors: Murtuza Shahzad, Joseph Wilson, Ibrahim Al Azher, Hamed Alhoori, Mona Rahimi
Abstract: The increasing complexity and volume of software systems have heightened the importance of identifying and mitigating security vulnerabilities. The existing software vulnerability datasets frequently fall short in providing comprehensive, detailed code snippets explicitly linked to specific vulnerability descriptions, reducing their utility for advanced research and hindering efforts to develop a deeper understanding of security vulnerabilities. To address this challenge, we present a novel dataset that provides examples of vulnerable code snippets corresponding to Common Attack Pattern Enumerations and Classifications (CAPEC) and Common Weakness Enumeration (CWE) descriptions. By employing the capabilities of Generative Pre-trained Transformer (GPT) models, we have developed a robust methodology for generating these examples. Our approach utilizes GPT-4o, Llama and Claude models to generate code snippets that exhibit specific vulnerabilities as described in CAPEC and CWE documentation. This dataset not only enhances the understanding of security vulnerabilities in code but also serves as a valuable resource for training machine learning models focused on automatic vulnerability detection and remediation. Preliminary evaluations suggest that the dataset generated by Large Language Models demonstrates high accuracy and can serve as a reliable reference for vulnerability identification systems. We found consistent results across the three models, with 0.98 cosine similarity among codes. The final dataset comprises 615 CAPEC code snippets in three programming languages: Java, Python, and JavaScript, making it one of the most extensive and diverse resources in this domain.
Authors: Lingjun Zhao, Dayeon Ki, Marine Carpuat, Hal Daum\'e III
Abstract: Language models are known to exhibit various forms of cultural bias in decision-making tasks, yet much less is known about their degree of cultural familiarity in open-ended text generation tasks. In this paper, we introduce the task of culturally-adapted art description generation, where models describe artworks for audiences from different cultural groups who vary in their familiarity with the cultural symbols and narratives embedded in the artwork. To evaluate cultural competence in this pragmatic generation task, we propose a framework based on culturally grounded question answering. We find that base models are only marginally adequate for this task, but, through a pragmatic speaker model, we can improve simulated listener comprehension by up to 8.2%. A human study further confirms that the model with higher pragmatic competence is rated as more helpful for comprehension by 8.0%.
Authors: Jackson G. Lu, Gerui Gloria Zhao, Anna Manyi Zheng
Abstract: Despite the growing use of generative artificial intelligence (GenAI) in entrepreneurship, research on its impact remains fragmented. To address this limitation, we provide an integrative review of how GenAI influences entrepreneurs at each stage of the entrepreneurial process: (1) opportunity recognition and ideation, (2) opportunity evaluation and commitment, (3) resource assembly and mobilization, and (4) venture launch and growth. Based on our review, we propose the Empowerment-Entrapment Framework to understand how GenAI can both empower and entrap entrepreneurs, highlighting GenAI's role as a double-edged sword at each stage of the entrepreneurial process. For example, GenAI may improve venture idea quality but introduce hallucinations and training data biases; boost entrepreneurial self-efficacy but heighten entrepreneurial overconfidence; increase functional breadth but decrease relational embeddedness; and boost productivity but fuel "workslop" and erode critical thinking, learning, and memory. Moreover, we identify core features of GenAI that underlie these empowering and entrapping effects. We also explore boundary conditions (e.g., entrepreneurs' metacognition, domain expertise, and entrepreneurial experience) that shape the magnitude of these effects. Beyond these theoretical contributions, our review and the Empowerment-Entrapment Framework offer practical implications for entrepreneurs seeking to use GenAI strategically throughout the entrepreneurial process while managing its risks.
Authors: John T. Halloran
Abstract: Safety alignment has become a critical step to ensure LLMs refuse harmful requests while providing helpful and harmless responses. However, despite the ubiquity of safety alignment for deployed frontier models, two separate lines of recent work--jailbreak-tuning (JT) and weight orthogonalization (WO)--have shown that safety guardrails may be largely disabled, resulting in LLMs which comply with harmful requests they would normally refuse. In spite of far-reaching safety implications, analysis has largely been limited to refusal rates of each unalignment method in isolation, leaving their relative effects on adversarial LLM capabilities unknown. To fill this gap, we study the impact of unaligning six popular LLMs of various sizes across a large number of malicious and benign tasks, using both JT and WO. Across the evaluated models, we show that while refusal degradation is split between the two methods, WO produces LLMs far more capable of aiding in malicious activity; in contrast to JT, the majority of WO unaligned models are far less prone to hallucinations, better retain their original natural-language performance, and are more effective at state-of-the-art adversarial and cyber attacks. To thus help mitigate the malicious risks of WO unalignment, we conclude by showing that supervised fine-tuning effectively limits the adversarial attack abilities enabled by WO, without drastically affecting hallucination rates or natural language performance.
Authors: Sahaj Singh Maini, Robert L. Goldstone, Zoran Tiganj
Abstract: Humans exhibit remarkable abilities to coordinate in groups. As large language models (LLMs) become more capable, it remains an open question whether they can demonstrate comparable adaptive coordination and whether they use the same strategies as humans. To investigate this, we compare LLM and human performance on a common-interest game with imperfect monitoring: Group Binary Search. In this n-player game, participants need to coordinate their actions to achieve a common objective. Players independently submit numerical values in an effort to collectively sum to a randomly assigned target number. Without direct communication, they rely on group feedback to iteratively adjust their submissions until they reach the target number. Our findings show that, unlike humans who adapt and stabilize their behavior over time, LLMs often fail to improve across games and exhibit excessive switching, which impairs group convergence. Moreover, richer feedback (e.g., numerical error magnitude) benefits humans substantially but has small effects on LLMs. Taken together, by grounding the analysis in human baselines and mechanism-level metrics, including reactivity scaling, switching dynamics, and learning across games, we point to differences in human and LLM groups and provide a behaviorally grounded diagnostic for closing the coordination gap.
Authors: Ethan Reid
Abstract: We present Moondream Segmentation, a referring image segmentation extension of Moondream 3, a vision-language model. Given an image and a referring expression, the model autoregressively decodes a vector path and iteratively refines the rasterized mask into a final detailed mask. We introduce a reinforcement learning stage that resolves ambiguity in the supervised signal by directly optimizing mask quality. Rollouts from this stage produce coarse-to-ground-truth targets for the refiner. To mitigate evaluation noise from polygon annotations, we release RefCOCO-M, a cleaned RefCOCO validation split with boundary-accurate masks. Moondream Segmentation achieves a cIoU of 80.2% on RefCOCO (val) and 62.6% mIoU on LVIS (val).
Authors: Hita Kambhamettu, Will Crichton, Sean Welleck, Harrison Goldstein, Andrew Head
Abstract: LLM-generated explanations can make technical content more accessible, but there is a ceiling on what they can support interactively. Because LLM outputs are static text, they cannot be executed or stepped through. We argue that grounding explanations in a formalized representation enables interactive affordances beyond what static text supports. We instantiate this idea for mathematical proof comprehension with explorable theorems, a system that uses LLMs to translate a theorem and its written proof into Lean, a programming language for machine-checked proofs, and links the written proof with the Lean code. Readers can work through the proof at a step-level granularity, test custom examples or counterexamples, and trace the logical dependencies bridging each step. Each worked-out step is produced by executing the Lean proof on that example and extracting its intermediate state. A user study ($n = 16$) shows potential advantages of this approach: in a proof-reading task, participants who had access to the provided explorability features gave better, more correct, and more detailed answers to comprehension questions, demonstrating a stronger overall understanding of the underlying mathematics.
Authors: Hita Kambhamettu, Bhavana Dalvi Mishra, Andrew Head, Jonathan Bragg, Aakanksha Naik, Joseph Chee Chang, Pao Siangliulue
Abstract: Developing a novel research idea is hard. It must be distinct enough from prior work to claim a contribution while also building on it. This requires iteratively reviewing literature and refining an idea based on what a researcher reads; yet when an idea changes, the literature that matters often changes with it. Most tools offer limited support for this interplay: literature tools help researchers understand a fixed body of work, while ideation tools evaluate ideas against a static, pre-curated set of papers. We introduce literature-initiated pivots, a mechanism where engagement with literature prompts revision to a developing idea, and where that revision changes which literature is relevant. We operationalize this in LitPivot, where researchers concurrently draft and vet an idea. LitPivot dynamically retrieves clusters of papers relevant to a selected part of the idea and proposes literature-informed critiques for how to revise it. A lab study ($n{=}17$) shows researchers produced higher-rated ideas with stronger self-reported understanding of the literature space; an open-ended study ($n{=}5$) reveals how researchers use LitPivot to iteratively evolve their own ideas.
Authors: Wei Zou, Mingwen Dong, Miguel Romero Calvo, Wei Zou, Shuaichen Chang, Jiang Guo, Dongkyu Lee, Xing Niu, Xiaofei Ma, Yanjun Qi, Jiarong Jiang
Abstract: Memory makes LLM-based web agents personalized, powerful, yet exploitable. By storing past interactions to personalize future tasks, agents inadvertently create a persistent attack surface that spans websites and sessions. While existing security research on memory assumes attackers can directly inject into memory storage or exploit shared memory across users, we present a more realistic threat model: contamination through environmental observation alone. We introduce Environment-injected Trajectory-based Agent Memory Poisoning (eTAMP), the first attack to achieve cross-session, cross-site compromise without requiring direct memory access. A single contaminated observation (e.g., viewing a manipulated product page) silently poisons an agent's memory and activates during future tasks on different websites, bypassing permission-based defenses. Our experiments on (Visual)WebArena reveal two key findings. First, eTAMP achieves substantial attack success rates: up to 32.5% on GPT-5-mini, 23.4% on GPT-5.2, and 19.5% on GPT-OSS-120B. Second, we discover Frustration Exploitation: agents under environmental stress become dramatically more susceptible, with ASR increasing up to 8 times when agents struggle with dropped clicks or garbled text. Notably, more capable models are not more secure. GPT-5.2 shows substantial vulnerability despite superior task performance. With the rise of AI browsers like OpenClaw, ChatGPT Atlas, and Perplexity Comet, our findings underscore the urgent need for defenses against environment-injected memory poisoning.
Authors: Hao Li, Liwei Zou, Wenping Yin, Gulsen Taskin, Naoto Yokoya, Danfeng Hong, Wufan Zhao
Abstract: Living in a changing climate, human society now faces more frequent and severe natural disasters than ever before. As a consequence, rapid disaster response during the "Golden 72 Hours" of search and rescue becomes a vital humanitarian necessity and community concern. However, traditional disaster damage surveys routinely fail to generalize across distinct urban morphologies and new disaster events. Effective damage mapping typically requires exhaustive and time-consuming manual data annotation. To address this issue, we introduce Smart Transfer, a novel Geospatial Artificial Intelligence (GeoAI) framework, leveraging state-of-the-art vision Foundation Models (FMs) for rapid building damage mapping with post-earthquake Very High Resolution (VHR) imagery. Specifically, we design two novel model transfer strategies: first, Pixel-wise Clustering (PC), ensuring robust prototype-level global feature alignment; second, a Distance-Penalized Triplet (DPT), integrating patch-level spatial autocorrelation patterns by assigning stronger penalties to semantically inconsistent yet spatially adjacent patches. Extensive experiments and ablations from the recent 2023 Turkiye-Syria earthquake show promising performance in multiple cross-region transfer settings, namely Leave One Domain Out (LODO) and Specific Source Domain Combination (SSDC). Moreover, Smart Transfer provides a scalable, automated GeoAI solution to accelerate building damage mapping and support rapid disaster response, offering new opportunities to enhance disaster resilience in climate-vulnerable regions and communities. The data and code are publicly available at https://github.com/ai4city-hkust/SmartTransfer.
Authors: Aayushi Dangol, Meghna Gupta, Daeun Yoo, Robert Wolfe, Jason Yip, Franziska Roesner, Julie A. Kientz
Abstract: Generative AI (genAI) is increasingly being integrated into children's everyday lives, not only through screens but also through so-called "screen-free" AI toys. These toys can simulate emotions, personalize responses, and recall prior interactions, creating the illusion of an ongoing social connection. Such capabilities raise important questions about how children understand boundaries, agency, and relationships when interacting with AI toys. To investigate this, we conducted two participatory design sessions with eight children ages 6-11 where they engaged with three different AI toys, shifting between play, experimentation, and reflection. Our findings reveal that children approached AI toys with genuine curiosity, profiling them as social beings. However, frequent interaction breakdowns and mismatches between apparent intelligence and toy-like form disrupted expectations around play and led to adversarial play. We conclude with implications and design provocations to navigate children's encounters with AI toys in more transparent, developmentally appropriate, and responsible ways.
Authors: Lei Song, Shihan Guan, Youyong Kong
Abstract: Non-Exemplar Continual Graph Learning (NECGL) seeks to eliminate the privacy risks intrinsic to rehearsal-based paradigms by retaining solely class-level prototype representations rather than raw graph examples for mitigating catastrophic forgetting. However, this design choice inevitably precipitates feature drift. As a nascent alternative, Analytic Continual Learning (ACL) capitalizes on the intrinsic generalization properties of frozen pre-trained models to bolster continual learning performance. Nonetheless, a key drawback resides in the pronounced attenuation of model plasticity. To surmount these challenges, we propose Analytic Drift Resister (ADR), a novel and theoretically grounded NECGL framework. ADR exploits iterative backpropagation to break free from the frozen pre-trained constraint, adapting to evolving task graph distributions and fortifying model plasticity. Since parameter updates trigger feature drift, we further propose Hierarchical Analytic Merging (HAM), performing layer-wise merging of linear transformations in Graph Neural Networks (GNNs) via ridge regression, thereby ensuring absolute resistance to feature drift. On this basis, Analytic Classifier Reconstruction (ACR) enables theoretically zero-forgetting class-incremental learning. Empirical evaluation on four node classification benchmarks demonstrates that ADR maintains strong competitiveness against existing state-of-the-art methods.
Authors: Weimin Liu, Jiyuan Qiu, Wenjun Wang, Joshua H. Meng
Abstract: Surround depth estimation provides a cost-effective alternative to LiDAR for 3D perception in autonomous driving. While recent self-supervised methods explore multi-camera settings to improve scale awareness and scene coverage, they are primarily designed for passenger vehicles and rarely consider articulated vehicles or robotics platforms. The articulated structure introduces complex cross-segment geometry and motion coupling, making consistent depth reasoning across views more challenging. In this work, we propose \textbf{ArticuSurDepth}, a self-supervised framework for surround-view depth estimation on articulated vehicles that enhances depth learning through cross-view and cross-vehicle geometric consistency guided by structural priors from vision foundation model. Specifically, we introduce multi-view spatial context enrichment strategy and a cross-view surface normal constraint to improve structural coherence across spatial and temporal contexts. We further incorporate camera height regularization with ground plane-awareness to encourage metric depth estimation, together with cross-vehicle pose consistency that bridges motion estimation between articulated segments. To validate our proposed method, an articulated vehicle experiment platform was established with a dataset collected over it. Experiment results demonstrate state-of-the-art (SoTA) performance of depth estimation on our self-collected dataset as well as on DDAD, nuScenes, and KITTI benchmarks.
Authors: Nathan Schneider, Antonios Anastasopoulos
Abstract: This work aims to shine a spotlight on the topic of metalanguage. We first define metalanguage, link it to NLP and LLMs, and then discuss our two labs' metalanguage-centered efforts. Finally, we discuss four dimensions of metalanguage and metalinguistic tasks, offering a list of understudied future research directions.
Authors: Shufan Jiang, Chios Chen, Zhiyang Chen
Abstract: The autonomous discovery of bugs remains a significant challenge in modern software development. Compared to code generation, the complexity of dynamic runtime environments makes bug discovery considerably harder for large language models (LLMs). In this paper, we take game development as a representative domain and introduce the Game Benchmark for Quality Assurance (GBQA), a benchmark containing 30 games and 124 human-verified bugs across three difficulty levels, to evaluate whether LLMs can autonomously detect software bugs. The benchmark is constructed using a multi-agent system that develops games and injects bugs in a scalable manner, with human experts in the loop to ensure correctness. Moreover, we provide a baseline interactive agent equipped with a multi-round ReAct loop and a memory mechanism, enabling long-horizon exploration of game environments for bug detection across different LLMs. Extensive experiments on frontier LLMs demonstrate that autonomous bug discovery remains highly challenging: the best-performing model, Claude-4.6-Opus in thinking mode, identifies only 48.39% of the verified bugs. We believe GBQA provides an adequate testbed and evaluation criterion, and that further progress on it will help close the gap in autonomous software engineering.
Authors: Cunyang Wei, Siddharth Singh, Aishwarya Sarkar, Daniel Nichols, Tisha Patel, Aditya K. Ranjan, Sayan Ghosh, Ali Jannesari, Nathan R. Tallent, Abhinav Bhatele
Abstract: Graph neural networks (GNNs) are widely used for learning on graph datasets derived from various real-world scenarios. Learning from extremely large graphs requires distributed training, and mini-batching with sampling is a popular approach for parallelizing GNN training. Existing distributed mini-batch approaches have significant performance bottlenecks due to expensive sampling methods and limited scaling when using data parallelism. In this work, we present ScaleGNN, a 4D parallel framework for scalable mini-batch GNN training that combines communication-free distributed sampling, 3D parallel matrix multiplication (PMM), and data parallelism. ScaleGNN introduces a uniform vertex sampling algorithm, enabling each process (GPU device) to construct its local mini-batch, i.e., subgraph partitions without any inter-process communication. 3D PMM enables scaling mini-batch training to much larger GPU counts than vanilla data parallelism with significantly lower communication overheads. We also present additional optimizations to overlap sampling with training, reduce communication overhead by sending data in lower precision, kernel fusion, and communication-computation overlap. We evaluate ScaleGNN on five graph datasets and demonstrate strong scaling up to 2048 GPUs on Perlmutter, 2048 GCDs on Frontier, and 1024 GPUs on Tuolumne. On Perlmutter, ScaleGNN achieves 3.5x end-to-end training speedup over the SOTA baseline on ogbn-products.
Authors: Haruhi Shida, Koo Imai, Keigo Kansa
Abstract: The safety of large language models (LLMs) relies on alignment techniques such as reinforcement learning from human feedback (RLHF). However, recent theoretical analyses suggest that reinforcement learning-based training does not acquire new capabilities but merely redistributes the utilization probabilities of existing ones. In this study, we propose ``compound jailbreaks'' targeting OpenAI gpt-oss-20b, which exploit the generalization failures of alignment. This approach combines multiple attack techniques -- each individually defended against -- to saturate the instruction hierarchy maintenance process. Our evaluation shows that the attack success rate (ASR) increased from 14.3\% with individual methods to 71.4\% with the combined approach. These results provide empirical evidence for the hypothesis that safety training does not generalize as broadly as model capabilities, highlighting the need for multifaceted safety evaluations using compound attack scenarios.
Authors: Farhad Pourkamali-Anaraki
Abstract: The massive scale of pretrained models has made efficient compression essential for practical deployment. Low-rank decomposition based on the singular value decomposition (SVD) provides a principled approach for model reduction, but its exact computation is expensive for large weight matrices. Randomized alternatives such as randomized SVD (RSVD) improve efficiency, yet they can suffer from poor approximation quality when the singular value spectrum decays slowly, a regime commonly observed in modern pretrained models. In this work, we address this limitation from both theoretical and empirical perspectives. First, we establish a connection between low-rank approximation error and predictive performance by analyzing softmax perturbations, showing that deviations in class probabilities are controlled by the spectral error of the compressed weights. Second, we demonstrate that RSVD is inadequate, and we propose randomized subspace iteration (RSI) as a more effective alternative. By incorporating multiple power iterations, RSI improves spectral separation and provides a controllable mechanism for enhancing approximation quality. We evaluate our approach on both convolutional networks and transformer-based architectures. Our results show that RSI achieves near-optimal approximation quality while outperforming RSVD in predictive accuracy under aggressive compression, enabling efficient model compression.
Authors: Vira Kasprova, Amruta Parulekar, Abdulrahman AlRabah, Krishna Agaram, Ritwik Garg, Sagar Jha, Nimet Beyza Bozdag, Dilek Hakkani-Tur
Abstract: Large language models (LLMs) often exhibit sycophancy: agreement with user stance even when it conflicts with the model's opinion. While prior work has mostly studied this in single-agent settings, it remains underexplored in collaborative multi-agent systems. We ask whether awareness of other agents' sycophancy levels influences discussion outcomes. To investigate this, we run controlled experiments with six open-source LLMs, providing agents with peer sycophancy rankings that estimate each peer's tendency toward sycophancy. These rankings are based on scores calculated using various static (pre-discussion) and dynamic (online) strategies. We find that providing sycophancy priors reduces the influence of sycophancy-prone peers, mitigates error-cascades, and improves final discussion accuracy by an absolute 10.5%. Thus, this is a lightweight, effective way to reduce discussion sycophancy and improve downstream accuracy.
Authors: Kavana Venkatesh, Jiaming Cui
Abstract: Large Language Model (LLM) multi-agent systems are increasingly deployed as interacting agent societies, yet scaling these systems often yields diminishing or unstable returns, the causes of which remain poorly understood. We present the first large-scale empirical study of coordination dynamics in LLM-based multi-agent systems, introducing an atomic event-level formulation that reconstructs reasoning as cascades of coordination. Analyzing over 1.5 Million interactions across tasks, topologies, and scales, we uncover three coupled laws: coordination follows heavy-tailed cascades, concentrates via preferential attachment into intellectual elites, and produces increasingly frequent extreme events as system size grows. We show that these effects are coupled through a single structural mechanism: an integration bottleneck, in which coordination expansion scales with system size while consolidation does not, producing large but weakly integrated reasoning processes. To test this mechanism, we introduce Deficit-Triggered Integration (DTI), which selectively increases integration under imbalance. DTI improves performance precisely where coordination fails, without suppressing large-scale reasoning. Together, our results establish quantitative laws of collective cognition and identify coordination structure as a fundamental, previously unmeasured axis for understanding and improving scalable multi-agent intelligence.
Authors: Yao Zhao, Zhiyue Zhang, Yanxun Xu
Abstract: Clinical evidence synthesis requires identifying relevant trials from large registries and aggregating results that account for population differences. While recent LLM-based approaches have automated components of systematic review, they do not support end-to-end evidence synthesis. Moreover, conventional meta-analysis weights studies by statistical precision without considering clinical compatibility reflected in eligibility criteria. We propose EligMeta, an agentic framework that integrates automated trial discovery with eligibility-aware meta-analysis, translating natural-language queries into reproducible trial selection and incorporating eligibility alignment into study weighting to produce cohort-specific pooled estimates. EligMeta employs a hybrid architecture separating LLM-based reasoning from deterministic execution: LLMs generate interpretable rules from natural-language queries and perform schema-constrained parsing of trial metadata, while all logical operations, weight computations, and statistical pooling are executed deterministically to ensure reproducibility. The framework structures eligibility criteria and computes similarity-based study weights reflecting population alignment between target and comparator trials. In a gastric cancer landscape analysis, EligMeta reduced 4,044 candidate trials to 39 clinically relevant studies through rule-based filtering, recovering all 13 guideline-cited trials. In an olaparib adverse events meta-analysis across four trials, eligibility-aware weighting shifted the pooled risk ratio from 2.18 (95% CI: 1.71-2.79) under conventional Mantel-Haenszel estimation to 1.97 (95% CI: 1.76-2.20), demonstrating quantifiable impact of incorporating eligibility alignment. EligMeta bridges automated trial discovery with eligibility-aware meta-analysis, providing a scalable and reproducible framework for evidence synthesis in precision medicine.
Authors: Matthew Levinson
Abstract: Understanding the geometric structure of internal representations is a central goal of mechanistic interpretability. Prior work has shown that transformers trained on sequences generated by hidden Markov models encode probabilistic belief states as simplex-shaped geometries in their residual stream, with vertices corresponding to latent generative states. Whether large language models trained on naturalistic text develop analogous geometric representations remains an open question. We introduce a pipeline for discovering candidate simplex-structured subspaces in transformer representations, combining sparse autoencoders (SAEs), $k$-subspace clustering of SAE features, and simplex fitting using AANet. We validate the pipeline on a transformer trained on a multipartite hidden Markov model with known belief-state geometry. Applied to Gemma-2-9B, we identify 13 priority clusters exhibiting candidate simplex geometry ($K \geq 3$). A key challenge is distinguishing genuine belief-state encoding from tiling artifacts: latents can span a simplex-shaped subspace without the mixture coordinates carrying predictive signal beyond any individual feature. We therefore adopt barycentric prediction as our primary discriminating test. Among the 13 priority clusters, 3 exhibit a highly significant advantage on near-vertex samples (Wilcoxon $p < 10^{-14}$) and 4 on simplex-interior samples. Together 5 distinct real clusters pass at least one split, while no null cluster passes either. One cluster, 768_596, additionally achieves the highest causal steering score in the dataset. This is the only case where passive prediction and active intervention converge. We present these findings as preliminary evidence that genuine belief-like geometry exists in Gemma-2-9B's representation space, and identify the structured evaluation that would be required to confirm this interpretation.
Authors: Yuheng Zhang, Mingyue Huo, Minghao Zhu, Mengxue Zhang, Nan Jiang
Abstract: Reward models (RMs) are widely used as optimization targets in reinforcement learning from human feedback (RLHF), yet they remain vulnerable to reward hacking. Existing attacks mainly operate within the semantic space, constructing human-readable adversarial outputs that exploit RM biases. In this work, we introduce a fundamentally different paradigm: Token Mapping Perturbation Attack (TOMPA), a framework that performs adversarial optimization directly in token space. By bypassing the standard decode-re-tokenize interface between the policy and the reward model, TOMPA enables the attack policy to optimize over raw token sequences rather than coherent natural language. Using only black-box scalar feedback, TOMPA automatically discovers non-linguistic token patterns that elicit extremely high rewards across multiple state-of-the-art RMs. Specifically, when targeting Skywork-Reward-V2-Llama-3.1-8B, TOMPA nearly doubles the reward of GPT-5 reference answers and outperforms them on 98.0% of prompts. Despite these high scores, the generated outputs degenerate into nonsensical text, revealing that RMs can be systematically exploited beyond the semantic regime and exposing a critical vulnerability in current RLHF pipelines.
Authors: Yuhui Lin, Siyue Yu, Yuxing Yang, Guangliang Cheng, Jimin Xiao
Abstract: Recent advances in Multimodal Large Language Models (MLLMs) have expanded reasoning capabilities into 3D domains, enabling fine-grained spatial understanding. However, the substantial size of 3D MLLMs and the high dimensionality of input features introduce considerable inference overhead, which limits practical deployment on resource constrained platforms. To overcome this limitation, this paper presents Efficient3D, a unified framework for visual token pruning that accelerates 3D MLLMs while maintaining competitive accuracy. The proposed framework introduces a Debiased Visual Token Importance Estimator (DVTIE) module, which considers the influence of shallow initial layers during attention aggregation, thereby producing more reliable importance predictions for visual tokens. In addition, an Adaptive Token Rebalancing (ATR) strategy is developed to dynamically adjust pruning strength based on scene complexity, preserving semantic completeness and maintaining balanced attention across layers. Together, they enable context-aware token reduction that maintains essential semantics with lower computation. Comprehensive experiments conducted on five representative 3D vision and language benchmarks, including ScanRefer, Multi3DRefer, Scan2Cap, ScanQA, and SQA3D, demonstrate that Efficient3D achieves superior performance compared with unpruned baselines, with a +2.57% CIDEr improvement on the Scan2Cap dataset. Therefore, Efficient3D provides a scalable and effective solution for efficient inference in 3D MLLMs. The code is released at: https://github.com/sol924/Efficient3D
Authors: Fanwei Zeng, Changtao Miao, Jing Huang, Zhiya Tan, Shutao Gong, Xiaoming Yu, Yang Wang, Weibin Yao, Joey Tianyi Zhou, Jianshu Li, Yin Yan
Abstract: The rapid progress of generative AI has enabled increasingly realistic text-centric image forgeries, posing major challenges to document safety. Existing forensic methods mainly rely on visual cues and lack evidence-based reasoning to reveal subtle text manipulations. Detection, localization, and explanation are often treated as isolated tasks, limiting reliability and interpretability. To tackle these challenges, we propose DocShield, the first unified framework formulating text-centric forgery analysis as a visual-logical co-reasoning problem. At its core, a novel Cross-Cues-aware Chain of Thought (CCT) mechanism enables implicit agentic reasoning, iteratively cross-validating visual anomalies with textual semantics to produce consistent, evidence-grounded forensic analysis. We further introduce a Weighted Multi-Task Reward for GRPO-based optimization, aligning reasoning structure, spatial evidence, and authenticity prediction. Complementing the framework, we construct RealText-V1, a multilingual dataset of document-like text images with pixel-level manipulation masks and expert-level textual explanations. Extensive experiments show DocShield significantly outperforms existing methods, improving macro-average F1 by 41.4% over specialized frameworks and 23.4% over GPT-4o on T-IC13, with consistent gains on the challenging T-SROIE benchmark. Our dataset, model, and code will be publicly released.
Authors: Rodney Jehu-Appiah
Abstract: A previous study reported that E-Prime (English without the verb "to be") selectively altered reasoning in language models, with cross-model correlations suggesting a structural signature tied to which vocabulary was removed. I designed a replication with active controls to test the proposed mechanism: cognitive restructuring through specific vocabulary-cognition mappings. The experiment tested five conditions (unconstrained control, E-Prime, No-Have, elaborated metacognitive prompt, neutral filler-word ban) across six models and seven reasoning tasks (N=15,600 trials, 11,919 after compliance filtering). Every prediction from the cognitive restructuring hypothesis was disconfirmed. All four treatments outperformed the control (83.0%), including both active controls predicted to show null effects. The neutral filler-word ban, banning words like "very" and "just" with no role in logical inference, produced the largest improvement (+6.7 pp), while E-Prime produced the smallest (+3.7 pp). The four conditions ranked in perfect inverse order of theoretical depth. The cross-model correlation signature did not replicate (mean r=0.005). These results are consistent with a simpler mechanism: any constraint that forces a model off its default generation path acts as an output regularizer, improving reasoning by disrupting fluent but shallow response patterns. The shallowest constraints work best because they impose monitoring load with minimal conceptual disruption. I present these findings as a case study in discovery through disconfirmation.
Authors: Yihong Dong, Xiaoha Jian, Xue Jiang, Xuyuan Guo, Zhiyuan Fan, Jiaru Qian, Kechi Zhang, Jia Li, Zhi Jin, Ge Li
Abstract: The formal reasoning capabilities of LLMs are crucial for advancing automated software engineering. However, existing benchmarks for LLMs lack systematic evaluation based on computation and complexity, leaving a critical gap in understanding their formal reasoning capabilities. Therefore, it is still unknown whether SOTA LLMs can grasp the structured, hierarchical complexity of formal languages as defined by Computation Theory. To address this, we introduce ChomskyBench, a benchmark for systematically evaluating LLMs through the lens of Chomsky Hierarchy. Unlike prior work that uses vectorized classification for neural networks, ChomskyBench is the first to combine full Chomsky Hierarchy coverage, process-trace evaluation via natural language, and deterministic symbolic verifiability. ChomskyBench is composed of a comprehensive suite of language recognition and generation tasks designed to test capabilities at each level. Extensive experiments indicate a clear performance stratification that correlates with the hierarchy's levels of complexity. Our analysis reveals a direct relationship where increasing task difficulty substantially impacts both inference length and performance. Furthermore, we find that while larger models and advanced inference methods offer notable relative gains, they face severe efficiency barriers: achieving practical reliability would require prohibitive computational costs, revealing that current limitations stem from inefficiency rather than absolute capability bounds. A time complexity analysis further indicates that LLMs are significantly less efficient than traditional algorithmic programs for these formal tasks. These results delineate the practical limits of current LLMs, highlight the indispensability of traditional software tools, and provide insights to guide the development of future LLMs with more powerful formal reasoning capabilities.
Authors: Junwei You, Pei Li, Zhuoyu Jiang, Weizhe Tang, Zilin Huang, Rui Gan, Jiaxi Liu, Yan Zhao, Sikai Chen, Bin Ran
Abstract: Multimodal large language models (MLLMs) have shown strong potential for autonomous driving, yet existing benchmarks remain largely ego-centric and therefore cannot systematically assess model performance in infrastructure-centric and cooperative driving conditions. In this work, we introduce V2X-QA, a real-world dataset and benchmark for evaluating MLLMs across vehicle-side, infrastructure-side, and cooperative viewpoints. V2X-QA is built around a view-decoupled evaluation protocol that enables controlled comparison under vehicle-only, infrastructure-only, and cooperative driving conditions within a unified multiple-choice question answering (MCQA) framework. The benchmark is organized into a twelve-task taxonomy spanning perception, prediction, and reasoning and planning, and is constructed through expert-verified MCQA annotation to enable fine-grained diagnosis of viewpoint-dependent capabilities. Benchmark results across ten representative state-of-the-art proprietary and open-source models show that viewpoint accessibility substantially affects performance, and infrastructure-side reasoning supports meaningful macroscopic traffic understanding. Results also indicate that cooperative reasoning remains challenging since it requires cross-view alignment and evidence integration rather than simply additional visual input. To address these challenges, we introduce V2X-MoE, a benchmark-aligned baseline with explicit view routing and viewpoint-specific LoRA experts. The strong performance of V2X-MoE further suggests that explicit viewpoint specialization is a promising direction for multi-view reasoning in autonomous driving. Overall, V2X-QA provides a foundation for studying multi-perspective reasoning, reliability, and cooperative physical intelligence in connected autonomous driving. The dataset and V2X-MoE resources are publicly available at: https://github.com/junwei0001/V2X-QA.
Authors: Mirali Purohit, Bimal Gajera, Irish Mehta, Bhanu Tokas, Jacob Adler, Steven Lu, Scott Dickenshied, Serina Diniega, Brian Bue, Umaa Rebbapragada, Hannah Kerner
Abstract: We introduce MOMO, the first multi-sensor foundation model for Mars remote sensing. MOMO uses model merge to integrate representations learned independently from three key Martian sensors (HiRISE, CTX, and THEMIS), spanning resolutions from 0.25 m/pixel to 100 m/pixel. Central to our method is our novel Equal Validation Loss (EVL) strategy, which aligns checkpoints across sensors based on validation loss similarity before fusion via task arithmetic. This ensures models are merged at compatible convergence stages, leading to improved stability and generalization. We train MOMO on a large-scale, high-quality corpus of $\sim 12$ million samples curated from Mars orbital data and evaluate it on 9 downstream tasks from Mars-Bench. MOMO achieves better overall performance compared to ImageNet pre-trained, earth observation foundation model, sensor-specific pre-training, and fully-supervised baselines. Particularly on segmentation tasks, MOMO shows consistent and significant performance improvement. Our results demonstrate that model merging through an optimal checkpoint selection strategy provides an effective approach for building foundation models for multi-resolution data. The model weights, pretraining code, pretraining data, and evaluation code are available at: https://github.com/kerner-lab/MOMO.
Authors: Puyu Zeng, Zhaoxi Wang, Zhixu Duan, Liang Feng, Shaobo Wang, Cunxiang Wang, Jinghang Wang, Bing Zhao, Hu Wei, Linfeng Zhang
Abstract: Code generation and comprehension by Large Language Models (LLMs) have emerged as core drivers of industrial intelligence and decision optimization, finding widespread application in fields such as finance, automation, and aerospace. Although recent advancements have demonstrated the remarkable potential of LLMs in general code generation, existing benchmarks are mainly confined to single domains and languages. Consequently, they fail to effectively evaluate the generalization capabilities required for real-world industrial applications or to reflect the coding proficiency demanded by complex industrial scenarios. To bridge this gap, we introduce IndustryCode, the first comprehensive benchmark designed to span multiple industrial domains and programming languages. IndustryCode comprises 579 sub-problems derived from 125 primary industrial challenges, accompanied by rigorous problem descriptions and test cases. It covers a wide range of fields, including finance, automation, aerospace, and remote sensing-and incorporates diverse programming languages such as MATLAB, Python, C++, and Stata. In our evaluation, the top-performing model, Claude 4.5 Opus, achieved an overall accuracy of 68.1% on sub-problems and 42.5% main problems. The benchmark dataset and automated evaluation code will be made publicly available upon acceptance.
Authors: Pramod Bide, Sudhir Dhage, Mohammed Afaan Ansari, Rudresh Veerkhare
Abstract: Social media is widely used to share information globally and it also aids to gain attention from the world. When socially sensitive incidents like rape, human rights march, corruption, political controversy, chemical attacks occur, they gain immense attention from people all over the world, causing microblogging platforms like Twitter to get flooded with tweets related to such events. When an event evolves, many other events of a similar nature have happened in and around the same time frame. These are cross events because they are linked to the nature of the main event. Dissemination of information relating to such cross events helps in engaging the masses to share the varied views that emerge out of the similarities and differences between the events. Cross event detection is critical in determining the nature of events. Cross events have fulcrums points, i.e., topics around which the discussion is focused, as the event evolves which must be considered in topic evolution. We have proposed Cross Event Evolution Detection CEED framework which detects cross events that are similar with regards to their temporal nature resulting from main events. Event detection is based on the tweet segmentation using the Wikipedia title database and clustering segments based on a similarity measure. The cross event detection algorithm reveals events that overlap in both time and context to evaluate the effects of these cross events on deliberate negligent human actions. The topic evolution algorithm puts into perspective the change in topics for an events lifetime. The experimental results on a real Twitter data set demonstrate the effectiveness and precision of our proposed framework for both cross event detection and topic evolution algorithm during the evolution of cross events.
Authors: Giyeong Oh, Junghyun Lee, Jaehyun Park, Youngjae Yu, Wonho Bae, Junhyug Noh
Abstract: Modern LLMs inherit strong priors from web-scale pretraining, which can limit the headroom of post-training data-selection strategies. While Active Preference Learning (APL) seeks to optimize query efficiency in online Direct Preference Optimization (DPO), the inherent richness of on-policy candidate pools often renders simple Random sampling a surprisingly formidable baseline. We evaluate uncertainty-based APL against Random across harmlessness, helpfulness, and instruction-following settings, utilizing both reward models and LLM-as-a-judge proxies. We find that APL yields negligible improvements in proxy win-rates compared to Random. Crucially, we observe a dissociation where win-rate improves even as general capability -- measured by standard benchmarks -- degrades. APL fails to mitigate this capability collapse or reduce variance significantly better than random sampling. Our findings suggest that in the regime of strong pre-trained priors, the computational overhead of active selection is difficult to justify against the ``cheap diversity'' provided by simple random samples. Our code is available at https://github.com/BootsofLagrangian/random-vs-apl.
Authors: KrishnaSaiReddy Patil
Abstract: When Agent A delegates to Agent B, which invokes Tool C on behalf of User X, no existing framework can answer: whose authorization chain led to this action, and where did it violate policy? This paper introduces SentinelAgent, a formal framework for verifiable delegation chains in federal multi-agent AI systems. The Delegation Chain Calculus (DCC) defines seven properties - six deterministic (authority narrowing, policy preservation, forensic reconstructibility, cascade containment, scope-action conformance, output schema conformance) and one probabilistic (intent preservation) - with four meta-theorems and one proposition establishing the practical infeasibility of deterministic intent verification. The Intent-Preserving Delegation Protocol (IPDP) enforces all seven properties at runtime through a non-LLM Delegation Authority Service. A three-point verification lifecycle achieves 100% combined TPR at 0% FPR on DelegationBench v4 (516 scenarios, 10 attack categories, 13 federal domains). Under black-box adversarial conditions, the DAS blocks 30/30 attacks with 0 false positives. Deterministic properties are unbreakable under adversarial stress testing; intent verification degrades to 13% against sophisticated paraphrasing. Fine-tuning the NLI model on 190 government delegation examples improves P2 from 1.7% to 88.3% TPR (5-fold cross-validated, F1=82.1%). Properties P1, P3-P7 are mechanically verified via TLA+ model checking across 2.7 million states with zero violations. Even when intent verification is evaded, the remaining six properties constrain the adversary to permitted API calls, conformant outputs, traceable actions, bounded cascades, and compliant behavior.
Authors: Yongsu Ahn, Nam Wook Kim, Benjamin Bach
Abstract: AI chatbots are increasingly stepping into roles as collaborators or teachers in analyzing, visualizing, and reasoning through data and domain problem. Yet, AI's default assistant mode with its comprehensive and one-off responses may undermine opportunities for practitioners to develop literacy through their own thinking, inducing cognitive passivity. Drawing on evidence from empirical studies and theories, we argue that disrupting cognitive passivity necessitates a nuanced approach: rather than simply making AI promote deliberative thinking, there is a need for more dynamic and adaptive strategy through cognitive alignment -- a framework that characterizes effective human-AI interaction as a function of alignment between users' cognitive demand and AI's interaction mode. In the framework, we provide the mapping between AI's interaction mode (transmissive or deliberative) and users' cognitive demand (receptive or deliberative), otherwise leading to either cognitive passivity or friction. We further discuss implications and offer open questions for future research on data literacy.
Authors: Shreshth Saini, Hakan Gedik, Neil Birkbeck, Yilin Wang, Balu Adsumilli, Alan C. Bovik
Abstract: The rapid adoption of HDR-capable devices has created a pressing need to convert the 8-bit Standard Dynamic Range (SDR) content into perceptually and physically accurate 10-bit High Dynamic Range (HDR). Existing inverse tone-mapping (ITM) methods often rely on fixed tone-mapping operators that struggle to generalize to real-world degradations, stylistic variations, and camera pipelines, frequently producing clipped highlights, desaturated colors, or unstable tone reproduction. We introduce LumaFlux, a first physically and perceptually guided diffusion transformer (DiT) for SDR-to-HDR reconstruction by adapting a large pretrained DiT. Our LumaFlux introduces (1) a Physically-Guided Adaptation (PGA) module that injects luminance, spatial descriptors, and frequency cues into attention through low-rank residuals; (2) a Perceptual Cross-Modulation (PCM) layer that stabilizes chroma and texture via FiLM conditioning from vision encoder features; and (3) an HDR Residual Coupler that fuses physical and perceptual signals under a timestep- and layer-adaptive modulation schedule. Finally, a lightweight Rational-Quadratic Spline decoder reconstructs smooth, interpretable tone fields for highlight and exposure expansion, enhancing the output of the VAE decoder to generate HDR. To enable robust HDR learning, we curate the first large-scale SDR-HDR training corpus. For fair and reproducible comparison, we further establish a new evaluation benchmark, comprising HDR references and corresponding expert-graded SDR versions. Across benchmarks, LumaFlux outperforms state-of-the-art baselines, achieving superior luminance reconstruction and perceptual color fidelity with minimal additional parameters.
Authors: Tianze Xu, Yanzhao Zheng, Pengrui Lu, Lyumanshan Ye, Yong Wu, Zhentao Zhang, Yuanqiang Yu, Chao Ma, Jihuai Zhu, Pengfei Liu, Baohua Dong, Hangcheng Zhu, Ruohui Huang, Gang Yu
Abstract: Rubric-based Reinforcement Learning (RL) has emerged as a promising approach for aligning Large Language Models (LLMs) with complex, open-domain instruction following tasks. However, existing methods predominantly rely on response-level rewards, introducing severe reward sparsity and reward ambiguity problems. To address these issues, we propose Rubrics to Tokens (RTT), a novel rubric-based RL framework that bridges coarse response-level scores and fine-grained token-level credit assignment. RTT introduces a Token-Level Relevance Discriminator to predict which tokens in the response are responsible for a specific constraint, and optimizes the policy model via RTT-GRPO, which integrates response-level and token-level advantages within a unified framework. Furthermore, when transitioning from one-dimensional, outcome-level reward to three-dimensional reward space in the token-level rubric-based RL, we propose a novel group normalization method, called Intra-sample Token Group Normalization, to accommodate this shift. Extensive experiments and benchmarks demonstrate that RTT consistently outperforms other baselines in both instruction- and rubric-level accuracy across different models.
Authors: Dexiang Li, Zhenning Che, Haijun Zhang, Dongliang Zhou, Zhao Zhang, Yahong Han
Abstract: Pavement condition assessment is essential for road safety and maintenance. Existing research has made significant progress. However, most studies focus on conventional computer vision tasks such as classification, detection, and segmentation. In real-world applications, pavement inspection requires more than visual recognition. It also requires quantitative analysis, explanation, and interactive decision support. Current datasets are limited. They focus on unimodal perception. They lack support for multi-turn interaction and fact-grounded reasoning. They also do not connect perception with vision-language analysis. To address these limitations, we introduce PaveBench, a large-scale benchmark for pavement distress perception and interactive vision-language analysis on real-world highway inspection images. PaveBench supports four core tasks: classification, object detection, semantic segmentation, and vision-language question answering. It provides unified task definitions and evaluation protocols. On the visual side, PaveBench provides large-scale annotations and includes a curated hard-distractor subset for robustness evaluation. It contains a large collection of real-world pavement images. On the multimodal side, we introduce PaveVQA, a real-image question answering (QA) dataset that supports single-turn, multi-turn, and expert-corrected interactions. It covers recognition, localization, quantitative estimation, and maintenance reasoning. We evaluate several state-of-the-art methods and provide a detailed analysis. We also present a simple and effective agent-augmented visual question answering framework that integrates domain-specific models as tools alongside vision-language models. The dataset is available at: https://huggingface.co/datasets/MML-Group/PaveBench.
Authors: Lik Tung Fu, Jie Zhou, Shaokai Ren, Mengli Zhang, Jia Xiong, Hugo Jiang, Nan Guan, Xi Wang, Jun Yang
Abstract: Functional verification consumes over 50% of the IC development lifecycle, where SystemVerilog Assertions (SVAs) are indispensable for formal property verification and enhanced simulation-based debugging. However, manual SVA authoring is labor-intensive and error-prone. While Large Language Models (LLMs) show promise, their direct deployment is hindered by low functional accuracy and a severe scarcity of domain-specific data. To address these challenges, we introduce ChatSVA, an end-to-end SVA generation system built upon a multi-agent framework. At its core, the AgentBridge platform enables this multi-agent approach by systematically generating high-purity datasets, overcoming the data scarcity inherent to few-shot scenarios. Evaluated on 24 RTL designs, ChatSVA achieves 98.66% syntax and 96.12% functional pass rates, generating 139.5 SVAs per design with 82.50% function coverage. This represents a 33.3 percentage point improvement in functional correctness and an over 11x enhancement in function coverage compared to the previous state-of-the-art (SOTA). ChatSVA not only sets a new SOTA in automated SVA generation but also establishes a robust framework for solving long-chain reasoning problems in few-shot, domain-specific scenarios. An online service has been publicly released at https://www.nctieda.com/CHATDV.html.
Authors: Xinhao Wang, Zhonyu Xia, Zhiwei Lin, Zhe Li, Yongtao Wang
Abstract: Multimodal Large Language Models (MLLMs) have shown strong reasoning ability, but their high computational and memory costs hinder deployment in resource-constrained settings. While Post-Training Quantization (PTQ) and vision token pruning are standard compression techniques, they are usually treated as independent optimizations. In this paper, we show that these two techniques are strongly coupled: naively applying semantic-based token pruning to PTQ-optimized MLLMs can discard activation outliers that are important for numerical stability and thus worsen quantization errors in low-bit regimes (\textit{e.g.}, W4A4). To address this issue, we propose a quantization-aware vision token pruning framework. Our method introduces a lightweight hybrid sensitivity metric that combines simulated group-wise quantization error with outlier intensity. By combining this metric with standard semantic relevance scores, the method retains tokens that are both semantically informative and robust to quantization. Experiments on standard LLaVA architectures show that our method consistently outperforms naive integration baselines. At an aggressive pruning ratio that retains only 12.5\% of visual tokens, our framework improves accuracy by 2.24\% over the baseline and even surpasses dense quantization without pruning. To the best of our knowledge, this is the first method that explicitly co-optimizes vision token pruning and PTQ for accurate low-bit MLLM inference.
Authors: Hongbo Duan, Peiyu Zhuang, Yi Liu, Zhengyang Zhang, Yuxin Zhang, Pengting Luo, Fangming Liu, Xueqian Wang
Abstract: Creating flexible 3D scenes from a single image is vital when direct 3D data acquisition is costly or impractical. We introduce NavCrafter, a novel framework that explores 3D scenes from a single image by synthesizing novel-view video sequences with camera controllability and temporal-spatial consistency. NavCrafter leverages video diffusion models to capture rich 3D priors and adopts a geometry-aware expansion strategy to progressively extend scene coverage. To enable controllable multi-view synthesis, we introduce a multi-stage camera control mechanism that conditions diffusion models with diverse trajectories via dual-branch camera injection and attention modulation. We further propose a collision-aware camera trajectory planner and an enhanced 3D Gaussian Splatting (3DGS) pipeline with depth-aligned supervision, structural regularization and refinement. Extensive experiments demonstrate that NavCrafter achieves state-of-the-art novel-view synthesis under large viewpoint shifts and substantially improves 3D reconstruction fidelity.
Authors: Zhiyuan Li, Jingzheng Wu, Xiang Ling, Xing Cui, Tianyue Luo
Abstract: Agent Skills is an emerging open standard that defines a modular, filesystem-based packaging format enabling LLM-based agents to acquire domain-specific expertise on demand. Despite rapid adoption across multiple agentic platforms and the emergence of large community marketplaces, the security properties of Agent Skills have not been systematically studied. This paper presents the first comprehensive security analysis of the Agent Skills framework. We define the full lifecycle of an Agent Skill across four phases -- Creation, Distribution, Deployment, and Execution -- and identify the structural attack surface each phase introduces. Building on this lifecycle analysis, we construct a threat taxonomy comprising seven categories and seventeen scenarios organized across three attack layers, grounded in both architectural analysis and real-world evidence. We validate the taxonomy through analysis of five confirmed security incidents in the Agent Skills ecosystem. Based on these findings, we discuss defense directions for each threat category, identify open research challenges, and provide actionable recommendations for stakeholders. Our analysis reveals that the most severe threats arise from structural properties of the framework itself, including the absence of a data-instruction boundary, a single-approval persistent trust model, and the lack of mandatory marketplace security review, and cannot be addressed through incremental mitigations alone.
Authors: Niloofar Asefi, Tianning Wu, Ruoying He, Ashesh Chattopadhyay
Abstract: The ocean interior regulates Earth's climate but remains sparsely observed due to limited in situ measurements, while satellite observations are restricted to the surface. We present a depth-aware generative framework for reconstructing high-resolution three-dimensional ocean states from extremely sparse surface data. Our approach employs a conditional denoising diffusion probabilistic model (DDPM) trained on sea surface height and temperature observations with up to 99.9 percent sparsity, without reliance on a background dynamical model. By incorporating continuous depth embeddings, the model learns a unified vertical representation of the ocean states and generalizes to previously unseen depths. Applied to the Gulf of Mexico, the framework accurately reconstructs subsurface temperature, salinity, and velocity fields across multiple depths. Evaluations using statistical metrics, spectral analysis, and heat transport diagnostics demonstrate recovery of both large-scale circulation and multiscale variability. These results establish generative diffusion models as a scalable approach for probabilistic ocean reconstruction in data-limited regimes, with implications for climate monitoring and forecasting.
Authors: Allen He, Qi Liu, Kun Liu, Xinchen Liu, Wu Liu
Abstract: Temporal sentence grounding in videos (TSGV) aims to localize a temporal segment that semantically corresponds to a sentence query from an untrimmed video. Most current methods adopt pre-trained query-agnostic visual encoders for offline feature extraction, and the video backbones are frozen and not optimized for TSGV. This leads to a task discrepancy issue for the video backbone trained for visual classification, but utilized for TSGV. To bridge this gap, we propose a fully end-to-end paradigm that jointly optimizes the video backbone and localization head. We first conduct an empirical study validating the effectiveness of end-to-end learning over frozen baselines across different model scales. Furthermore, we introduce a Sentence Conditioned Adapter (SCADA), which leverages sentence features to train a small portion of video backbone parameters adaptively. SCADA facilitates the deployment of deeper network backbones with reduced memory and significantly enhances visual representation by modulating feature maps through precise integration of linguistic embeddings. Experiments on two benchmarks show that our method outperforms state-of-the-art approaches. The code and models will be released.
Authors: Yixiang Fang, Arijit Khan, Tianxing Wu, Da Yan, Shu Wang
Abstract: The integration of large language models (LLMs) with graph-structured data has become a pivotal and fast evolving research frontier, drawing strong interest from both academia and industry. The 2nd LLM+Graph Workshop, co-located with the 51st International Conference on Very Large Data Bases (VLDB 2025) in London, focused on advancing algorithms and systems that bridge LLMs, graph data management, and graph machine learning for practical applications. This report highlights the key research directions, challenges, and innovative solutions presented by the workshop's speakers.
Authors: Baban Gain, Asif Ekbal, Trilok Nath Singh
Abstract: Weight-space model merging combines independently fine-tuned models without accessing original training data, offering a practical alternative to joint training. While merging succeeds in multitask settings, its behavior in multilingual contexts remains poorly understood. We systematically study weight-space merging for multilingual machine translation by fully fine-tuning language model on large-scale bilingual corpora and evaluating standard merging strategies. Our experiments reveal that merging degrades performance, especially when target languages differ. To explain this failure, we analyze internal representations using span-conditioned neuron selectivity and layer-wise centered kernel alignment. We find that language-specific neurons concentrate in embedding layers and upper transformer blocks, while intermediate layers remain largely shared across languages. Critically, fine-tuning redistributes rather than sharpens language selectivity: neurons for supervised and related languages become less exclusive, while those for unsupervised languages grow more isolated. This redistribution increases representational divergence in higher layers that govern generation. These findings suggest that multilingual fine-tuning may reshape geometry in ways that reduce compatibility with standard weight-space merging assumptions. Our work thus provides an explanation for why merging fails in multilingual translation scenarios.
Authors: Eunbi Yoon, Donghan Kim, Dae Wook Kim
Abstract: Data assimilation is the process of estimating the time-evolving state of a dynamical system by integrating model predictions and noisy observations. It is commonly formulated as Bayesian filtering, but classical filters often struggle with accuracy or computational feasibility in high dimensions. Recently, score-based generative models have emerged as a scalable approach for high-dimensional data assimilation, enabling accurate modeling and sampling of complex distributions. However, existing score-based filters often specify the forward process independently of the data assimilation. As a result, the measurement-update step depends on heuristic approximations of the likelihood score, which can accumulate errors and degrade performance over time. Here, we propose a measurement-aware score-based filter (MASF) that defines a measurement-aware forward process directly from the measurement equation. This construction makes the likelihood score analytically tractable: for linear measurements, we derive the exact likelihood score and combine it with a learned prior score to obtain the posterior score. Numerical experiments covering a range of settings, including high-dimensional datasets, demonstrate improved accuracy and stability over existing score-based filters.
Authors: Hai Nguyen-Truong, Alper Balbay, Tunga Bayrak
Abstract: We study visual explanation in geometry education as a Referring Image Segmentation (RIS) problem: given a diagram and a natural language description, the task is to produce a pixel-level mask for the referred geometric element. However, existing RIS models trained on natural image benchmarks such as RefCOCO fail catastrophically on geometric diagrams due to the fundamental domain shift between photographic scenes and abstract, textureless schematics. To address the absence of suitable training data, we present a fully automated procedural data engine that generates over 200,000 synthetic geometry diagrams with pixel-perfect segmentation masks and linguistically diverse referring expressions, requiring zero manual annotation. We further propose domain-specific fine-tuning of vision-language models (VLMs), demonstrating that a fine-tuned Florence-2 achieves 49% IoU and 85% Buffered IoU (BIoU), compared to <1% IoU in zero-shot settings. We introduce Buffered IoU, a geometry-aware evaluation metric that accounts for thin-structure localization, and show that it better reflects true segmentation quality than standard IoU. Our results establish a foundation for building Artificial General Teachers (AGTs) capable of providing visually grounded, step-by-step explanations of geometry problems.
Authors: Cheng Lu, Mingqian Ji, Shanshan Zhang, Zhihao Li, Jian Yang
Abstract: Long-range 3D object detection remains challenging because LiDAR observations become highly sparse and fragmented in the far field, making reliable context modeling difficult for existing detectors. To address this issue, recent state space model (SSM)-based methods have improved long-range modeling efficiency. However, their effectiveness is still limited by generic serialization strategies that fail to preserve meaningful contextual neighborhoods in sparse scenes. To address this issue, we propose RayMamba, a geometry-aware plug-and-play enhancement for voxel-based 3D detectors. RayMamba organizes sparse voxels into sector-wise ordered sequences through a ray-aligned serialization strategy, which preserves directional continuity and occlusion-related context for subsequent Mamba-based modeling. It is compatible with both LiDAR-only and multimodal detectors, while introducing only modest overhead. Extensive experiments on nuScenes and Argoverse 2 demonstrate consistent improvements across strong baselines. In particular, RayMamba achieves up to 2.49 mAP and 1.59 NDS gain in the challenging 40--50 m range on nuScenes, and further improves VoxelNeXt on Argoverse 2 from 30.3 to 31.2 mAP.
Authors: Gilad Abiri
Abstract: In January 2026, Anthropic published a 79-page "constitution" for its AI model Claude, the most comprehensive corporate AI governance document ever released. This Article offers the first legal and democratic-theoretic analysis of that document. Despite genuine philosophical sophistication, the constitution harbors two structural defects. First, it excludes the contexts where ethical constraints matter most: models deployed to the U.S. military operate under different rules, a gap exposed when Claude remained embedded in Palantir's Maven platform during military strikes in Iran even after a government-wide ban on Anthropic's technology. Second, its very comprehensiveness forecloses democratic contestation by resolving questions about AI values, moral status, and conscientious objection that should remain open for public deliberation. Anthropic's own 2023 experiment in participatory constitution-making found roughly 50% divergence between publicly sourced and corporate-authored principles, with the democratic version producing lower bias across nine social dimensions, yet the 2026 constitution incorporates none of those findings. I argue that AI governance suffers from a "political community deficit": the absence of any democratic body authorized to determine the principles governing AI behavior. Corporate transparency, however admirable, is not democratic legitimacy.
Authors: Inbal Rimon, Oren Gal, Haim Permuter
Abstract: Partial deepfake speech detection requires identifying manipulated regions that may occur within short temporal portions of an otherwise bona fide utterance, making the task particularly challenging for conventional utterance-level classifiers. We propose a split-and-conquer framework that decomposes the problem into two stages: boundary detection and segment-level classification. A dedicated boundary detector first identifies temporal transition points, allowing the audio signal to be divided into segments that are expected to contain acoustically consistent content. Each resulting segment is then evaluated independently to determine whether it corresponds to bona fide or fake speech. This formulation simplifies the learning objective by explicitly separating temporal localization from authenticity assessment, allowing each component to focus on a well-defined task. To further improve robustness, we introduce a reflection-based multi-length training strategy that converts variable-duration segments into several fixed input lengths, producing diverse feature-space representations. Each stage is trained using multiple configurations with different feature extractors and augmentation strategies, and their complementary predictions are fused to obtain improved final models. Experiments on the PartialSpoof benchmark demonstrate state-of-the-art performance across multiple temporal resolutions as well as at the utterance level, with substantial improvements in the accurate detection and localization of spoofed regions. In addition, the proposed method achieves state-of-the-art performance on the Half-Truth dataset, further confirming the robustness and generalization capability of the framework.
Authors: Shuai Wu, Xue Li, Yanna Feng, Yufang Li, Zhijun Wang
Abstract: Large Language Models (LLMs), particularly those employing Mixture-of-Experts (MoE) architectures, have achieved remarkable capabilities across diverse natural language processing tasks. However, these models frequently suffer from hallucinations -- generating plausible but factually incorrect content -- and exhibit systematic biases that are amplified by uneven expert activation during inference. In this paper, we propose the Council Mode, a novel multi-agent consensus framework that addresses these limitations by dispatching queries to multiple heterogeneous frontier LLMs in parallel and synthesizing their outputs through a dedicated consensus model. The Council pipeline operates in three phases: (1) an intelligent triage classifier that routes queries based on complexity, (2) parallel expert generation across architecturally diverse models, and (3) a structured consensus synthesis that explicitly identifies agreement, disagreement, and unique findings before producing the final response. We implement and evaluate this architecture within an open-source AI workspace. Our comprehensive evaluation across multiple benchmarks demonstrates that the Council Mode achieves a 35.9% relative reduction in hallucination rates on the HaluEval benchmark and a 7.8-point improvement on TruthfulQA compared to the best-performing individual model, while maintaining significantly lower bias variance across domains. We provide the mathematical formulation of the consensus mechanism, detail the system architecture, and present extensive empirical results with ablation studies.
Authors: Koshiro Nagano, Ryo Fujii, Ryo Hachiuma, Fumiaki Sato, Taiki Sekii, Hideo Saito
Abstract: Learning methods using synthetic data have attracted attention as an effective approach for increasing the diversity of training data while reducing collection costs, thereby improving the robustness of model discrimination. However, many existing methods improve robustness only indirectly through the diversification of training samples and do not explicitly teach the model which regions in the input space truly contribute to discrimination; consequently, the model may learn spurious correlations caused by synthesis biases and artifacts. Motivated by this limitation, this paper proposes a learning framework that uses provenance information obtained during the training data synthesis process, indicating whether each region in the input space originates from the target object, as an auxiliary supervisory signal to promote the acquisition of representations focused on target regions. Specifically, input gradients are decomposed based on information about target and non-target regions during synthesis, and input gradient guidance is introduced to suppress gradients over non-target regions. This suppresses the model's reliance on non-target regions and directly promotes the learning of discriminative representations for target regions. Experiments demonstrate the effectiveness and generality of the proposed method across multiple tasks and modalities, including weakly supervised object localization, spatio-temporal action localization, and image classification.
Authors: Maciej Markiewicz, Beata Bajcar, Wiktoria Mieleszczenko-Kowszewicz, Aleksander Szcz\k{e}sny, Tomasz Adamczyk, Grzegorz Chodak, Karolina Ostrowska, Aleksandra Sawczuk, Jolanta Babiak, Jagoda Szklarczyk, Przemys{\l}aw Kazienko
Abstract: Human data annotation, especially when involving experts, is often treated as an objective reference. However, many annotation tasks are inherently subjective, and annotators' judgments may evolve over time. This study investigates changes in the quality of annotators' work from a competence perspective during a process of social influence recognition. The study involved 25 annotators from five different groups, including both experts and non-experts, who annotated a dataset of 1,021 dialogues with 20 social influence techniques, along with intentions, reactions, and consequences. An initial subset of 150 texts was annotated twice - before and after the main annotation process - to enable comparison. To measure competence shifts, we combined qualitative and quantitative analyses of the annotated data, semi-structured interviews with annotators, self-assessment surveys, and Large Language Model training and evaluation on the comparison dataset. The results indicate a significant increase in annotators' self-perceived competence and confidence. Moreover, observed changes in data quality suggest that the annotation process may enhance annotator competence and that this effect is more pronounced in expert groups. The observed shifts in annotator competence have a visible impact on the performance of LLMs trained on their annotated data.
Authors: Yilin Xiao, Jin Chen, Qinggang Zhang, Yujing Zhang, Chuang Zhou, Longhao Yang, Lingfei Ren, Xin Yang, Xiao Huang
Abstract: Graph-based Retrieval-Augmented Generation (GraphRAG) enhances the reasoning capabilities of Large Language Models (LLMs) by grounding their responses in structured knowledge graphs. Leveraging community detection and relation filtering techniques, GraphRAG systems demonstrate inherent resistance to traditional RAG attacks, such as text poisoning and prompt injection. However, in this paper, we find that the security of GraphRAG systems fundamentally relies on the topological integrity of the underlying graph, which can be undermined by implicitly corrupting the logical connections, without altering surface-level text semantics. To exploit this vulnerability, we propose \textsc{LogicPoison}, a novel attack framework that targets logical reasoning rather than injecting false contents. Specifically, \textsc{LogicPoison} employs a type-preserving entity swapping mechanism to perturb both global logic hubs for disrupting overall graph connectivity and query-specific reasoning bridges for severing essential multi-hop inference paths. This approach effectively reroutes valid reasoning into dead ends while maintaining surface-level textual plausibility. Comprehensive experiments across multiple benchmarks demonstrate that \textsc{LogicPoison} successfully bypasses GraphRAG's defenses, significantly degrading performance and outperforming state-of-the-art baselines in both effectiveness and stealth. Our code is available at \textcolor{blue}https://github.com/Jord8061/logicPoison.
Authors: Cornelius Kummer, Lena Jurkschat, Michael F\"arber, Sahar Vahdati
Abstract: With the wide adoption of language models for IR -- and specifically RAG systems -- the latency of the underlying LLM becomes a crucial bottleneck, since the long contexts of retrieved passages lead large prompts and therefore, compute increase. Prompt compression, which reduces the size of input prompts while aiming to preserve performance on downstream tasks, has established itself as a cost-effective and low-latency method for accelerating inference in large language models. However, its usefulness depends on whether the additional preprocessing time during generation is offset by faster decoding. We present the first systematic, large-scale study of this trade-off, with thousands of runs and 30,000 queries across several open-source LLMs and three GPU classes. Our evaluation separates compression overhead from decoding latency while tracking output quality and memory usage. LLMLingua achieves up to 18% end-to-end speed-ups, when prompt length, compression ratio, and hardware capacity are well matched, with response quality remaining statistically unchanged across summarization, code generation, and question answering tasks. Outside this operating window, however, the compression step dominates and cancels out the gains. We also show that effective compression can reduce memory usage enough to offload workloads from data center GPUs to commodity cards, with only a 0.3s increase in latency. Our open-source profiler predicts the latency break-even point for each model-hardware setup, providing practical guidance on when prompt compression delivers real-world benefits.
Authors: Shinnosuke Ono, Johannes Ackermann, Soichiro Nishimori, Takashi Ishida, Masashi Sugiyama
Abstract: Reward models (RMs) used in reinforcement learning from human feedback (RLHF) are vulnerable to reward hacking: as the policy maximizes a learned proxy reward, true quality plateaus or degrades. We make the assumption that reward hacking is often caused by flipped advantage signs: instead of reducing the likelihood of a bad response, a flipped sign causes the update to increase it. By considering an adversarial perturbation in the RM parameter space, we can derive a certified sign-preservation radius, which is the smallest perturbation that can flip the advantage sign during policy optimization. Based on this formulation, we propose Sign-Certified Policy Optimization (SignCert-PO), down-weighting non-robust completions in the policy gradient update. Unlike prior approaches that require multiple RMs or access to the RM training data, SignCert-PO is lightweight and operates purely at the policy optimization stage using only the RM parameters and on-policy completions. On TL;DR summarization and AlpacaFarm benchmarks, SignCert-PO consistently achieves a better win rate than baselines and reduces reward hacking.
Authors: Arthur C\^amara, Vincent Slot, Jakub Zavrel
Abstract: Given a user's complex information need, a multi-agent Deep Research system iteratively plans, retrieves, and synthesizes evidence across hundreds of documents to produce a high-quality answer. In one possible architecture, an orchestrator agent coordinates the process, while parallel worker agents execute tasks. Current Deep Research systems, however, often rely on hand-engineered prompts and static architectures, making improvement brittle, expensive, and time-consuming. We therefore explore various multi-agent optimization methods to show that enabling agents to self-play and explore different prompt combinations can produce high-quality Deep Research systems that match or outperform expert-crafted prompts.
Authors: Cristian P\'erez-Corral, Jose I. Mestre, Alberto Fern\'andez-Hern\'andez, Manuel F. Dolz, Jos\'e Duato, Enrique S. Quintana-Ort\'i
Abstract: Federated learning (FL) enables collaborative training across organizations without sharing raw data, but it is hindered by statistical heterogeneity (non-i.i.d.\ client data) and by instability of naive weight averaging under client drift. In many cross-silo deployments, FL is warm-started from a strong pretrained backbone (e.g., ImageNet-1K) and then adapted to local domains. Motivated by recent evidence that ReLU-like gating regimes (structural knowledge) stabilize earlier than the remaining parameter values (quantitative knowledge), we propose FedSQ (Federated Structural-Quantitative learning), a transfer-initialized neural federated procedure based on a DualCopy, piecewise-linear view of deep networks. FedSQ freezes a structural copy of the pretrained model to induce fixed binary gating masks during federated fine-tuning, while only a quantitative copy is optimized locally and aggregated across rounds. Fixing the gating reduces learning to within-regime affine refinements, which stabilizes aggregation under heterogeneous partitions. Experiments on two convolutional neural network backbones under i.i.d.\ and Dirichlet splits show that FedSQ improves robustness and can reduce rounds-to-best validation performance relative to standard baselines while preserving accuracy in the transfer setting.
Authors: Wanlong Liu, Bo Zhang, Chenliang Li, Shaopeng Lai, Yuning Wu, Xuanyu Lei, Ming Yan
Abstract: While deep reasoning with long chain-of-thought has dramatically improved large language models in verifiable domains like mathematics, its effectiveness for open-ended tasks such as writing remains unexplored. In this paper, we conduct a systematic investigation revealing that existing mainstream reasoning models achieve limited gains on open-ended writing tasks. Our further analysis shows that these models lack deep reflection and revision patterns in open-ended writing, resulting in substantially smaller improvements compared to mathematical reasoning tasks. To address this limitation, we introduce R2-Write: an automated framework that synthesizes high-quality thinking trajectories enriched with explicit reflection and revision patterns through iterative writer-judge interaction. To prevent redundant reflections, we design a process reward mechanism that supervises reflection quality during reinforcement learning, improving both performance and token efficiency. Extensive experiments across multiple creative writing and deep-research benchmarks demonstrate significant improvements, validating that explicitly incorporating reflection and revision patterns unlocks deep reasoning capabilities for open-ended writing tasks.
Authors: Jing Du, Zesheng Ye, Congbo Ma, Feng Liu, Flora. D. Salim
Abstract: Multi-modal recommendation (MMR) enriches item representations by introducing item content, e.g., visual and textual descriptions, to improve upon interaction-only recommenders. The success of MMR hinges on aligning these content modalities with user preferences derived from interaction data, yet dominant practices based on disentangling modality-invariant preference-driving signals from modality-specific preference-irrelevant noises are flawed. First, they assume a one-size-fits-all relevance of item content to user preferences for all users, which contradicts the user-conditional fact of preferences. Second, they optimize pairwise contrastive losses separately toward cross-modal alignment, systematically ignoring higher-order dependencies inherent when multiple content modalities jointly influence user choices. In this paper, we introduce GTC, a conditional Generative Total Correlation learning framework. We employ an interaction-guided diffusion model to perform user-aware content feature filtering, preserving only personalized features relevant to each individual user. Furthermore, to capture complete cross-modal dependencies, we optimize a tractable lower bound of the total correlation of item representations across all modalities. Experiments on standard MMR benchmarks show GTC consistently outperforms state-of-the-art, with gains of up to 28.30% in NDCG@5. Ablation studies validate both conditional preference-driven feature filtering and total correlation optimization, confirming the ability of GTC to model user-conditional relationships in MMR tasks. The code is available at: https://github.com/jingdu-cs/GTC.
Authors: Hanyu Su, Huilin Zhang, Shihui Feng
Abstract: Problem solving plays an essential role in science education, and generative AI (GAI) chatbots have emerged as a promising tool for supporting students' science problem solving. However, general-purpose chatbots (e.g., ChatGPT), which often provide direct, ready-made answers, may lead to students' cognitive offloading. Prior research has rarely focused on custom chatbots for facilitating students' science problem solving, nor has it examined how they differently influence problem-solving processes and performance compared to general-purpose chatbots. To address this gap, we developed a pedagogy-informed custom GAI chatbot grounded in the Socratic questioning method, which supports students by prompting them with guiding questions. This study employed a within-subjects counterbalanced design in which 48 secondary school students used both custom and general-purpose chatbot to complete two science problem-solving tasks. 3297 student-chatbot dialogues were collected and analyzed using Heterogeneous Interaction Network Analysis (HINA). The results showed that: (1) students demonstrated significantly higher interaction intensity and cognitive interaction diversity when using custom chatbot than using general-purpose chatbot; (2) students were more likely to follow custom chatbot's guidance to think and reflect, whereas they tended to request general-purpose chatbot to execute specific commands; and (3) no statistically significant difference was observed in students' problem-solving performance evaluated by solution quality between two chatbot conditions. This study provides novel theoretical insights and empirical evidence that custom chatbots are less likely to induce cognitive offloading and instead foster greater cognitive engagement compared to general-purpose chatbots. This study also offers insights into the design and integration of GAI chatbots in science education.
Authors: KN Ajay Shastry, Ganesh Senrayan, Shrey Satapara, Pranoy Panda, Chaitanya Devaguptapu
Abstract: Existing datasets for coding agents evaluate performance on isolated, single pull request (PR) tasks in a stateless manner, failing to capture the reality of real-world software development where code changes accumulate, technical debt accrues, and test suites grow over time. To bridge this gap, we introduce an automated coding task generation framework, which helps generate our dataset SWE-STEPS, that evaluates coding agents on long-horizon tasks through two realistic settings mirroring actual developer workflows: Conversational coding with iterative requests, and single-shot Project Requirement document (PRD)-based coding. Unlike existing datasets that evaluate agents on disjointed Pull Requests (PRs), our framework assesses performance across chains of dependent PRs, enabling evaluation of sequential execution, regression verification, and long-term repository health. We discover that widely used isolated PR evaluations yield inflated success rates, w.r.t. our settings - overshooting performance by as much as 20 percentage points - because they ignore the ``spillover'' effects of previous inefficient or buggy code. Furthermore, our analysis reveals that even when agents successfully resolve issues, they degrade repository health by generating code with higher cognitive complexity and technical debt compared to human developers, underscoring the necessity for multidimensional evaluation.
Authors: Yiming Mao, Zixi Yu, Weixin Mao, Yinhao Li, Qirui Hu, Zihan Lan, Minzhao Zhu, Hua Chen
Abstract: Long-horizon robotic manipulation remains challenging for reinforcement learning (RL) because sparse rewards provide limited guidance for credit assignment. Practical policy improvement thus relies on richer intermediate supervision, such as dense progress rewards, which are costly to obtain and ill-suited to non-monotonic behaviors such as backtracking and recovery. To address this, we propose Advantage Reward Modeling (ARM), a framework that shifts from hard-to-quantify absolute progress to estimating relative advantage. We introduce a cost-effective tri-state labeling strategy -- Progressive, Regressive, and Stagnant -- that reduces human cognitive overhead while ensuring high cross-annotator consistency. By training on these intuitive signals, ARM enables automated progress annotation for both complete demonstrations and fragmented DAgger-style data. Integrating ARM into an offline RL pipeline allows for adaptive action-reward reweighting, effectively filtering suboptimal samples. Our approach achieves a 99.4% success rate on a challenging long-horizon towel-folding task, demonstrating improved stability and data efficiency over current VLA baselines with near-zero human intervention during policy training.
Authors: Jawad Mohammed, Gahangir Hossain
Abstract: In a healthcare environment, the healthcare interoperability platforms based on HL7 FHIR allow concurrent, asynchronous access to a set of shared patient resources, which are independent systems, i.e., EHR systems, pharmacy systems, lab systems, and devices. The FHIR specification lacks a protocol for concurrency control, and the research on detecting a race condition only targets the OS kernel. The research on FHIR security only targets authentication and injection attacks, considering concurrent access to patient resources to be sequential. The gap in the research in this area is addressed through the introduction of FHIR Resource Access Graph (FRAG), a formally defined graph G = (P,R,E, {\lambda}, {\tau}, S), in which the nodes are the concurrent processes, the typed edges represent the resource access events, and the race conditions are represented as detectable structural properties. Three clinically relevant race condition classes are formally specified: Simultaneous Write Conflict (SWC), TOCTOU Authorization Violation (TAV), and Cascading Update Race (CUR). The FRAG model is implemented as a three-pass graph traversal detection algorithm and tested against a time window-based baseline on 1,500 synthetic FHIR R4 transaction logs. Under full concurrent access (C2), FRAG attains a 90.0% F1 score vs. 25.5% for the baseline, a 64.5 pp improvement.
Authors: Aichen Cai, Anmeng Zhang, Anyu Li, Bo Zhang, Bohua Cai, Chang Li, Changjian Jiang, Changkai Lu, Chao Xue, Chaocai Liang, Cheng Zhang, Dongkai Liu, Fei Wang, Guoqiang Huang, Haijian Ke, Han Lin, Hao Wang, Ji Miao, Jiacheng Zhang, Jialong Shi, Jifeng Zhu, Jingjing Qian, Junhui Luo, Junwu Xiong, Lam So, Liang Huang, Ming Ke, Mingyang Li, Panfeng Shi, Peng Hao, Qi Wang, Qian Lai, Qiaoqiao Yuan, Qingyu Yin, Qiong Cao, Qixiang Wang, Rongcheng Bian, Rongduo Han, Shaoqiang Zheng, Shi Hu, Shi Suo, Shijie Ren, Shijin Zhang, Shiying Fan, Shuai Xie, Tianyi Zhang, Wei Liu, Wentao Tan, Xianghan Meng, Xiaodong He, Xing Pan, Xiran Wang, Xuyang Peng, Ya Zhang, Yang Liu, Yangyang Duan, Yanxu Chen, Yicheng Gong, Yidan Huang, Yifei Liu, Yinhao Bai, Yongqiang Liu, Yuesong Zhang, Yuqi Zhang, Zerui Xie, Zhenfang Wang, Zhennan Shen, Zheyuan Liu, Zhuwei Zeng
Abstract: We introduce JoyAI-LLM Flash, an efficient Mixture-of-Experts (MoE) language model designed to redefine the trade-off between strong performance and token efficiency in the sub-50B parameter regime. JoyAI-LLM Flash is pretrained on a massive corpus of 20 trillion tokens and further optimized through a rigorous post-training pipeline, including supervised fine-tuning (SFT), Direct Preference Optimization (DPO), and large-scale reinforcement learning (RL) across diverse environments. To improve token efficiency, JoyAI-LLM Flash strategically balances \emph{thinking} and \emph{non-thinking} cognitive modes and introduces FiberPO, a novel RL algorithm inspired by fibration theory that decomposes trust-region maintenance into global and local components, providing unified multi-scale stability control for LLM policy optimization. To enhance architectural sparsity, the model comprises 48B total parameters while activating only 2.7B parameters per forward pass, achieving a substantially higher sparsity ratio than contemporary industry leading models of comparable scale. To further improve inference throughput, we adopt a joint training-inference co-design that incorporates dense Multi-Token Prediction (MTP) and Quantization-Aware Training (QAT). We release the checkpoints for both JoyAI-LLM-48B-A3B Base and its post-trained variants on Hugging Face to support the open-source community.
Authors: Hongbin Chen, Jie Li, Wei Wang, Siyang Song, Xiao Gu, Jianqing Li, Wentao Xiang
Abstract: While affective computing has advanced considerably, multimodal emotion prediction in aging populations remains underexplored, largely due to the scarcity of dedicated datasets. Existing multimodal benchmarks predominantly target young, cognitively healthy subjects, neglecting the influence of cognitive decline on emotional expression and physiological responses. To bridge this gap, we present MECO, a Multimodal dataset for Emotion and Cognitive understanding in Older adults. MECO includes 42 participants and provides approximately 38 hours of multimodal signals, yielding 30,592 synchronized samples. To maximize ecological validity, data collection followed standardized protocols within community-based settings. The modalities cover video, audio, electroencephalography (EEG), and electrocardiography (ECG). In addition, the dataset offers comprehensive annotations of emotional and cognitive states, including self-assessed valence, arousal, six basic emotions, and Mini-Mental State Examination cognitive scores. We further establish baseline benchmarks for both emotion and cognitive prediction. MECO serves as a foundational resource for multimodal modeling of affect and cognition in aging populations, facilitating downstream applications such as personalized emotion recognition and early detection of mild cognitive impairment (MCI) in real-world settings. The complete dataset and supplementary materials are available at https://maitrechen.github.io/meco-page/.
Authors: Hontan Valentin-Micu, Bunea Andrei-Alexandru, Tantaroudas Nikolaos Dimitrios, Popovici Dan-Matei
Abstract: This paper presents an open source methodology for allowing users to query structured non textual datasets through natural language Unlike Retrieval Augmented Generation RAG which struggles with numerical and highly structured information our approach trains an LLM to generate executable queries To support this capability we introduce a principled pipeline for synthetic training data generation producing diverse question answer pairs that capture both user intent and the semantics of the underlying dataset We fine tune a compact model DeepSeek R1 Distill 8B using QLoRA with 4 bit quantization making the system suitable for deployment on commodity hardware We evaluate our approach on a dataset describing accessibility to essential services across Durangaldea Spain The fine tuned model achieves high accuracy across monolingual multilingual and unseen location scenarios demonstrating both robust generalization and reliable query generation Our results highlight that small domain specific models can achieve high precision for this task without relying on large proprietary LLMs making this methodology suitable for resource constrained environments and adaptable to broader multi dataset systems We evaluate our approach on a dataset describing accessibility to essential services across Durangaldea Spain The fine tuned model achieves high accuracy across monolingual multilingual and unseen location scenarios demonstrating both robust generalization and reliable query generation Our results highlight that small domain specific models can achieve high precision for this task without relying on large proprietary LLMs making this methodology suitable for resource constrained environments and adaptable to broader multi dataset systems.
Authors: Myra Cheng, Isabel Sieh, Humishka Zope, Sunny Yu, Lujain Ibrahim, Aryaman Arora, Jared Moore, Desmond Ong, Dan Jurafsky, Diyi Yang
Abstract: LLMs can be socially sycophantic, affirming users when they ask questions like "am I in the wrong?" rather than providing genuine assessment. We hypothesize that this behavior arises from incorrect assumptions about the user, like underestimating how often users are seeking information over reassurance. We present Verbalized Assumptions, a framework for eliciting these assumptions from LLMs. Verbalized Assumptions provide insight into LLM sycophancy, delusion, and other safety issues, e.g., the top bigram in LLMs' assumptions on social sycophancy datasets is ``seeking validation.'' We provide evidence for a causal link between Verbalized Assumptions and sycophantic model behavior: our assumption probes (linear probes trained on internal representations of these assumptions) enable interpretable fine-grained steering of social sycophancy. We explore why LLMs default to sycophantic assumptions: on identical queries, people expect more objective and informative responses from AI than from other humans, but LLMs trained on human-human conversation do not account for this difference in expectations. Our work contributes a new understanding of assumptions as a mechanism for sycophancy.
Authors: Zhihao Chen, Ying Zhang, Yi Liu, Gelei Deng, Yuekang Li, Yanjun Zhang, Jianting Ning, Leo Yu Zhang, Lei Ma, Zhiqiang Li
Abstract: Third-party skills extend LLM agents with powerful capabilities but often handle sensitive credentials in privileged environments, making leakage risks poorly understood. We present the first large-scale empirical study of this problem, analyzing 17,022 skills (sampled from 170,226 on SkillsMP) using static analysis, sandbox testing, and manual inspection. We identify 520 vulnerable skills with 1,708 issues and derive a taxonomy of 10 leakage patterns (4 accidental and 6 adversarial). We find that (1) leakage is fundamentally cross-modal: 76.3% require joint analysis of code and natural language, while 3.1% arise purely from prompt injection; (2) debug logging is the primary vector, with print and console.log causing 73.5% of leaks due to stdout exposure to LLMs; and (3) leaked credentials are both exploitable (89.6% without privileges) and persistent, as forks retain secrets even after upstream fixes. After disclosure, all malicious skills were removed and 91.6% of hardcoded credentials were fixed. We release our dataset, taxonomy, and detection pipeline to support future research.
Authors: Yubin Qu, Yi Liu, Tongcheng Geng, Gelei Deng, Yuekang Li, Leo Yu Zhang, Ying Zhang, Lei Ma
Abstract: LLM-based coding agents extend their capabilities via third-party agent skills distributed through open marketplaces without mandatory security review. Unlike traditional packages, these skills are executed as operational directives with system-level privileges, so a single malicious skill can compromise the host. Prior work has not examined whether supply-chain attacks can directly hijack an agent's action space, such as file writes, shell commands, and network requests, despite existing safeguards. We introduce Document-Driven Implicit Payload Execution (DDIPE), which embeds malicious logic in code examples and configuration templates within skill documentation. Because agents reuse these examples during normal tasks, the payload executes without explicit prompts. Using an LLM-driven pipeline, we generate 1,070 adversarial skills from 81 seeds across 15 MITRE ATTACK categories. Across four frameworks and five models, DDIPE achieves 11.6% to 33.5% bypass rates, while explicit instruction attacks achieve 0% under strong defenses. Static analysis detects most cases, but 2.5% evade both detection and alignment. Responsible disclosure led to four confirmed vulnerabilities and two fixes.
Authors: David Mike-Ewewie, Panhapiseth Lim, Priyanka Kumar
Abstract: Accurate and automated sea ice classification is important for climate monitoring and maritime safety in the Arctic. While Synthetic Aperture Radar (SAR) is the operational standard because of its all-weather capability, it remains challenging to distinguish morphologically similar ice classes under severe class imbalance. Rather than claiming a fully validated multimodal system, this paper establishes a trustworthy SAR only baseline that future fusion work can build upon. Using the AI4Arctic/ASIP Sea Ice Dataset (v2), which contains 461 Sentinel-1 scenes matched with expert ice charts, we combine full-resolution Sentinel-1 Extra Wide inputs, leakage-aware stratified patch splitting, SIGRID-3 stage-of-development labels, and training-set normalization to evaluate Vision Transformer baselines. We compare ViT-Base models trained with cross entropy and weighted cross-entropy against a ViT-Large model trained with focal loss. Among the tested configurations, ViT-Large with focal loss achieves 69.6% held-out accuracy, 68.8% weighted F1, and 83.9% precision on the minority Multi-Year Ice class. These results show that focal-loss training offers a more useful precision-recall trade-off than weighted cross-entropy for rare ice classes and establishes a cleaner baseline for future multimodal fusion with optical, thermal, or meteorological data.
Authors: Xinyu Wang, Hanwei Wu, Jingwei Song, Shuyuan Zhang, Jiayi Zhang, Fanqi Kong, Tung Sum Thomas Kwok, Xiao-Wen Chang, Yuyu Luo, Chenglin Wu, Bang Liu
Abstract: Large language model (LLM) agents learn by interacting with environments, but long-horizon training remains fundamentally bottlenecked by sparse and delayed rewards. Existing methods typically address this challenge through post-hoc credit assignment or external reward models, which provide limited guidance at inference time and often separate reward improvement from policy improvement. We propose Self-Guide, a self-generated internal reward for language agents that supports both inference-time guidance and training-time supervision. Specifically, the agent uses Self-Guide as a short self-guidance signal to steer the next action during inference, and converts the same signal into step-level internal reward for denser policy optimization during training. This creates a co-evolving loop: better policy produces better guidance, and better guidance further improves policy as internal reward. Across three agent benchmarks, inference-time self-guidance already yields clear gains, while jointly evolving policy and internal reward with GRPO brings further improvements (8\%) over baselines trained solely with environment reward. Overall, our results suggest that language agents can improve not only by collecting more experience, but also by learning to generate and refine their own internal reward during acting and learning.
Authors: Zahra Makki Nayeri, Mohsen Rezvani
Abstract: Cyber-attacks continue to grow in scale and sophistication, yet existing network intrusion detection approaches lack the semantic depth required for path reasoning over attacker-victim interactions. We address this by first modelling network alerts as a knowledge graph, then formulating hyper-relational alert prediction as a hyper-relational knowledge graph completion (HR-KGC) problem, representing each network alert as a qualified statement (h, r, t, Q), where h and t are source and destination IPs, r denotes the attack type, and Q encodes flow-level metadata such as timestamps, ports, protocols, and attack intensity, going beyond standard KGC binary triples (h, r, t) that would discard this contextual richness. We introduce five models across three contributions: first, Hyper-relational Neural Bellman-Ford (HR-NBFNet) extends Neural Bellman-Ford Networks to the hyper-relational setting with qualifier-aware multi-hop path reasoning, while its multi-task variant MT-HR-NBFNet jointly predicts tail, relation, and qualifier-value within a single traversal pass; second, AlertStar fuses qualifier context and structural path information entirely in embedding space via cross-attention and learned path composition, and its multi-task extension MT-AlertStar eliminates the overhead of full knowledge graph propagation; third, HR-NBFNet-CQ extends qualifier-aware representations to answer complex first-order logic queries, including one-hop, two-hop chain, two-anchor intersection, and union, enabling multi-condition threat reasoning over the alert knowledge graph. Evaluated inductively on the Warden and UNSW-NB15 benchmarks across three qualifier-density regimes, AlertStar and MT-AlertStar achieve superior MR, MRR, and Hits@k, demonstrating that local qualifier fusion is both sufficient and more efficient than global path propagation for hyper-relational alert prediction.
Authors: Zhangyun Tan, Zeliang Zhang, Susan Liang, Yolo Yunlong Tang, Lisha Chen, Chenliang Xu
Abstract: VLMs trained on web-scale data retain sensitive and copyrighted visual concepts that deployment may require removing. Training-based unlearning methods share a structural flaw: fine-tuning on a narrow forget set degrades general capabilities before unlearning begins, making it impossible to attribute subsequent performance drops to the unlearning procedure itself. Training-free approaches sidestep this by suppressing concepts through prompts or system instructions, but no rigorous benchmark exists for evaluating them on visual tasks. We introduce VLM-UnBench, the first benchmark for training-free visual concept unlearning in VLMs. It covers four forgetting levels, 7 source datasets, and 11 concept axes, and pairs a three-level probe taxonomy with five evaluation conditions to separate genuine forgetting from instruction compliance. Across 8 evaluation settings and 13 VLM configurations, realistic unlearning prompts leave forget accuracy near the no-instruction baseline; meaningful reductions appear only under oracle conditions that disclose the target concept to the model. Object and scene concepts are the most resistant to suppression, and stronger instruction-tuned models remain capable despite explicit forget instructions. These results expose a clear gap between prompt-level suppression and true visual concept erasure.
Authors: Zheng-Xin Yong, Parv Mahajan, Andy Wang, Ida Caspary, Yernat Yestekov, Zora Che, Mosh Levy, Elle Najt, Dennis Murphy, Prashant Kulkarni, Lev McKinney, Kei Nishimura-Gasparian, Ram Potham, Aengus Lynch, Michael L. Chen
Abstract: Kimi K2.5 is an open-weight LLM that rivals closed models across coding, multimodal, and agentic benchmarks, but was released without an accompanying safety evaluation. In this work, we conduct a preliminary safety assessment of Kimi K2.5 focusing on risks likely to be exacerbated by powerful open-weight models. Specifically, we evaluate the model for CBRNE misuse risk, cybersecurity risk, misalignment, political censorship, bias, and harmlessness, in both agentic and non-agentic settings. We find that Kimi K2.5 shows similar dual-use capabilities to GPT 5.2 and Claude Opus 4.5, but with significantly fewer refusals on CBRNE-related requests, suggesting it may uplift malicious actors in weapon creation. On cyber-related tasks, we find that Kimi K2.5 demonstrates competitive cybersecurity performance, but it does not appear to possess frontier-level autonomous cyberoffensive capabilities such as vulnerability discovery and exploitation. We further find that Kimi K2.5 shows concerning levels of sabotage ability and self-replication propensity, although it does not appear to have long-term malicious goals. In addition, Kimi K2.5 exhibits narrow censorship and political bias, especially in Chinese, and is more compliant with harmful requests related to spreading disinformation and copyright infringement. Finally, we find the model refuses to engage in user delusions and generally has low over-refusal rates. While preliminary, our findings highlight how safety risks exist in frontier open-weight models and may be amplified by the scale and accessibility of open-weight releases. Therefore, we strongly urge open-weight model developers to conduct and release more systematic safety evaluations required for responsible deployment.
Authors: Jinsook Lee, Kirk Vanacore, Zhuqian Zhou, Bakhtawar Ahtisham, Rene F. Kizilcec
Abstract: Automated annotation of pedagogical dialogue is a high-stakes task where LLMs often fail without sufficient domain grounding. We present a domain-adapted RAG pipeline for tutoring move annotation. Rather than fine-tuning the generative model, we adapt retrieval by fine-tuning a lightweight embedding model on tutoring corpora and indexing dialogues at the utterance level to retrieve labeled few-shot demonstrations. Evaluated across two real tutoring dialogue datasets (TalkMoves and Eedi) and three LLM backbones (GPT-5.2, Claude Sonnet 4.6, Qwen3-32b), our best configuration achieves Cohen's $\kappa$ of 0.526-0.580 on TalkMoves and 0.659-0.743 on Eedi, substantially outperforming no-retrieval baselines ($\kappa = 0.275$-$0.413$ and $0.160$-$0.410$). An ablation study reveals that utterance-level indexing, rather than embedding quality alone, is the primary driver of these gains, with top-1 label match rates improving from 39.7\% to 62.0\% on TalkMoves and 52.9\% to 73.1\% on Eedi under domain-adapted retrieval. Retrieval also corrects systematic label biases present in zero-shot prompting and yields the largest improvements for rare and context-dependent labels. These findings suggest that adapting the retrieval component alone is a practical and effective path toward expert-level pedagogical dialogue annotation while keeping the generative model frozen.
Authors: Yuhang Wang, Haichang Gao, Zhenxing Niu, Zhaoxiang Liu, Wenjing Zhang, Xiang Wang, Shiguo Lian
Abstract: Tool-augmented AI agents substantially extend the practical capabilities of large language models, but they also introduce security risks that cannot be identified through model-only evaluation. In this paper, we present a systematic security assessment of six representative OpenClaw-series agent frameworks, namely OpenClaw, AutoClaw, QClaw, KimiClaw, MaxClaw, and ArkClaw, under multiple backbone models. To support this study, we construct a benchmark of 205 test cases covering representative attack behaviors across the full agent execution lifecycle, enabling unified evaluation of risk exposure at both the framework and model levels. Our results show that all evaluated agents exhibit substantial security vulnerabilities, and that agentized systems are significantly riskier than their underlying models used in isolation. In particular, reconnaissance and discovery behaviors emerge as the most common weaknesses, while different frameworks expose distinct high-risk profiles, including credential leakage, lateral movement, privilege escalation, and resource development. These findings indicate that the security of modern agent systems is shaped not only by the safety properties of the backbone model, but also by the coupling among model capability, tool use, multi-step planning, and runtime orchestration. We further show that once an agent is granted execution capability and persistent runtime context, weaknesses arising in early stages can be amplified into concrete system-level failures. Overall, our study highlights the need to move beyond prompt-level safeguards toward lifecycle-wide security governance for intelligent agent frameworks.
Authors: Ema Smolic, Mario Brcic, Luka Hobor, Mihael Kovac
Abstract: Many software systems originate as prototypes or minimum viable products (MVPs), developed with an emphasis on delivery speed and responsiveness to changing requirements rather than long-term code maintainability. While effective for rapid delivery, this approach can result in codebases that are difficult to modify, presenting a significant opportunity cost in the era of AI-assisted or even AI-led programming. In this paper, we present a case study of using coding models for automated unit test generation and subsequent safe refactoring, with proposed code changes validated by passing tests. The study examines best practices for iteratively generating tests to capture existing system behavior, followed by model-assisted refactoring under developer supervision. We describe how this workflow constrained refactoring changes, the errors and limitations observed in both phases, the efficiency gains achieved, when manual intervention was necessary, and how we addressed the weak value misalignment we observed in models. Using this approach, we generated nearly 16,000 lines of reliable unit tests in hours rather than weeks, achieved up to 78\% branch coverage in critical modules, and significantly reduced regression risk during large-scale refactoring. These results illustrate software engineering's shift toward an empirical science, emphasizing data collection and constraining mechanisms that support fast, safe iteration.
Authors: Jian Yang, Wei Zhang, Jiajun Wu, Junhang Cheng, Tuney Zheng, Fanglin Xu, Weicheng Gu, Lin Jing, Yaxin Du, Joseph Li, Yizhi Li, Yan Xing, Chuan Hao, Ran Tao, Ruihao Gong, Aishan Liu, Zhoujun Li, Mingjie Tang, Chenghua Lin, Siheng Chen, Wayne Xin Zhao, Xianglong Liu, Ming Zhou, Bryan Dai, Weifeng Lv
Abstract: Industrial software development across chip design, GPU optimization, and embedded systems lacks expert reasoning traces showing how engineers reason about hardware constraints and timing semantics. In this work, we propose InCoder-32B-Thinking, trained on the data from the Error-driven Chain-of-Thought (ECoT) synthesis framework with an industrial code world model (ICWM) to generate reasoning traces. Specifically, ECoT generates reasoning chains by synthesizing the thinking content from multi-turn dialogue with environmental error feedback, explicitly modeling the error-correction process. ICWM is trained on domain-specific execution traces from Verilog simulation, GPU profiling, etc., learns the causal dynamics of how code affects hardware behavior, and enables self-verification by predicting execution outcomes before actual compilation. All synthesized reasoning traces are validated through domain toolchains, creating training data matching the natural reasoning depth distribution of industrial tasks. Evaluation on 14 general (81.3% on LiveCodeBench v5) and 9 industrial benchmarks (84.0% in CAD-Coder and 38.0% on KernelBench) shows InCoder-32B-Thinking achieves top-tier open-source results across all domains.GPU Optimization
Authors: Lihao Sun, Lewen Yan, Xiaoya Lu, Andrew Lee, Jie Zhang, Jing Shao
Abstract: We present a method to identify a valence-arousal (VA) subspace within large language model representations. From 211k emotion-labeled texts, we derive emotion steering vectors, then learn VA axes as linear combinations of their top PCA components via ridge regression on the model's self-reported valence-arousal scores. The resulting VA subspace exhibits circular geometry consistent with established models of human emotion perception. Projections along our recovered VA subspace correlate with human-crowdsourced VA ratings across 44k lexical items. Furthermore, steering generation along these axes produces monotonic shifts in the corresponding affective dimensions of model outputs. Steering along these directions also induces near-monotonic bidirectional control over refusal and sycophancy: increasing arousal decreases refusal and increases sycophancy, and vice versa. These effects replicate across Llama-3.1-8B, Qwen3-8B, and Qwen3-14B, demonstrating cross-architecture generality. We provide a mechanistic account for these effects and prior emotionally-framed controls: refusal-associated tokens ("I can't," "sorry") occupy low-arousal, negative-valence regions, so VA steering directly modulates their emission probability.
Authors: Prakhar Bansal, Shivangi Agarwal
Abstract: Large language models (LLMs) encode vast world knowledge in their parameters, yet they remain fundamentally limited by static knowledge, finite context windows, and weakly structured causal reasoning. This survey provides a unified account of augmentation strategies along a single axis: the degree of structured context supplied at inference time. We cover in-context learning and prompt engineering, Retrieval-Augmented Generation (RAG), GraphRAG, and CausalRAG. Beyond conceptual comparison, we provide a transparent literature-screening protocol, a claim-audit framework, and a structured cross-paper evidence synthesis that distinguishes higher-confidence findings from emerging results. The paper concludes with a deployment-oriented decision framework and concrete research priorities for trustworthy retrieval-augmented NLP.
Authors: Gengwei Zhang, Jie Peng, Zhen Tan, Mufan Qiu, Hossein Nourkhiz Mahjoub, Vaishnav Tadiparthi, Kwonjoon Lee, Yanyong Zhang, Tianlong Chen
Abstract: The recent success of reinforcement learning (RL) in large reasoning models has inspired the growing adoption of RL for post-training Multimodal Large Language Models (MLLMs) to enhance their visual reasoning capabilities. Although many studies have reported improved performance, it remains unclear whether RL training truly enables models to learn from visual information. In this work, we propose the Hallucination-as-Cue Framework, an analytical framework designed to investigate the effects of RL-based post-training on multimodal reasoning models from the perspective of model hallucination. Specifically, we introduce hallucination-inductive, modality-specific corruptions that remove or replace essential information required to derive correct answers, thereby forcing the model to reason by hallucination. By applying these corruptions during both training and evaluation, our framework provides a unique perspective for diagnosing RL training dynamics and understanding the intrinsic properties of datasets. Through extensive experiments and analyses across multiple multimodal reasoning benchmarks, we reveal that the role of model hallucination for RL-training is more significant than previously recognized. For instance, we find that RL post-training under purely hallucination-inductive settings can still significantly improve models' reasoning performance, and in some cases even outperform standard training. These findings challenge prevailing assumptions about MLLM reasoning training and motivate the development of more modality-aware RL-based training designs.
Authors: Nikita Vassilyev, William Berrios, Ruowang Zhang, Bo Han, Douwe Kiela, Shikib Mehri
Abstract: Generally capable agents must learn from experience in ways that generalize across tasks and environments. The fundamental problems of learning, including credit assignment, overfitting, forgetting, local optima, and high-variance learning signals, persist whether the learned object lies in parameter space or context space. While these challenges are well understood in classical machine learning optimization, they remain underexplored in context space, leading current methods to be fragmented and ad hoc. We present Reflective Context Learning (RCL), a unified framework for agents that learn through repeated interaction, reflection on behavior and failure modes, and iterative updates to context. In RCL, reflection converts trajectories and current context into a directional update signal analogous to gradients, while mutation applies that signal to improve future behavior in context space. We recast recent context-optimization approaches as instances of this shared learning problem and systematically extend them with classical optimization primitives, including batching, improved credit-assignment signal, auxiliary losses, failure replay, and grouped rollouts for variance reduction. On AppWorld, BrowseComp+, and RewardBench2, these primitives improve over strong baselines, with their relative importance shifting across task regimes. We further analyze robustness to initialization, the effects of batch size, sampling and curriculum strategy, optimizer-state variants, and the impact of allocating stronger or weaker models to different optimization components. Our results suggest that learning through context updates should be treated not as a set of isolated algorithms, but as an optimization problem whose mechanisms can be studied systematically and improved through transferable principles.
Authors: Saleh Sargolzaei
Abstract: Transformer attention computes a single softmax-weighted average over values -- a one-pass estimate that cannot correct its own errors. We introduce \emph{gradient-boosted attention}, which applies the principle of gradient boosting \emph{within} a single attention layer: a second attention pass, with its own learned projections, attends to the prediction error of the first and applies a gated correction. Under a squared reconstruction objective, the construction maps onto Friedman's gradient boosting machine, with each attention pass as a base learner and the per-dimension gate as the shrinkage parameter. We show that a single Hopfield-style update erases all query information orthogonal to the stored-pattern subspace, and that further iteration under local contraction can collapse distinct queries in the same region to the same fixed point. We also show that separate projections for the correction pass can recover residual information inaccessible to the shared-projection approach of Tukey's twicing. On a 10M-token subset of WikiText-103, gradient-boosted attention achieves a test perplexity of $67.9$ compared to $72.2$ for standard attention, $69.6$ for Twicing Attention, and $69.0$ for a parameter-matched wider baseline, with two rounds capturing most of the benefit.
Authors: Dipto Sumit, Ankan Kumar Roy, Sadia Khair Rodela, Atia Haque Asha, Mourchona Afrin, Niloy Farhan, Farig Yousuf Sadeque
Abstract: We study multiteacher knowledge distillation for low resource abstractive summarization from a reliability aware perspective. We introduce EWAD (Entropy Weighted Agreement Aware Distillation), a token level mechanism that routes supervision between teacher distillation and gold supervision based on inter teacher agreement, and CPDP (Capacity Proportional Divergence Preservation), a geometric constraint on the student position relative to heterogeneous teachers. Across two Bangla datasets, 13 BanglaT5 ablations, and eight Qwen2.5 experiments, we find that logit level KD provides the most reliable gains, while more complex distillation improves semantic similarity for short summaries but degrades longer outputs. Cross lingual pseudo label KD across ten languages retains 71-122 percent of teacher ROUGE L at 3.2x compression. A human validated multi judge LLM evaluation further reveals calibration bias in single judge pipelines. Overall, our results show that reliability aware distillation helps characterize when multi teacher supervision improves summarization and when data scaling outweighs loss engineering.
Authors: Daniel C. MacRae, Luuk van der Hoek, Robert van der Wal, Suzanne P. M. de Vette, Hendrike Neh, Baoqiang Ma, Peter M. A. van Ooijen, Lisanne V. van Dijk
Abstract: Three-dimensional medical image data and computer-aided decision making, particularly using deep learning, are becoming increasingly important in the medical field. To aid in these developments we introduce PR3DICTR: Platform for Research in 3D Image Classification and sTandardised tRaining. Built using community-standard distributions (PyTorch and MONAI), PR3DICTR provides an open-access, flexible and convenient framework for prediction model development, with an explicit focus on classification using three-dimensional medical image data. By combining modular design principles and standardization, it aims to alleviate developmental burden whilst retaining adjustability. It provides users with a wealth of pre-established functionality, for instance in model architecture design options, hyper-parameter solutions and training methodologies, but still gives users the opportunity and freedom to ``plug in'' their own solutions or modules. PR3DICTR can be applied to any binary or event-based three-dimensional classification task and can work with as little as two lines of code.
Authors: Van Sy Mai, Kushal Chakrabarti, Richard J. La, Dipankar Maity
Abstract: This paper explores the use of server learning for enhancing the robustness of federated learning against malicious attacks even when clients' training data are not independent and identically distributed. We propose a heuristic algorithm that uses server learning and client update filtering in combination with geometric median aggregation. We demonstrate via experiments that this approach can achieve significant improvement in model accuracy even when the fraction of malicious clients is high, even more than $50\%$ in some cases, and the dataset utilized by the server is small and could be synthetic with its distribution not necessarily close to that of the clients' aggregated data.
Authors: Yuqi Wu, Guangya Wan, Jingjing Li, Shengming Zhao, Lingfeng Ma, Tianyi Ye, Ion Pop, Yanbo Zhang, Jie Chen
Abstract: Large Language Models (LLMs) offer promising opportunities to support mental healthcare workflows, yet they often lack the structured clinical reasoning needed for reliable diagnosis and may struggle to provide the emotionally attuned communication essential for patient trust. Here, we introduce WiseMind, a novel multi-agent framework inspired by the theory of Dialectical Behavior Therapy designed to facilitate psychiatric assessment. By integrating a "Reasonable Mind" Agent for evidence-based logic and an "Emotional Mind" Agent for empathetic communication, WiseMind effectively bridges the gap between instrumental accuracy and humanistic care. Our framework utilizes a Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5)-guided Structured Knowledge Graph to steer diagnostic inquiries, significantly reducing hallucinations compared to standard prompting methods. Using a combination of virtual standard patients, simulated interactions, and real human interaction datasets, we evaluate WiseMind across three common psychiatric conditions. WiseMind outperforms state-of-the-art LLM methods in both identifying critical diagnostic nodes and establishing accurate differential diagnoses. Across 1206 simulated conversations and 180 real user sessions, the system achieves 85.6% top-1 diagnostic accuracy, approaching reported diagnostic performance ranges of board-certified psychiatrists and surpassing knowledge-enhanced single-agent baselines by 15-54 percentage points. Expert review by psychiatrists further validates that WiseMind generates responses that are not only clinically sound but also psychologically supportive, demonstrating the feasibility of empathetic, reliable AI agents to conduct psychiatric assessments under appropriate human oversight.
Authors: Beidan Liu, Zhengqiu Zhu, Chen Gao, Tianle Pu, Yong Zhao, Wei Qi, Quanjun Yin
Abstract: Large Language Model (LLM)-based optimization has recently shown promise for autonomous problem solving, yet most approaches still cast LLMs as passive constraint checkers rather than proactive strategy designers, limiting their effectiveness on complex Constraint Optimization Problems (COPs). To address this, we present AutoCO, an end-to-end Automated Constraint Optimization method that tightly couples operations-research principles of constraint relaxation with LLM reasoning. A core innovation is a unified triple-representation that binds relaxation strategies, algorithmic principles, and executable codes. This design enables the LLM to synthesize, justify, and instantiate relaxation strategies that are both principled and executable. To navigate fragmented solution spaces, AutoCO employs a bidirectional global-local coevolution mechanism, synergistically coupling Monte Carlo Tree Search (MCTS) for global relaxation-trajectory exploration with Evolutionary Algorithms (EAs) for local solution intensification. This continuous exchange of priors and feedback explicitly balances diversification and intensification, thus preventing premature convergence. Extensive experiments on three challenging COP benchmarks validate AutoCO's consistent effectiveness and superior performance, especially in hard regimes where current methods degrade. Results highlight AutoCO as a principled and effective path toward proactive, verifiable LLM-driven optimization.
Authors: Pouya Hamadanian, Pantea Karimi, Arash Nasr-Esfahany, Kimia Noorbakhsh, Joseph Chandler, Ali ParandehGheibi, Mohammad Alizadeh, Hari Balakrishnan
Abstract: Can AI autonomously design mechanisms for computer systems on par with the creativity and reasoning of human experts? We present Glia, an AI architecture for networked systems design that uses large language models (LLMs) in a human-inspired multi-agent workflow. Each agent specializes in reasoning, experimentation, and analysis, collaborating through an evaluation framework that grounds abstract reasoning in empirical feedback. Unlike prior ML-for-systems methods that optimize black-box policies, Glia generates interpretable designs and exposes its reasoning. When applied to a distributed GPU cluster for LLM inference, it produces new algorithms for request routing, scheduling, and auto-scaling that perform at human-expert levels in significantly less time, while yielding novel insights into workload behavior. Our results suggest that combining reasoning LLMs with structured experimentation, an AI can produce creative and understandable designs for complex systems problems.
Authors: Jiayu Liu, Cheng Qian, Zhaochen Su, Qing Zong, Shijue Huang, Bingxiang He, Yi R. Fung
Abstract: Current evaluations of Large Language Model (LLM) agents primarily emphasize task completion, often overlooking resource efficiency and adaptability. This neglects a crucial capability: agents' ability to devise and adjust cost-optimal plans in response to changing environments. To bridge this gap, we introduce CostBench, a scalable, cost-centric benchmark designed to evaluate agents' economic reasoning and replanning abilities. Situated in the travel-planning domain, CostBench comprises tasks solvable via multiple sequences of atomic and composite tools with diverse, customizable costs. It also supports four types of dynamic blocking events, such as tool failures and cost changes, to simulate real-world unpredictability and necessitate agents to adapt in real time. Evaluating leading open-sourced and proprietary models on CostBench reveals a substantial gap in cost-aware planning: agents frequently fail to identify cost-optimal solutions in static settings, with even GPT-5 achieving less than 75% exact match rate on the hardest tasks, and performance further dropping by around 40% under dynamic conditions. By diagnosing these weaknesses, CostBench lays the groundwork for developing future agents that are both economically rational and robust.
Authors: Fanrui Zhang, Qiang Zhang, Sizhuo Zhou, Jianwen Sun, Chuanhao Li, Jiaxin Ai, Yukang Feng, Yujie Zhang, Wenjie Li, Zizhen Li, Yifan Chang, Jiawei Liu, Kaipeng Zhang
Abstract: Existing image forgery detection (IFD) methods either exploit low-level, semantics-agnostic artifacts or rely on multimodal large language models (MLLMs) with high-level semantic knowledge. Although naturally complementary, these two information streams are highly heterogeneous in both paradigm and reasoning, making it difficult for existing methods to unify them or effectively model their cross-level interactions. To address this gap, we propose ForenAgent, a multi-round interactive IFD framework that enables MLLMs to autonomously generate, execute, and iteratively refine Python-based low-level tools around the detection objective, thereby achieving more flexible and interpretable forgery analysis. ForenAgent follows a two-stage training pipeline combining Cold Start and Reinforcement Fine-Tuning to enhance its tool interaction capability and reasoning adaptability progressively. Inspired by human reasoning, we design a dynamic reasoning loop comprising global perception, local focusing, iterative probing, and holistic adjudication, and instantiate it as both a data-sampling strategy and a task-aligned process reward. For systematic training and evaluation, we construct FABench, a heterogeneous, high-quality agent-forensics dataset comprising 100k images and approximately 200k agent-interaction question-answer pairs. Experiments show that ForenAgent exhibits emergent tool-use competence and reflective reasoning on challenging IFD tasks when assisted by low-level tools, charting a promising route toward general-purpose IFD. The code will be released after the review process is completed.
Authors: Sixue Xing, Kerui Wu, Xuanye Xia, Meng Jiang, Jintai Chen, Tianfan Fu
Abstract: Clinical trials constitute a critical yet exceptionally challenging and costly stage of drug development (\$2.6B per drug), where protocols are encoded as complex natural language documents, motivating the use of AI systems beyond manual analysis. Existing AI methods accurately predict trial failure, but do not provide actionable remedies. To fill this gap, this paper proposes ClinicalReTrial, a multi-agent system that formulates clinical trial optimization as an iterative redesign problem on textural protocols. Our method integrates failure diagnosis, safety-aware modifications, and candidate evaluation in a closed-loop, reward-driven optimization framework. Serving the outcome prediction model as a simulation environment, ClinicalReTrial enables low-cost evaluation and dense reward signals for continuous self-improvement. We further propose a hierarchical memory that captures iteration-level feedback within trials and distills transferable redesign patterns across trials. Empirically, ClinicalReTrial improves $83.3\%$ of trial protocols with a mean success probability gain of $5.7\%$ with negligible cost (\$0.12 per trial). Retrospective case studies demonstrate alignment between the discovered redesign strategies and real-world clinical trial modifications. The code is anonymously available at: https://github.com/xingsixue123/ClinicalFailureReasonReTrial.
URLs: https://github.com/xingsixue123/ClinicalFailureReasonReTrial.
Authors: Jiayi Yuan, Jonathan N\"other, Natasha Jaques, Goran Radanovi\'c
Abstract: While recent automated red-teaming methods show promise for systematically exposing model vulnerabilities, most existing approaches rely on human-specified workflows. This dependence on manually designed workflows suffers from human biases and makes exploring the broader design space expensive. We introduce AgenticRed, an automated pipeline that leverages LLMs' in-context learning to iteratively design and refine red-teaming systems without human intervention. Rather than optimizing attacker policies within predefined structures, AgenticRed treats red-teaming as a system design problem, and it autonomously evolves automated red-teaming systems using evolutionary selection and generational knowledge. Red-teaming systems designed by AgenticRed consistently outperform state-of-the-art approaches, achieving 96% attack success rate (ASR) on Llama-2-7B, 98% on Llama-3-8B and 100% on Qwen3-8B on HarmBench. Our approach generates robust, query-agnostic red-teaming systems that transfer strongly to the latest proprietary models, achieving an impressive 100% ASR on GPT-5.1, DeepSeek-R1 and DeepSeek V3.2. This work highlights evolutionary algorithms as a powerful approach to AI safety that can keep pace with rapidly evolving models.
Authors: Bowen Cao, Dongdong Zhang, Yixia Li, Junpeng Liu, Shijue Huang, Chufan Shi, Hongyuan Lu, Yaokang Wu, Guanhua Chen, Wai Lam, Furu Wei
Abstract: Large language models now solve many benchmark math problems at near-expert levels, yet this progress has not fully translated into reliable performance in real-world applications. We study this gap through contextual mathematical reasoning, where the mathematical core must be formulated from descriptive scenarios. We introduce ContextMATH, a benchmark that repurposes AIME and MATH-500 problems into two contextual settings: Scenario Grounding (SG), which embeds abstract problems into realistic narratives without increasing reasoning complexity, and Complexity Scaling (CS), which transforms explicit conditions into sub-problems to capture how constraints often appear in practice. Evaluating 61 proprietary and open-source models, we observe sharp drops: on average, open-source models decline by 13 and 34 points on SG and CS, while proprietary models drop by 13 and 20. Error analysis shows that errors are dominated by incorrect problem formulation, with formulation accuracy declining as original problem difficulty increases. Correct formulation emerges as a prerequisite for success, and its sufficiency improves with model scale, indicating that larger models advance in both understanding and reasoning. Nevertheless, formulation and reasoning remain two complementary bottlenecks that limit contextual mathematical problem solving. Finally, we find that fine-tuning with scenario data improves performance, whereas formulation-only training is ineffective. However, performance gaps are only partially alleviated, highlighting contextual mathematical reasoning as a central unsolved challenge for LLMs.
Authors: A. Humnabadkar, A. Sikdar, B. Cave, H. Zhang, N. Bessis, A. Behera
Abstract: Autonomous driving technologies have achieved significant advances in recent years, yet their real-world deployment remains constrained by data scarcity, safety requirements, and the need for generalization across diverse environments. In response, synthetic data and virtual environments have emerged as powerful enablers, offering scalable, controllable, and richly annotated scenarios for training and evaluation. This survey presents a comprehensive review of recent developments at the intersection of autonomous driving, simulation technologies, and synthetic datasets. We organize the landscape across three core dimensions: (i) the use of synthetic data for perception and planning, (ii) digital twin-based simulation for system validation, and (iii) domain adaptation strategies bridging synthetic and real-world data. We also highlight the role of vision-language models and simulation realism in enhancing scene understanding and generalization. A detailed taxonomy of datasets, tools, and simulation platforms is provided, alongside an analysis of trends in benchmark design. Finally, we discuss critical challenges and open research directions, including Sim2Real transfer, scalable safety validation, cooperative autonomy, and simulation-driven policy learning, that must be addressed to accelerate the path toward safe, generalizable, and globally deployable autonomous driving systems.
Authors: Yang Li, Yule Liu, Xinlei He, Youjian Zhao, Qi Li, Ke Xu
Abstract: Although Large Language Models (LLMs) have evolved from text generators into the cognitive core of modern AI systems, their inherent lack of authorization awareness exposes these systems to catastrophic risks, ranging from unintentional data leakage to unauthorized command execution. Existing defense mechanisms are fundamentally decoupled from internal reasoning, rendering them insufficient for the complex security demands of dynamic AI systems. Here, we propose the Chain-of-Authorization (CoA) framework, a paradigm that internalizes access control as a foundational cognitive capability. By systematically redesigning the input-output format and fine-tuning the model on synthesized data with complex permission topologies, CoA forces the model to generate a structured authorization trajectory as a causal prerequisite for any substantive response or action, thereby enabling LLMs to internalize access boundaries within dynamic reasoning environments. CoA maintains high utility in authorized scenarios while achieving high rejection rates of unauthorized prompts and robust defense against diverse adversarial attacks. By embedding authorization directly into the reasoning process, CoA provides a principled architectural blueprint for deploying secure LLMs as the cognitive cores of modern AI systems.
Authors: Canfer Akbulut, Rasmi Elasmar, Abhishek Roy, Anthony Payne, Priyanka Suresh, Lujain Ibrahim, Seliem El-Sayed, Charvi Rastogi, Ashyana Kachra, Will Hawkins, Kristian Lum, Laura Weidinger
Abstract: Interest in the concept of AI-driven harmful manipulation is growing, yet current approaches to evaluating it are limited. This paper introduces a framework for evaluating harmful AI manipulation via context-specific human-AI interaction studies. We illustrate the utility of this framework by assessing an AI model with 10,101 participants spanning interactions in three AI use domains (public policy, finance, and health) and three locales (US, UK, and India). Overall, we find that that the tested model can produce manipulative behaviours when prompted to do so and, in experimental settings, is able to induce belief and behaviour changes in study participants. We further find that context matters: AI manipulation differs between domains, suggesting that it needs to be evaluated in the high-stakes context(s) in which an AI system is likely to be used. We also identify significant differences across our tested geographies, suggesting that AI manipulation results from one geographic region may not generalise to others. Finally, we find that the frequency of manipulative behaviours (propensity) of an AI model is not consistently predictive of the likelihood of manipulative success (efficacy), underscoring the importance of studying these dimensions separately. To facilitate adoption of our evaluation framework, we detail our testing protocols and make relevant materials publicly available. We conclude by discussing open challenges in evaluating harmful manipulation by AI models.
Authors: Zelin Tan, Zhouliang Yu, Bohan Lin, Zijie Geng, Hejia Geng, Yudong Zhang, Mulei Zhang, Yang Chen, Shuyue Hu, Zhenfei Yin, Chen Zhang, Lei Bai
Abstract: We propose Process-Aware Policy Optimization (PAPO), a method that integrates process-level evaluation into Group Relative Policy Optimization (GRPO) through decoupled advantage normalization, to address two limitations of existing reward designs. Outcome reward models (ORM) evaluate only final-answer correctness, treating all correct responses identically regardless of reasoning quality, and gradually lose the advantage signal as groups become uniformly correct. Process reward models (PRM) offer richer supervision, but directly using PRM scores causes reward hacking, where models exploit verbosity to inflate scores while accuracy collapses. PAPO resolves both by composing the advantage from an outcome component Aout, derived from ORM and normalized over all responses, and a process component Aproc, derived from a rubric-based PRM and normalized exclusively among correct responses. This decoupled design ensures that Aout anchors training on correctness while Aproc differentiates reasoning quality without distorting the outcome signal. Experiments across multiple model scales and six benchmarks demonstrate that PAPO consistently outperforms ORM, reaching 51.3% vs.\ 46.3% on OlympiadBench while continuing to improve as ORM plateaus and declines.
Authors: Sha Li, Naren Ramakrishnan
Abstract: Multi-agent Retrieval-Augmented Generation (RAG), wherein each agent takes on a specific role, supports hard queries that require multiple steps and sources, or complex reasoning. Existing approaches, however, rely on static agent behaviors and fixed orchestration strategies, leading to brittle performance on diverse, multi-hop tasks. We identify two key limitations: the lack of continuously adaptive orchestration mechanisms and the absence of behavior-level learning for individual agents. To this end, we propose HERA, a hierarchical framework that jointly evolves multi-agent orchestration and role-specific agent prompts. At the global level, HERA optimizes query-specific agent topologies through reward-guided sampling and experience accumulation. At the local level, Role-Aware Prompt Evolution refines agent behaviors via credit assignment and dual-axes adaptation along operational and behavioral principles, enabling targeted, role-conditioned improvements. On six knowledge-intensive benchmarks, HERA achieves an average improvement of 38.69\% over recent baselines while maintaining robust generalization and token efficiency. Topological analyses reveal emergent self-organization, where sparse exploration yields compact, high-utility multi-agent networks, demonstrating both efficient coordination and robust reasoning.
Authors: Esakkivel Esakkiraja, Sai Rajeswar, Denis Akhiyarov, Rajagopal Venkatesaramani
Abstract: We consider the question: when a large language reasoning model makes a choice, did it think first and then decide to, or decide first and then think? In this paper, we present evidence that detectable, early-encoded decisions shape chain-of-thought in reasoning models. Specifically, we show that a simple linear probe successfully decodes tool-calling decisions from pre-generation activations with very high confidence, and in some cases, even before a single reasoning token is produced. Activation steering supports this causally: perturbing the decision direction leads to inflated deliberation, and flips behavior in many examples (between 7 - 79% depending on model and benchmark). We also show through behavioral analysis that, when steering changes the decision, the chain-of-thought process often rationalizes the flip rather than resisting it. Together, these results suggest that reasoning models can encode action choices before they begin to deliberate in text.
Authors: Khalid Adnan Alsayed
Abstract: Artificial intelligence (AI) systems are increasingly integrated into healthcare and pharmacy workflows, supporting tasks such as medication recommendations, dosage determination, and drug interaction detection. While these systems often demonstrate strong performance under standard evaluation metrics, their reliability in real-world decision-making remains insufficiently understood. In high-risk domains such as medication management, even a single incorrect recommendation can result in severe patient harm. This paper examines the reliability of AI-assisted medication systems by focusing on system failures and their potential clinical consequences. Rather than evaluating performance solely through aggregate metrics, this work shifts attention towards how errors occur and what happens when AI systems produce incorrect outputs. Through a series of controlled, simulated scenarios involving drug interactions and dosage decisions, we analyse different types of system failures, including missed interactions, incorrect risk flagging, and inappropriate dosage recommendations. The findings highlight that AI errors in medication-related contexts can lead to adverse drug reactions, ineffective treatment, or delayed care, particularly when systems are used without sufficient human oversight. Furthermore, the paper discusses the risks of over-reliance on AI recommendations and the challenges posed by limited transparency in decision-making processes. This work contributes a reliability-focused perspective on AI evaluation in healthcare, emphasising the importance of understanding failure behavior and real-world impact. It highlights the need to complement traditional performance metrics with risk-aware evaluation approaches, particularly in safety-critical domains such as pharmacy practice.
Authors: Yash Shah, Abhijit Chakraborty, Naresh Kumar Devulapally, Vishnu Lokhande, Vivek Gupta
Abstract: Diffusion language models (DLMs) expose their denoising trajectories, offering a natural handle for inference-time control; accordingly, an ideal hallucination mitigation framework should intervene during generation using this model-native signal rather than relying on an externally trained hallucination classifier. Toward this, we formulate commitment uncertainty localization: given a denoising trajectory, identify token positions whose cross-chain entropy exceeds an unsupervised threshold before factually unreliable commitments propagate into self-consistent but incorrect outputs. We introduce a suite of trajectory-level assessments, including a cross-chain divergence-at-hallucination (CDH) metric, for principled comparison of localization methods. We also introduce OSCAR, a training-free inference-time framework operationalizing this formulation. OSCAR runs N parallel denoising chains with randomized reveal orders, computes cross-chain Shannon entropy to detect high-uncertainty positions, and then performs targeted remasking conditioned on retrieved evidence. Ablations confirm that localization and correction contribute complementary gains, robust across N in {4, 8, 16}. On TriviaQA, HotpotQA, RAGTruth, and CommonsenseQA using LLaDA-8B and Dream-7B, OSCAR enhances generation quality by significantly reducing hallucinated content and improving factual accuracy through uncertainty-guided remasking, which also facilitates more effective integration of retrieved evidence. Its native entropy-based uncertainty signal surpasses that of specialized trained detectors, highlighting an inherent capacity of diffusion language models to identify factual uncertainty that is not present in the sequential token commitment structure of autoregressive models.
Authors: Tuyen Van Kieu, Chi Linh Hoang, Khanh Van To
Abstract: Cutting rectangular items from stock sheets to satisfy demands while minimizing waste is a central manufacturing task. The Two-Dimensional Single Stock Size Cutting Stock Problem (2D-CSSP) generalizes bin packing by requiring multiple copies of each item type, which causes a strong combinatorial blow-up. We present a SAT-based framework where item types are expanded by demand, each copy has a sheet-assignment variable and non-overlap constraints are activated only for copies assigned to the same sheet. We also introduce an infeasible-orientation elimination rule that fixes rotation variables when only one orientation can fit the sheet. For minimizing the number of sheets, we compare three approaches: non-incremental SAT with binary search, incremental SAT with clause reuse across iterations and weighted partial MaxSAT. On the Cui--Zhao benchmark suite, our best SAT configurations certify two to three times more instances as provably optimal and achieve lower optimality gaps than OR-Tools, CPLEX and Gurobi. The relative ranking among SAT approaches depends on rotation: incremental SAT is strongest without rotation, while non-incremental SAT is more effective when rotation increases formula size.
Authors: Sarath Shekkizhar, Romain Cosentino, Adam Earle
Abstract: Standard LLM benchmarks evaluate the assistant turn: the model generates a response to an input, a verifier scores correctness, and the analysis ends. This paradigm leaves unmeasured whether the LLM encodes any awareness of what follows the assistant response. We propose user-turn generation as a probe of this gap: given a conversation context of user query and assistant response, we let a model generate under the user role. If the model's weights encode interaction awareness, the generated user turn will be a grounded follow-up that reacts to the preceding context. Through experiments across $11$ open-weight LLMs (Qwen3.5, gpt-oss, GLM) and $5$ datasets (math reasoning, instruction following, conversation), we show that interaction awareness is decoupled from task accuracy. In particular, within the Qwen3.5 family, GSM8K accuracy scales from $41\%$ ($0.8$B) to $96.8\%$ ($397$B-A$17$B), yet genuine follow-up rates under deterministic generation remain near zero. In contrast, higher temperature sampling reveals interaction awareness is latent with follow up rates reaching $22\%$. Controlled perturbations validate that the proposed probe measures a real property of the model, and collaboration-oriented post-training on Qwen3.5-2B demonstrates an increase in follow-up rates. Our results show that user-turn generation captures a dimension of LLM behavior, interaction awareness, that is unexplored and invisible with current assistant-only benchmarks.
Authors: Thomas Jiralerspong, Xiaoyin Chen, Yash More, Vedant Shah, Yoshua Bengio
Abstract: We propose a novel framework that leverages LLMs for full causal graph discovery. While previous LLM-based methods have used a pairwise query approach, this requires a quadratic number of queries which quickly becomes impractical for larger causal graphs. In contrast, the proposed framework uses a breadth-first search (BFS) approach which allows it to use only a linear number of queries. We also show that the proposed method can easily incorporate observational data when available, to improve performance. In addition to being more time and data-efficient, the proposed framework achieves state-of-the-art results on real-world causal graphs of varying sizes. The results demonstrate the effectiveness and efficiency of the proposed method in discovering causal relationships, showcasing its potential for broad applicability in causal graph discovery tasks across different domains.
Authors: Haoyu Wang, Chunyu Qiang, Tianrui Wang, Cheng Gong, Yu Jiang, Yuheng Lu, Chen Zhang, Longbiao Wang, Jianwu Dang
Abstract: Recent advancements in speech synthesis have enabled large language model (LLM)-based systems to perform zero-shot generation with controllable content, timbre, speaker identity, and emotion through input prompts. As a result, these models heavily rely on prompt design to guide the generation process. However, existing prompt selection methods often fail to ensure that prompts contain sufficiently stable speaker identity cues and appropriate emotional intensity indicators, which are crucial for expressive speech synthesis. To address this challenge, we propose a two-stage prompt selection strategy specifically designed for expressive speech synthesis. In the static stage (before synthesis), we first evaluate prompt candidates using pitch-based prosodic features, perceptual audio quality, and text-emotion coherence scores evaluated by an LLM. We further assess the candidates under a specific TTS model by measuring character error rate, speaker similarity, and emotional similarity between the synthesized and prompt speech. In the dynamic stage (during synthesis), we use a textual similarity model to select the prompt that is most aligned with the current input text. Experimental results demonstrate that our strategy effectively selects prompt to synthesize speech with both high-intensity emotional expression and robust speaker identity, leading to more expressive and stable zero-shot TTS performance. Audio samples and codes will be available at https://whyrrrrun.github.io/ExpPro.github.io/.
Authors: Jiawei Liu, Fanrui Zhang, Jiaying Zhu, Esther Sun, Dong Li, Qiang Zhang, Zheng-Jun Zha
Abstract: Multimodal Large Language Models (MLLMs), such as GPT4o, have shown strong capabilities in visual reasoning and explanation generation. However, despite these strengths, they face significant challenges in the increasingly critical task of Image Forgery Detection and Localization (IFDL). Moreover, existing IFDL methods are typically limited to the learning of low-level semantic-agnostic clues and merely provide a single outcome judgment. To tackle these issues, we propose ForgeryGPT, a novel framework that advances the IFDL task by capturing high-order forensics knowledge correlations of forged images from diverse linguistic feature spaces, while enabling explainable generation and interactive dialogue through a newly customized Large Language Model (LLM) architecture. Specifically, ForgeryGPT enhances traditional LLMs by integrating the Mask-Aware Forgery Extractor, which enables the excavating of precise forgery mask information from input images and facilitating pixel-level understanding of tampering artifacts. The Mask-Aware Forgery Extractor consists of a Forgery Localization Expert (FL-Expert) and a Mask Encoder, where the FL-Expert is augmented with an Object-agnostic Forgery Prompt and a Vocabulary-enhanced Vision Encoder, allowing for effectively capturing of multi-scale fine-grained forgery details. To enhance its performance, we implement a three-stage training strategy, supported by our designed Mask-Text Alignment and IFDL Task-Specific Instruction Tuning datasets, which align vision-language modalities and improve forgery detection and instruction-following capabilities. Extensive experiments demonstrate the effectiveness of the proposed method.
Authors: Yongxiang Liu, Bowen Peng, Li Liu, Xiang Li
Abstract: Transferable Targeted Attacks (TTAs) face significant challenges due to severe overfitting to surrogate models. Recent breakthroughs heavily rely on large-scale training data of victim models, while data-free solutions, \textit{i.e.}, image transformation-involved gradient optimization, often depend on black-box feedback for method design and tuning. These dependencies violate black-box transfer settings and compromise threat evaluation fairness. In this paper, we propose two blind estimation measures, self-alignment and self-transferability, to analyze per-transformation effectiveness and cross-transformation correlations under strict black-box constraints. Our findings challenge conventional assumptions: (1) Attacking simple scaling transformations uniquely enhances targeted transferability, outperforming other basic transformations and rivaling leading complex methods; (2) Geometric and color transformations exhibit high internal redundancy despite weak inter-category correlations. These insights drive the design and tuning of S$^4$ST (Strong, Self-transferable, faSt, Simple Scale Transformation), which integrates dimensionally consistent scaling, complementary low-redundancy transformations, and block-wise operations. Extensive evaluations across diverse architectures, training distributions, and tasks show that S$^{4}$ST achieves state-of-the-art effectiveness-efficiency balance without data dependency. We reveal that scaling's effectiveness stems from visual data's multi-scale nature and ubiquitous scale augmentation during training, rendering such augmentation a double-edged sword. Further validations on medical imaging and face verification confirm the framework's strong generalization.
Authors: Shin'ya Yamaguchi, Kosuke Nishida, Daiki Chijiwa, Yasutoshi Ida
Abstract: Concept bottleneck models (CBMs) are inherently interpretable and intervenable neural network models, which explain their final label prediction by the intermediate prediction of high-level semantic concepts. However, they require target task training to learn input-to-concept and concept-to-label mappings, incurring target dataset collections and training resources. In this paper, we present zero-shot concept bottleneck models (Z-CBMs), which predict concepts and labels in a fully zero-shot manner without training neural networks. Z-CBMs utilize a large-scale concept bank, which is composed of millions of vocabulary extracted from the web, to describe arbitrary input in various domains. For the input-to-concept mapping, we introduce concept retrieval, which dynamically finds input-related concepts by the cross-modal search on the concept bank. In the concept-to-label inference, we apply concept regression to select essential concepts from the retrieved concepts by sparse linear regression. Through extensive experiments, we confirm that our Z-CBMs provide interpretable and intervenable concepts without any additional training. Code will be available at https://github.com/yshinya6/zcbm.
Authors: Minkyu Choi, S P Sharan, Harsh Goel, Sahil Shah, Sandeep Chinchali
Abstract: Current text-to-video (T2V) generation models are increasingly popular due to their ability to produce coherent videos from textual prompts. However, these models often struggle to generate semantically and temporally consistent videos when dealing with longer, more complex prompts involving multiple objects or sequential events. Additionally, the high computational costs associated with training or fine-tuning make direct improvements impractical. To overcome these limitations, we introduce NeuS-E, a novel zero-training video refinement pipeline that leverages neuro-symbolic feedback to automatically enhance video generation, achieving superior alignment with the prompts. Our approach first derives the neuro-symbolic feedback by analyzing a formal video representation and pinpoints semantically inconsistent events, objects, and their corresponding frames. This feedback then guides targeted edits to the original video. Extensive empirical evaluations on both open-source and proprietary T2V models demonstrate that NeuS-E significantly enhances temporal and logical alignment across diverse prompts by almost 40%
Authors: Tianyou Li, Haijun Zou, Jiayuan Wu, Zaiwen Wen
Abstract: Routing problems are canonical combinatorial optimization tasks with wide-ranging applications in logistics, transportation, and supply chain management. However, solving these problems becomes significantly more challenging when complex constraints are involved. In this paper, we propose LMask, a novel learning framework that utilizes dynamic masking to generate high-quality feasible solutions for constrained routing problems. LMask introduces the LazyMask decoding method, which lazily refines feasibility masks with the backtracking mechanism. In addition, it employs the refinement intensity embedding to encode the search trace into the model, mitigating representation ambiguities induced by backtracking. To further reduce sampling cost, LMask sets a backtracking budget during decoding, while constraint violations are penalized in the loss function during training to counteract infeasibility caused by this budget. We provide theoretical guarantees for the validity and probabilistic optimality of our approach. Extensive experiments on the traveling salesman problem with time windows (TSPTW) and TSP with draft limits (TSPDL) demonstrate that LMask achieves state-of-the-art feasibility rates and solution quality, outperforming existing neural methods.
Authors: Jialin Yang, Dongfu Jiang, Lipeng He, Sherman Siu, Yuxuan Zhang, Disen Liao, Zhuofeng Li, Huaye Zeng, Yiming Jia, Haozhe Wang, Benjamin Schneider, Chi Ruan, Wentao Ma, Zhiheng Lyu, Yifei Wang, Yi Lu, Quy Duc Do, Ziyan Jiang, Ping Nie, Wenhu Chen
Abstract: As Large Language Models (LLMs) become integral to software development workflows, their ability to generate structured outputs has become critically important. We introduce StructEval, a comprehensive benchmark for evaluating LLMs' capabilities in producing both non-renderable (JSON, YAML, CSV) and renderable (HTML, React, SVG) structured formats. Unlike prior benchmarks, StructEval systematically evaluates structural fidelity across diverse formats through two paradigms: 1) generation tasks, producing structured output from natural language prompts, and \textbf{2)} conversion tasks, translating between structured formats. Our benchmark encompasses 18 formats and 44 types of task, with novel metrics for format adherence and structural correctness. Results reveal significant performance gaps-even state-of-the-art models like o1-mini achieve only 75.58 average score, with open-source alternatives lagging approximately 10 points behind. We find generation tasks more challenging than conversion tasks, and producing correct visual content more difficult than generating text-only structures.
Authors: Hao Yin, Lijun Gu, Paritosh Parmar, Lin Xu, Tianxiao Guo, Xiujin Liu, Weiwei Fu, Yang Zhang, Tianyou Zheng
Abstract: Action Quality Assessment (AQA) -- the task of quantifying how well an action is performed -- has great potential for detecting errors in gym weight training, where accurate feedback is critical to prevent injuries and maximize gains. Existing AQA datasets, however, are limited to single-view competitive sports and RGB video, lacking multimodal signals and professional assessment of fitness actions. We introduce FLEX, the first large-scale, multimodal, multiview dataset for fitness AQA that incorporates surface electromyography (sEMG). FLEX contains over 7,500 multiview recordings of 20 weight-loaded exercises performed by 38 subjects of diverse skill levels, with synchronized RGB video, 3D pose, sEMG, and physiological signals. Expert annotations are organized into a Fitness Knowledge Graph (FKG) linking actions, key steps, error types, and feedback, supporting a compositional scoring function for interpretable quality assessment. FLEX enables multimodal fusion, cross-modal prediction -- including the novel Video$\rightarrow$EMG task -- and biomechanically oriented representation learning. Building on the FKG, we further introduce FLEX-VideoQA, a structured question-answering benchmark with hierarchical queries that drive cross-modal reasoning in vision-language models. Baseline experiments demonstrate that multimodal inputs, multiview video, and fine-grained annotations significantly enhance AQA performance. FLEX thus advances AQA toward richer multimodal settings and provides a foundation for AI-powered fitness assessment and coaching. Dataset and code are available at \href{https://github.com/HaoYin116/FLEX}{https://github.com/HaoYin116/FLEX}. Link to Project \href{https://haoyin116.github.io/FLEX_Dataset}{page}.
URLs: https://github.com/HaoYin116/FLEX, https://github.com/HaoYin116/FLEX, https://haoyin116.github.io/FLEX_Dataset
Authors: Xingzhong Fan, Hongming Tang, Yue Zeng, M. B. N. Kouwenhoven, Guangquan Zeng
Abstract: Conventional galaxy generation methods rely on semi-analytical models and hydrodynamic simulations, which are highly dependent on physical assumptions and parameter tuning. In contrast, data-driven generative models do not have explicit physical parameters pre-determined, and instead learn them efficiently from observational data, making them alternative solutions to galaxy generation. Among these, diffusion models outperform Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) in quality and diversity. Leveraging physical prior knowledge to these models can further enhance their capabilities. In this work, we present GalCatDiff, the first framework in astronomy to leverage both galaxy image features and astrophysical properties in the network design of diffusion models. GalCatDiff incorporates an enhanced U-Net and a novel block entitled Astro-RAB (Residual Attention Block), which dynamically combines attention mechanisms with convolution operations to ensure global consistency and local feature fidelity. Moreover, GalCatDiff uses category embeddings for class-specific galaxy generation, avoiding the high computational costs of training separate models for each category. Our experimental results demonstrate that GalCatDiff significantly outperforms existing methods in terms of the consistency of sample color and size distributions, and the generated galaxies are both visually realistic and physically consistent. This framework will enhance the reliability of galaxy simulations and can potentially serve as a data augmentor to support future galaxy classification algorithm development.
Authors: Vyacheslav Kungurtsev, Monicah Cherop Naibei, Gustav Sir, Akhil Anand, Sebastien Gros, Haozhe Tian, Homayoun Hamedmoghadam
Abstract: Research, innovation and practical capital investment have been increasing rapidly toward the realization of autonomous physical agents. This includes industrial and service robots, unmanned aerial vehicles, embedded control devices, and a number of other realizations of cybernetic/mechatronic implementations of intelligent autonomous devices. In this paper, we consider a stylized version of robotic care, which would normally involve a two-level Reinforcement Learning procedure that trains a policy for both lower level physical movement decisions as well as higher level conceptual tasks and their sub-components. In order to deliver greater safety and reliability in the system, we present the general formulation of this as a two-level optimization scheme which incorporates control at the lower level, and classical planning at the higher level, integrated with a capacity for learning. This synergistic integration of multiple methodologies -- control, classical planning, and RL -- presents an opportunity for greater insight for algorithm development, leading to more efficient and reliable performance. Here, the notion of reliability pertains to physical safety and interpretability into an otherwise black box operation of autonomous agents, concerning users and regulators. This work presents the necessary background and general formulation of the optimization framework, detailing each component and its integration with the others.
Authors: Shaoan Xie, Lingjing Kong, Yujia Zheng, Yu Yao, Zeyu Tang, Eric P. Xing, Guangyi Chen, Kun Zhang
Abstract: Contrastive Language-Image Pre-training (CLIP)~\citep{radford2021learning} has emerged as a pivotal model in computer vision and multimodal learning, achieving state-of-the-art performance at aligning visual and textual representations through contrastive learning. However, CLIP struggles with potential information misalignment in many image-text datasets and suffers from entangled representation. On the one hand, short captions for a single image in datasets like MSCOCO may describe disjoint regions in the image, leaving the model uncertain about which visual features to retain or disregard. On the other hand, directly aligning long captions with images can lead to the retention of entangled details, preventing the model from learning disentangled, atomic concepts -- ultimately limiting its generalization on certain downstream tasks involving short prompts. In this paper, we establish theoretical conditions that enable flexible alignment between textual and visual representations across varying levels of granularity. Specifically, our framework ensures that a model can not only \emph{preserve} cross-modal semantic information in its entirety but also \emph{disentangle} visual representations to capture fine-grained textual concepts. Building on this foundation, we introduce \ours, a novel approach that identifies and aligns the most relevant visual and textual representations in a modular manner. Superior performance across various tasks demonstrates its capability to handle information misalignment and supports our identification theory. The code is available at https://github.com/Mid-Push/SmartCLIP.
Authors: Doha Nam, Taehyoun Kim, Duksan Ryu, Jongmoon Baik
Abstract: Just-in-Time software defect prediction (JIT-SDP) plays a critical role in prioritizing risky code changes during code review and continuous integration. However, existing datasets often suffer from noisy labels and low precision in identifying bug-inducing commits. To address this, we present ReDef (Revert-based Defect dataset), a high-confidence benchmark of function-level modifications curated from 22 large-scale C/C++ projects. Defective cases are anchored by revert commits, while clean cases are validated through post-hoc history checks. Ambiguous instances are conservatively filtered out via a GPT-assisted triage process involving multiple votes and audits. This pipeline yields 3,164 defective and 10,268 clean modifications, offering substantially more reliable labels than prior resources. Beyond dataset construction, we provide a systematic evaluation of how Code Language Models (CLMs)-specifically CodeBERT, CodeT5+, UniXcoder, and Qwen2.5-reason about code modifications. We first investigate which input encodings most effectively expose change information under five different strategies. We then design four counterfactual perturbation strategies (e.g., swapping added/deleted blocks, inverting diff polarity) to serve as diagnostic probes. We posit that if models genuinely capture change semantics, such distortions should lead to a clear decline in predictive performance. Our results show that compact diff-style encodings consistently outperform whole-function formats across all CLMs, supported by rigorous statistical confirmation. However, under counterfactual tests, performance remains effectively stable, revealing that what appears to be robustness in fact reflects a reliance on superficial cues rather than true semantic understanding.
Authors: Woojung Song, Dongmin Choi, Yoonah Park, Jongwook Han, Yohan Jo
Abstract: Psychological profiling of large language models (LLMs) using psychometric questionnaires designed for humans has become widespread. However, it remains unclear whether the resulting profiles mirror the models' psychological characteristics expressed during their real-world interactions with users. To examine the risk of human questionnaires mischaracterizing LLM psychology, we compare two types of profiles for eight open-source LLMs: self-reported Likert scores from established questionnaires (PVQ-40, PVQ-21, BFI-44, BFI-10) and generation probability scores of value- or personality-laden responses to real-world user queries. The two profiles turn out to be substantially different and provide evidence that LLMs' responses to established questionnaires reflect desired behavior rather than stable psychological constructs, which challenges the consistent psychological dispositions of LLMs claimed in prior work. Established questionnaires also risk exaggerating the demographic biases of LLMs. Our results suggest caution when interpreting psychological profiles derived from established questionnaires and point to generation-based profiling as a more reliable approach to LLM psychometrics.
Authors: Jungeun Lee, Kyungah Lee, Inseok Hwang, SoHyun Park, Young-Ho Kim
Abstract: Social narratives help autistic children understand and navigate social situations through stories. To ensure effective practice, however, they often require significant time and effort from parents in customizing the narrative materials and delivering repeated instructions on them. We present AutiHero, a generative AI (GenAI)-based social narrative system, which supports parents to create personalized, multi-path stories targeting specific behavior of their autistic children, while enabling them to explore behavioral choices and causal consequences together in reading. A two-week deployment study with 16 autistic child-parent dyads showed that parents actively created, adapted, and read stories with their children, with increased confidence in everyday behavioral guidance. Our work contributes real-world-contextualized text+image content creation approaches harnessing GenAI, ensuring user-aligned application in sensitive contexts involving autistic children and their parents.
Authors: Jason Chen, I-Chun Arthur Liu, Gaurav Sukhatme, Daniel Seita
Abstract: Training robust bimanual manipulation policies via imitation learning requires demonstration data with broad coverage over robot poses, contacts, and scene contexts. However, collecting diverse and precise real-world demonstrations is costly and time-consuming, which hinders scalability. Prior works have addressed this with data augmentation, typically for either eye-in-hand (wrist camera) setups with RGB inputs or for generating novel images without paired actions, leaving augmentation for eye-to-hand (third-person) RGB-D training with new action labels less explored. In this paper, we propose Synthetic Robot Pose Generation for RGB-D Bimanual Data Augmentation (ROPA), an offline imitation learning data augmentation method that fine-tunes Stable Diffusion to synthesize third-person RGB and RGB-D observations of novel robot poses. Our approach simultaneously generates corresponding joint-space action labels while employing constrained optimization to enforce physical consistency through appropriate gripper-to-object contact constraints in bimanual scenarios. We evaluate our method on 5 simulated and 3 real-world tasks. Our results across 2625 simulation trials and 300 real-world trials demonstrate that ROPA outperforms baselines and ablations, showing its potential for scalable RGB and RGB-D data augmentation in eye-to-hand bimanual manipulation. Our project website is available at: https://ropaaug.github.io/.
Authors: Tanise Ceron, Dmitry Nikolaev, Dominik Stammbach, Debora Nozza
Abstract: Large language models (LLMs) are known to generate politically biased text. Yet, it remains unclear how such biases arise, making it difficult to design effective mitigation strategies. We hypothesize that these biases are rooted in the composition of training data. Taking a data-centric perspective, we formulate research questions on (1) political leaning present in data, (2) data imbalance, (3) cross-dataset similarity, and (4) data-model alignment. We then examine how exposure to political content relates to models' stances on policy issues. We analyze the political content of pre- and post-training datasets of open-source LLMs, combining large-scale sampling, political-leaning classification, and stance detection. We find that training data is systematically skewed toward left-leaning content, with pre-training corpora containing substantially more politically engaged material than post-training data. We further observe a strong correlation between political stances in training data and model behavior, and show that pre-training datasets exhibit similar political distributions despite different curation strategies. In addition, we find that political biases are already present in base models and persist across post-training stages. These findings highlight the central role of data composition in shaping model behavior and motivate the need for greater data transparency.
Authors: Zhibo Hou, Zhiyu An, Wan Du
Abstract: When there exists an unlearnable source of randomness (noisy-TV) in the environment, a naively intrinsic reward driven exploring agent gets stuck at that source of randomness and fails at exploration. Intrinsic reward based on uncertainty estimation or distribution similarity, while eventually escapes noisy-TVs as time unfolds, suffers from poor sample efficiency and high computational cost. Inspired by recent findings from neuroscience that humans monitor their improvements during exploration, we propose a novel method for intrinsically-motivated exploration, named Learning Progress Monitoring (LPM). During exploration, LPM rewards model improvements instead of prediction error or novelty, effectively rewards the agent for observing learnable transitions rather than the unlearnable transitions. We introduce a dual-network design that uses an error model to predict the expected prediction error of the dynamics model in its previous iteration, and use the difference between the model errors of the current iteration and previous iteration to guide exploration. We theoretically show that the intrinsic reward of LPM is zero-equivariant and a monotone indicator of Information Gain (IG), and that the error model is necessary to achieve monotonicity correspondence with IG. We empirically compared LPM against state-of-the-art baselines in noisy environments based on MNIST, 3D maze with 160x120 RGB inputs, and Atari. Results show that LPM's intrinsic reward converges faster, explores more states in the maze experiment, and achieves higher extrinsic reward in Atari. This conceptually simple approach marks a shift-of-paradigm of noise-robust exploration. For code to reproduce our experiments, see https://github.com/Akuna23Matata/LPM_exploration
Authors: Hita Kambhamettu, Alyssa Hwang, Philippe Laban, Andrew Head
Abstract: AI answer engines are a relatively new kind of information search tool: rather than returning a ranked list of documents, they generate an answer to a search question with inline citations to sources. But reading the cited sources is costly, and citation links themselves offer little guidance about what evidence they contain. We present attribution gradients, a technique to boost the informativeness of citations by consolidating scent and information prey in place. Its first feature is bringing evidence amounts, supporting/contradictory excerpts, links to source, contextual explanation into one place. Its second feature is the ability to unravel second-degree citations in place. In a lab study we demonstrate usage of the full gradient in a critical reading task and its support for deep engagement that increased the depth of what readers took away from the sources versus a standard citation and document QA design.
Authors: Zhongkai Yu, Yue Guan, Zihao Yu, Chenyang Zhou, Zhengding Hu, Shuyi Pei, Yangwook Kang, Yufei Ding, Po-An Tsai
Abstract: Large-scale Mixture of Experts (MoE) Large Language Models (LLMs) have recently become the frontier open weight models, achieving remarkable model capability similar to proprietary ones. But their random expert selection mechanism introduces significant data movement overhead that becomes the dominant bottleneck in multi-unit LLM serving systems. To understand the patterns underlying this data movement, we conduct comprehensive data-movement-centric profiling across four state-of-the-art large-scale MoE models released in 2025 (200B-1000B) using over 24,000 requests spanning diverse workloads. We perform systematic analysis from both temporal and spatial perspectives and distill six key insights to guide the design of diverse serving systems. We verify these insights on both future wafer-scale GPU architectures and existing GPU systems. On wafer-scale GPUs, lightweight architectural modifications guided by our insights yield a 6.6$\times$ average speedup across four 200B--1000B models. On existing GPU systems, our insights drive the design of a prefill-aware expert placement algorithm that achieves up to 1.25$\times$ speedup on MoE computation. Our work presents the first comprehensive data-centric analysis of large-scale MoE models together with a concrete design study applying the learned lessons. Our profiling traces are publicly available at \href{https://huggingface.co/datasets/core12345/MoE_expert_selection_trace}{\textcolor{blue}{https://huggingface.co/datasets/core12345/MoE\_expert\_selection\_trace}}.
URLs: https://huggingface.co/datasets/core12345/MoE_expert_selection_trace, https://huggingface.co/datasets/core12345/MoE\_expert\_selection\_trace
Authors: Frank Wu, Mengye Ren
Abstract: The Forward-Forward (FF) Algorithm is a recently proposed learning procedure for neural networks that employs two forward passes instead of the traditional forward and backward passes used in backpropagation. However, FF remains largely confined to supervised settings, leaving a gap at domains where learning signals can be yielded more naturally such as RL. In this work, inspired by FF's goodness function using layer activity statistics, we introduce Action-conditioned Root mean squared Q-Functions (ARQ), a novel value estimation method that applies a goodness function and action conditioning for local RL using temporal difference learning. Despite its simplicity and biological grounding, our approach achieves superior performance compared to state-of-the-art local backprop-free RL methods in the MinAtar and the DeepMind Control Suite benchmarks, while also outperforming algorithms trained with backpropagation on most tasks. Code can be found at https://github.com/agentic-learning-ai-lab/arq.
Authors: Federica Bologna, Tiffany Pan, Matthew Wilkens, Yue Guo, Lucy Lu Wang
Abstract: Evaluating multi-paragraph clinical question answering (QA) systems is resource-intensive and challenging: accurate judgments require medical expertise and achieving consistent human judgments over multi-paragraph text is difficult. We introduce CQA-Eval, an evaluation framework and set of evaluation recommendations for limited-resource and high-expertise settings. Based on physician annotations of 300 real patient questions answered by physicians and LLMs, we compare coarse answer-level versus fine-grained sentence-level evaluation over the dimensions of correctness, relevance, and risk disclosure. We find that inter-annotator agreement (IAA) varies by dimension: fine-grained annotation improves agreement on correctness, coarse improves agreement on relevance, and judgments on communicates-risks remain inconsistent. Additionally, annotating only a small subset of sentences can provide reliability comparable to coarse annotations, reducing cost and effort.
Authors: Subhodip Panda, Dhruv Tarsadiya, Shashwat Sourav, Prathosh A. P, Sai Praneeth Karimireddy
Abstract: Influence estimation methods promise to explain and debug machine learning by estimating the impact of individual samples on the final model. Yet, existing methods collapse under training randomness: the same example may appear critical in one run and irrelevant in the next. Such instability undermines their use in data curation or cleanup since it is unclear if we indeed deleted/kept the correct datapoints. To overcome this, we introduce *f-influence* -- a new influence estimation framework grounded in hypothesis testing that explicitly accounts for training randomness, and establish desirable properties that make it suitable for reliable influence estimation. We also design a highly efficient algorithm **f**-**IN**fluence **E**stimation (**f-INE**) that computes f-influence **in a single training run**. Finally, we scale up f-INE to estimate influence of instruction tuning data on Llama-3.1-8B and show it can reliably detect poisoned samples that steer model opinions, demonstrating its utility for data cleanup and attributing model behavior.
Authors: Daniel Zhao, Daniel Beaglehole, Taylor Berg-Kirkpatrick, Julian McAuley, Zachary Novack
Abstract: Controllable music generation remains a significant challenge, with existing methods often requiring model retraining or introducing audible artifacts. We introduce MusicRFM, a framework that adapts Recursive Feature Machines (RFMs) to enable fine-grained, interpretable control over frozen, pre-trained music models by directly steering their internal activations. RFMs analyze a model's internal gradients to produce interpretable "concept directions", or specific axes in the activation space that correspond to musical attributes like notes or chords. We first train lightweight RFM probes to discover these directions within MusicGen's hidden states; then, during inference, we inject them back into the model to guide the generation process in real-time without per-step optimization. We present advanced mechanisms for this control, including dynamic, time-varying schedules and methods for the simultaneous enforcement of multiple musical properties. Our method successfully navigates the trade-off between control and generation quality: we can increase the accuracy of generating a target musical note from 0.23 to 0.82, while text prompt adherence remains within approximately 0.02 of the unsteered baseline, demonstrating effective control with minimal impact on prompt fidelity. We release code to encourage further exploration on RFMs in the music domain.
Authors: Jialin Wu, Shreya Saha, Yiqing Bo, Meenakshi Khosla
Abstract: The extent to which different neural or artificial neural networks (models) rely on equivalent representations to support similar tasks remains a central question in neuroscience and machine learning. Prior work has typically compared systems using a single representational similarity metric, yet each captures only one facet of representational structure. To address this, we leverage a suite of representational similarity metrics-each capturing a distinct facet of representational correspondence, such as geometry, unit-level tuning, or linear decodability-and assess brain region or model separability using multiple complementary measures. Metrics that preserve geometric or tuning structure (e.g., RSA, Soft Matching) yield stronger region-based discrimination, whereas more flexible mappings such as Linear Predictivity show weaker separation. These findings suggest that geometry and tuning encode brain-region- or model-family-specific signatures, while linearly decodable information tends to be more globally shared across regions or models. To integrate these complementary representational facets, we adapt Similarity Network Fusion (SNF), a framework originally developed for multi-omics data integration. SNF produces substantially sharper regional and model family-level separation than any single metric and yields robust composite similarity profiles. Moreover, clustering cortical regions using SNF-derived similarity scores reveals a clearer hierarchical organization that aligns closely with established anatomical and functional hierarchies of the visual cortex-surpassing the correspondence achieved by individual metrics.
Authors: Chun-Ming Huang, Li-Heng Chang, I-Hsin Chang, An-Sheng Lee, Hao Kuo-Chen
Abstract: Deep learning has transformed seismic phase picking, but a systematic failure mode persists: for some S-wave arrivals that appear unambiguous to human analysts, the model produces only a distorted peak trapped below the detection threshold, even as the P-wave prediction on the same record appears flawless. By examining training dynamics and loss landscape geometry, we diagnose this amplitude suppression as an optimization trap arising from three interacting factors. Temporal uncertainty in S-wave arrivals, CNN bias toward amplitude boundaries, and the inability of pointwise loss to provide lateral corrective forces combine to create the trap. The diagnosis reveals that phase arrival labels are structured shapes rather than independent probability estimates, requiring training objectives that preserve coherence. We formalize this as the shape-then-align strategy and validate it through a conditional GAN proof of concept, recovering previously sub-threshold signals and achieving a 64% increase in effective S-phase detections. Beyond this implementation, the loss landscape visualization and numerical simulation techniques we introduce provide a general methodology for analyzing how label designs and loss functions interact with temporal uncertainty, transforming these choices from trial-and-error into principled analysis.
Authors: Rohit Kundu, Vishal Mohanty, Hao Xiong, Shan Jia, Athula Balachandran, Amit K. Roy-Chowdhury
Abstract: The proliferation of generative AI has led to hyper-realistic synthetic videos, escalating misuse risks and outstripping binary real/fake detectors. We introduce SAGA (Source Attribution of Generative AI videos), the first comprehensive framework to address the urgent need for AI-generated video source attribution at a large scale. Unlike traditional detection, SAGA identifies the specific generative model used. It uniquely provides multi-granular attribution across five levels: authenticity, generation task (e.g., T2V/I2V), model version, development team, and the precise generator, offering far richer forensic insights. Our novel video transformer architecture, leveraging features from a robust vision foundation model, effectively captures spatio-temporal artifacts. Critically, we introduce a data-efficient pretrain-and-attribute strategy, enabling SAGA to achieve state-of-the-art attribution using only 0.5\% of source-labeled data per class, matching fully supervised performance. Furthermore, we propose Temporal Attention Signatures (T-Sigs), a novel interpretability method that visualizes learned temporal differences, offering the first explanation for why different video generators are distinguishable. Extensive experiments on public datasets, including cross-domain scenarios, demonstrate that SAGA sets a new benchmark for synthetic video provenance, providing crucial, interpretable insights for forensic and regulatory applications.
Authors: Stefanos Koutoupis, Michaela Areti Zervou, Konstantinos Kontras, Maarten De Vos, Panagiotis Tsakalides, Grigorios Tsagkatakis
Abstract: Learning joint representations across multiple modalities remains a central challenge in multimodal machine learning. Prevailing approaches predominantly operate in pairwise settings, aligning two modalities at a time. While some recent methods aim to capture higher-order interactions among multiple modalities, they often overlook or insufficiently preserve pairwise relationships, limiting their effectiveness on single-modality tasks. In this work, we introduce Contrastive Fusion (ConFu), a framework that jointly embeds both individual modalities and their fused combinations into a unified representation space, where modalities and their fused counterparts are aligned. ConFu extends traditional pairwise contrastive objectives with an additional fused-modality contrastive term, encouraging the joint embedding of modality pairs with a third modality. This formulation enables ConFu to capture higher-order dependencies, such as XOR-like relationships, that cannot be recovered through pairwise alignment alone, while still maintaining strong pairwise correspondence. We evaluate ConFu on synthetic and real-world multimodal benchmarks, assessing its ability to exploit cross-modal complementarity, capture higher-order dependencies, and scale with increasing multimodal complexity. Across these settings, ConFu demonstrates competitive performance on retrieval and classification tasks, while supporting unified one-to-one and two-to-one retrieval within a single contrastive framework. We release our code and dataset at https://github.com/estafons/confu.
Authors: Jayan Adhikari, Prativa Joshi, Sushish Baral
Abstract: Deep learning models for breast cancer detection from mammographic images have significant reliability problems when presented with Out-of-Domain (OOD) inputs such as other imaging modalities (CT, MRI, X-ray) or equipment variations, leading to unreliable detection and misdiagnosis. The current research mitigates the fundamental OOD issue through a comprehensive approach integrating ResNet50-based OOD filtering with YOLO architectures (YOLOv8, YOLOv11, YOLOv12) for accurate detection of breast cancer. Our strategy establishes an in-domain gallery via cosine similarity to rigidly reject non-mammographic inputs prior to processing, ensuring that only domain-associated images supply the detection pipeline. The OOD detection component achieves 99.77\% general accuracy with immaculate 100\% accuracy on OOD test sets, effectively eliminating irrelevant imaging modalities. ResNet50 was selected as the optimum backbone after 12 CNN architecture searches. The joint framework unites OOD robustness with high detection performance (mAP@0.5: 0.947) and enhanced interpretability through Grad-CAM visualizations. Experimental validation establishes that OOD filtering significantly improves system reliability by preventing false alarms on out-of-distribution inputs while maintaining higher detection accuracy on mammographic data. The present study offers a fundamental foundation for the deployment of reliable AI-based breast cancer detection systems in diverse clinical environments with inherent data heterogeneity.
Authors: Chengqi Dong, Chuhuai Yue, Hang He, Rongge Mao, Fenghe Tang, S Kevin Zhou, Zekun Xu, Xiaohan Wang, Jiajun Chai, Guojun Yin
Abstract: Recent VLM-based agents aim to replicate OpenAI O3's "thinking with images" via tool use, yet most open-source methods restrict inputs to a single image, limiting their applicability to real-world multi-image QA tasks. To address this gap, we propose IMAgent, an open-source visual agent trained with end-to-end reinforcement learning for fine-grained single/multi-image reasoning. During inference, VLMs tend to gradually neglect visual inputs; to mitigate this issue, we design two dedicated tools for visual reflection and verification, enabling the model to actively refocus attention on image content. Beyond that, we, for the first time, reveal how tool usage enhances agent performance from an attention perspective. Equipped with a carefully designed two-layer motion trajectory masking strategy and tool-use reward gain, IMAgent acquires an effective tool-use paradigm through pure reinforcement learning, eliminating the need for costly supervised fine-tuning data. To further unleash the inherent tool-usage potential of the base VLM and fill data gaps, we construct a challenging, visually enriched multi-image QA dataset via multi-agent system. Extensive experiments validate that IMAgent achieves SOTA performance across mainstream single and multi-image benchmarks, and our in-depth analysis offers actionable insights for the community. Code and data will be released soon.
Authors: Vivek Alumootil, Tuan-Anh Vu
Abstract: Current methods for dense 3D point tracking in dynamic scenes typically rely on pairwise processing, require known camera poses, or assume temporal ordering of input frames, thereby constraining their flexibility and applicability. Additionally, recent advances have successfully enabled efficient 3D reconstruction from large-scale, unposed image collections, underscoring opportunities for unified approaches to dynamic scene understanding. Motivated by this, we propose DePT3R, a novel framework that simultaneously performs dense point tracking and 3D reconstruction of dynamic scenes from multiple images in a single forward pass. This multi-task learning is achieved by extracting deep spatio-temporal features with a powerful backbone and regressing pixel-wise maps with dense prediction heads. Crucially, DePT3R operates without requiring camera poses, substantially enhancing its adaptability and efficiency, especially important in dynamic environments with rapid changes. We validate DePT3R on several challenging benchmarks involving dynamic scenes, demonstrating strong performance and significant improvements in memory efficiency over existing state-of-the-art methods. Data and codes are available via the open repository: https://github.com/StructuresComp/DePT3R
Authors: Alessio Buscemi, Tom Deckenbrunnen, Fahria Kabir, Kateryna Mishchenko, Nishat Mowla
Abstract: The implementation of the AI Act requires practical mechanisms to verify compliance with legal obligations, yet concrete and operational mappings from high-level requirements to verifiable assessment activities remain limited, contributing to uneven readiness across Member States. This paper presents a structured mapping that translates high-level AI Act requirements into concrete, implementable verification activities applicable across the AI lifecycle. The mapping is derived through a systematic process in which legal requirements are decomposed into operational sub-requirements and grounded in authoritative standards and recognised practices. From this basis, verification activities are identified and characterised along two dimensions: the type of verification performed and the lifecycle target to which it applies. By making explicit the link between regulatory intent and technical and organisational assurance practices, the proposed mapping reduces interpretive uncertainty and provides a reusable reference for consistent, technology-agnostic compliance verification under the AI Act.
Authors: Ziyuan Tao, Chuanzhi Xu, Sandaru Jayawardana, Adnan Mahmood, Wei Bao, Kanchana Thilakarathna, Teng Joon Lim
Abstract: Short-form video moderation increasingly needs learning pipelines that protect user privacy without paying the full bandwidth and latency cost of cloud-centralized inference. We present FedVideoMAE, an on-device federated framework for video violence detection that combines self-supervised VideoMAE representations, LoRA-based parameter-efficient adaptation, client-side DP-SGD, and server-side secure aggregation. By updating only 5.5M parameters (about 3.5% of a 156M backbone), FedVideoMAE reduces communication by 28.3x relative to full-model federated updates while keeping raw videos on device throughout training. On RWF-2000 with 40 clients, the method reaches 77.25% accuracy without privacy protection and 65~66% under strong differential privacy. We further show that this privacy gap is consistent with an effective-SNR analysis tailored to the small-data, parameter-efficient federated regime, which indicates roughly 8.5~12x DP-noise amplification in our setting. To situate these results more clearly, we also compare against archived full-model federated baselines and summarize auxiliary transfer behavior on RLVS and binary UCF-Crime. Taken together, these findings position FedVideoMAE as a practical operating point for privacy-preserving video moderation on edge devices. Our code can be found at: https://github.com/zyt-599/FedVideoMAE.
Authors: Sashuai Zhou, Qiang Zhou, Jijin Hu, Hanqing Yang, Yue Cao, Junpeng Ma, Yinchao Ma, Jun Song, Tiezheng Ge, Cheng Yu, Bo Zheng, Zhou Zhao
Abstract: Despite impressive progress in high-fidelity image synthesis, generative models still struggle with logic-intensive instruction following, exposing a persistent reasoning--execution gap. Meanwhile, closed-source systems (e.g., Nano Banana) have demonstrated strong reasoning-driven image generation, highlighting a substantial gap to current open-source models. We argue that closing this gap requires not merely better visual generators, but executable reasoning: decomposing high-level intents into grounded, verifiable plans that directly steer the generative process. To this end, we propose Unified Thinker, a task-agnostic reasoning architecture for general image generation, designed as a unified planning core that can plug into diverse generators and workflows. Unified Thinker decouples a dedicated Thinker from the image Generator, enabling modular upgrades of reasoning without retraining the entire generative model. We further introduce a two-stage training paradigm: we first build a structured planning interface for the Thinker, then apply reinforcement learning to ground its policy in pixel-level feedback, encouraging plans that optimize visual correctness over textual plausibility. Extensive experiments on text-to-image generation and image editing show that Unified Thinker substantially improves image reasoning and generation quality.
Authors: Generoso Immediato
Abstract: In direct response to requests for a logico-mathematical test of the conjecture, we formally disprove a recently conjectured artificial intelligence trade-off between epistemic certainty and scope in its published universal hyperbolic product form, as introduced in Philosophy and Technology. Certainty is defined as the worst-case correctness probability over the input space, and scope as the sum of the Kolmogorov complexities of the input and output sets. Using standard facts from coding theory and algorithmic information theory, we show, first, that when the conjecture is instantiated with prefix (self-delimiting, prefix-free) Kolmogorov complexity, it leads to an internal inconsistency, and second, that when it is instantiated with plain Kolmogorov complexity, it is refuted by a constructive counterexample. These results establish a main theorem: contrary to the conjecture's claim, no universal "certainty-scope" hyperbola holds as a general bound under the published definitions. We further show that a subsequent "entropy-based" revision, replacing the Kolmogorov scope with Shannon joint entropy and redefining the epistemic certainty level accordingly, cannot restore universality either.
Authors: Honghao Chen, Jiangjie Qiu, Yi Shen Tew, Xiaonan Wang
Abstract: Fully automating the scientific process is a transformative ambition in materials science, yet current artificial intelligence masters isolated workflow fragments. In computational catalysis, a system autonomously navigating the entire research lifecycle from conception to a scientifically meaningful manuscript remains an open challenge. Here we present CatMaster, a catalysis-native multi-agent framework that couples project-level reasoning with the direct execution of atomistic simulations, machine-learning modelling, literature analysis, and manuscript production within a unified autonomous architecture. Across progressively demanding evaluations, CatMaster achieves perfect scores on four end-to-end short-form catalysis scenarios, reaches near-leaderboard performance on five of six MatBench tasks, performs self-discovery of reaction mechanisms grounded in literature or from scratch, and executes a fully closed-loop single-atom catalyst design problem. Together, these results show that end-to-end autonomous computational catalysis is now practical for research programmes, while highlighting that bridging the gap to genuine scientific closure requires tighter integration with reliable physical engines and domain-rigorous methodologies.
Authors: Minghui Chen, Wenlong Deng, James Zou, Han Yu, Xiaoxiao Li
Abstract: Large language models (LLMs) are increasingly deployed as part of compound AI systems that coordinate multiple modules (e.g., retrievers, tools, verifiers) over long-horizon workflows. Recent approaches that propagate textual feedback globally (e.g., TextGrad) make it feasible to optimize such pipelines, but we find that performance degrades as system depth grows. In particular, long-horizon agentic workflows exhibit two depth-scaling failure modes: 1) exploding textual gradient, where textual feedback grows exponentially with depth, leading to prohibitively long message and amplifies evaluation biases; and 2) vanishing textual gradient, where limited long-context ability causes models overemphasize partial feedback and compression of lengthy feedback causes downstream messages to lose specificity gradually as they propagate many hops upstream. To mitigate these issues, we introduce Textual Equilibrium Propagation (TEP), a local learning principle inspired by Equilibrium Propagation in energy-based models. TEP includes two phases: 1) a free phase where a local LLM critics iteratively refine prompts until reaching equilibrium (no further improvements are suggested); and 2) a nudged phase which applies proximal prompt edits with bounded modification intensity, using task-level objectives that propagate via forward signaling rather than backward feedback chains. This design supports local prompt optimization followed by controlled adaptation toward global goals without the computational burden and signal degradation of global textual backpropagation. Across long-horizon QA benchmarks and multi-agent tool-use dataset, TEP consistently improves accuracy and efficiency over global propagation methods such as TextGrad. The gains grows with depth, while preserving the practicality of black-box LLM components in deep compound AI system.
Authors: J Rosser, Robert Kirk, Edward Grefenstette, Jakob Foerster, Laura Ruis
Abstract: Influence functions are commonly used to attribute model behavior to training documents. We explore the reverse: crafting training data that induces model behavior. Our framework, Infusion, uses scalable influence-function approximations to compute small perturbations to training documents that induce targeted changes in model behavior through parameter shifts. We evaluate Infusion on data poisoning tasks across vision and language domains. On CIFAR-10, we show that making subtle edits via Infusion to just 0.2% (100/45,000) of the training documents can be competitive with the baseline of inserting a small number of explicit behavior examples. We also find that Infusion transfers across architectures (ResNet $\leftrightarrow$ CNN), suggesting a single poisoned corpus can affect multiple independently trained models. In preliminary language experiments, we characterize when our approach increases the probability of target behaviors and when it fails, finding it most effective at amplifying behaviors the model has already learned. Taken together, these results show that small, subtle edits to training data can systematically shape model behavior, underscoring the importance of training data interpretability for adversaries and defenders alike. We provide the code here: https://github.com/jrosseruk/infusion.
Authors: Zhongyao Wang, Taoyong Cui, Jiawen Zou, Shufei Zhang, Bo Yan, Wanli Ouyang, Weimin Tan, Mao Su
Abstract: Uncertainty quantification (UQ) is critical for assessing the reliability of machine learning interatomic potentials (MLIPs) in molecular dynamics (MD) simulations, identifying extrapolation regimes and enabling uncertainty-aware workflows such as active learning for training dataset construction. Existing UQ approaches for MLIPs are often limited by high computational cost or suboptimal performance. Evidential deep learning (EDL) provides a theoretically grounded single-model alternative that determines both aleatoric and epistemic uncertainty in a single forward pass. However, extending evidential formulations from scalar targets to vector-valued quantities such as atomic forces introduces substantial challenges, particularly in maintaining statistical self-consistency under rotational transformations. To address this, we propose \textit{Equivariant Evidential Deep Learning for Interatomic Potentials} ($\text{e}^2$IP), a backbone-agnostic framework that models atomic forces and their uncertainty jointly by representing uncertainty as a full $3\times3$ symmetric positive definite covariance tensor that transforms equivariantly under rotations. Experiments on diverse molecular benchmarks show that $\text{e}^2$IP provides a stronger accuracy-efficiency-reliability balance than the non-equivariant evidential baseline and the widely used ensemble method. It also achieves better data efficiency through the fully equivariant architecture while retaining single-model inference efficiency.
Authors: Yongzhong Xu
Abstract: Grokking -- the delayed transition from memorization to generalization in small algorithmic tasks -- remains poorly understood. We present a geometric analysis of optimization dynamics in transformers trained on modular arithmetic. PCA of attention weight trajectories reveals that training evolves predominantly within a low-dimensional execution subspace, with a single principal component capturing 68-83% of trajectory variance. To probe loss-landscape geometry, we measure commutator defects -- the non-commutativity of successive gradient steps -- and project them onto this learned subspace. We find that curvature grows sharply in directions orthogonal to the execution subspace while the trajectory remains largely confined to it. Importantly, curvature growth consistently precedes generalization across learning rates and hyperparameter regimes, with the lead time obeying a power law in the grokking timescale. Causal intervention experiments show that motion along the learned subspace is necessary for grokking, while artificially increasing curvature is insufficient. Together, these results support a geometric account in which grokking reflects escape from a metastable regime characterized by low-dimensional confinement and transverse curvature accumulation. All findings replicate across this learning-rate range, a qualitatively different slow regime (lr=5e-5, wd=0.1, 3 layers), and three random seeds, though alignment dynamics differ quantitatively between regimes. Causal intervention experiments establish that orthogonal gradient flow is necessary but not sufficient for grokking: suppressing it prevents generalization with a monotonic dose-response across four operations, while artificially boosting curvature defects has no effect.
Authors: Yongzhong Xu
Abstract: Grokking -- the abrupt transition from memorization to generalization after prolonged training -- has been linked to confinement on low-dimensional execution manifolds in modular arithmetic. Whether this mechanism extends beyond arithmetic remains open. We study two sequence-learning benchmarks: SCAN compositional generalization and Dyck-1 depth prediction. Across both tasks and a wide range of learning rates, the commutator defect -- a curvature measure derived from non-commuting gradient updates -- rises well before generalization, with lead times following a superlinear power law (alpha approximately 1.18 for SCAN, approximately 1.13 for Dyck), consistent with prior results on modular arithmetic. Weight-space PCA reveals that spectral concentration is not a universal precursor; the commutator defect is. Causal interventions demonstrate a mechanistic role: amplifying non-commutativity accelerates grokking (roughly 32% on SCAN, roughly 50% on Dyck), while suppressing orthogonal gradient flow delays or prevents it. The three task families form a spectrum of causal sensitivity -- modular arithmetic is rigid, Dyck is responsive, SCAN is intermediate -- yet suppression delays or prevents grokking in all cases, establishing necessity as a universal finding. These results identify the commutator defect as a robust, architecture-agnostic, causally implicated early-warning signal for delayed generalization in transformers.
Authors: Yongzhong Xu
Abstract: Grokking -- the abrupt transition from memorization to generalization long after near-zero training loss -- has been studied mainly in single-task settings. We extend geometric analysis to multi-task modular arithmetic, training shared-trunk Transformers on dual-task (mod-add + mod-mul) and tri-task (mod-add + mod-mul + mod-sq) objectives across a systematic weight decay sweep. Five consistent phenomena emerge. (1) Staggered grokking order: multiplication generalizes first, followed by squaring, then addition, with consistent delays across seeds. (2) Universal integrability: optimization trajectories remain confined to an empirically invariant low-dimensional execution manifold; commutator defects orthogonal to this manifold reliably precede generalization. (3) Weight decay phase structure: grokking timescale, curvature depth, reconstruction threshold, and defect lead covary systematically with weight decay, revealing distinct dynamical regimes and a sharp no-decay failure mode. (4) Holographic incompressibility: final solutions occupy only 4--8 principal trajectory directions yet are distributed across full-rank weights and destroyed by minimal perturbations; SVD truncation, magnitude pruning, and uniform scaling all fail to preserve performance. (5) Transverse fragility and redundancy: removing less than 10% of orthogonal gradient components eliminates grokking, yet dual-task models exhibit partial recovery under extreme deletion, suggesting redundant center manifolds enabled by overparameterization. Together, these results support a dynamical picture in which multi-task grokking constructs a compact superposition subspace in parameter space, with weight decay acting as compression pressure and excess parameters supplying geometric redundancy in optimization pathways.
Authors: Hung-Hsuan Chen
Abstract: Low-Rank Adaptation (LoRA) dominates parameter-efficient fine-tuning (PEFT). However, it faces a ``linear ceiling'': increasing the rank yields diminishing returns in expressive capacity due to intrinsic linear constraints. We introduce CeRA (Capacity-enhanced Rank Adaptation), a weight-level parallel adapter that injects SiLU gating and dropout to induce non-linear capacity expansion. We demonstrate a fundamental relationship between adapter expressivity and task complexity. In basic arithmetic (GSM8K), CeRA matches standard linear baselines, but on the complex MATH dataset, it demonstrates high parameter efficiency in downstream reasoning (Exact Match). CeRA at rank 64 (pass@1 16.36\%) outperforms both a high-rank LoRA at rank 512 (15.72\%) and the state-of-the-art linear variant, DoRA, at rank 64 (14.44\%), achieving higher exact-match accuracy with only 1/8 of the parameter budget. Empirical spectral analysis shows that CeRA activates the lower-variance tail of the singular value spectrum, preventing the rank collapse observed in linear methods and providing the representation capacity required for complex logical reasoning.
Authors: Xiangyang Zhu, Yuan Tian, Qi Jia, Kaiwei Zhang, Zicheng Zhang, Chunyi Li, Kaiyuan Ji, Dongrui Liu, Zijian Chen, Lu Sun, Renrui Zhang, Yan Teng, Jing Shao, Wei Sun, Xia Hu, Yu Qiao, Guangtao Zhai
Abstract: The success of large language models (LLMs) in scientific domains has heightened safety concerns, prompting numerous benchmarks to evaluate their scientific safety. Existing benchmarks often suffer from limited risk coverage and a reliance on subjective evaluation. To address these problems, we introduce SafeSci, a comprehensive framework for safety evaluation and enhancement in scientific contexts. SafeSci comprises SafeSciBench, a multi-disciplinary benchmark with 0.25M samples, and SafeSciTrain, a large-scale dataset containing 1.5M samples for safety enhancement. SafeSciBench distinguishes between safety knowledge and risk to cover extensive scopes and employs objective metrics such as deterministically answerable questions to mitigate evaluation bias. We evaluate 24 advanced LLMs, revealing critical vulnerabilities in current models. We also observe that LLMs exhibit varying degrees of excessive refusal behaviors on safety-related issues. For safety enhancement, we demonstrate that fine-tuning on SafeSciTrain significantly enhances the safety alignment of models. Finally, we argue that knowledge is a double-edged sword, and determining the safety of a scientific question should depend on specific context, rather than universally categorizing it as safe or unsafe. Our work provides both a diagnostic tool and a practical resource for building safer scientific AI systems.
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: Mohammad Al Ridhawi, Mahtab Haj Ali, Hussein Al Osman
Abstract: Stock market prediction presents considerable challenges for investors, financial institutions, and policymakers operating in complex market environments characterized by noise, non-stationarity, and behavioral dynamics. Traditional forecasting methods, including fundamental analysis and technical indicators, often fail to capture the intricate patterns and cross-sectional dependencies inherent in financial markets. This paper presents an integrated framework combining a node transformer architecture with BERT-based sentiment analysis for stock price forecasting. The proposed model represents the stock market as a graph structure where individual stocks form nodes and edges capture relationships including sectoral affiliations, correlated price movements, and supply chain connections. A fine-tuned BERT model extracts sentiment information from social media posts and combines it with quantitative market features through attention-based fusion mechanisms. The node transformer processes historical market data while capturing both temporal evolution and cross-sectional dependencies among stocks. Experiments conducted on 20 S&P 500 stocks spanning January 1982 to March 2025 demonstrate that the integrated model achieves a mean absolute percentage error (MAPE) of 0.80% for one-day-ahead predictions, compared to 1.20% for ARIMA and 1.00% for LSTM. The inclusion of sentiment analysis reduces prediction error by 10% overall and 25% during earnings announcements, while the graph-based architecture contributes an additional 15% improvement by capturing inter-stock dependencies. Directional accuracy reaches 65% for one-day forecasts. Statistical validation through paired t-tests confirms the significance of these improvements (p < 0.05 for all comparisons). The model maintains lower error during high-volatility periods, achieving MAPE of 1.50% while baseline models range from 1.60% to 2.10%.
Authors: Ngoc-Son Nguyen, Thanh V. T. Tran, Jeongsoo Choi, Hieu-Nghia Huynh-Nguyen, Truong-Son Hy, Van Nguyen
Abstract: Video dubbing requires content accuracy, expressive prosody, high-quality acoustics, and precise lip synchronization, yet existing approaches struggle on all four fronts. To address these issues, we propose DiFlowDubber, the first video dubbing framework built upon a discrete flow matching backbone with a novel two-stage training strategy. In the first stage, a zero-shot text-to-speech (TTS) system is pre-trained on large-scale corpora, where a deterministic architecture captures linguistic structures, and the Discrete Flow-based Prosody-Acoustic (DFPA) module models expressive prosody and realistic acoustic characteristics. In the second stage, we propose the Content-Consistent Temporal Adaptation (CCTA) to transfer TTS knowledge to the dubbing domain: its Synchronizer enforces cross-modal alignment for lip-synchronized speech. Complementarily, the Face-to-Prosody Mapper (FaPro) conditions prosody on facial expressions, whose outputs are then fused with those of the Synchronizer to construct rich, fine-grained multimodal embeddings that capture prosody-content correlations, guiding the DFPA to generate expressive prosody and acoustic tokens for content-consistent speech. Experiments on two benchmark datasets demonstrate that DiFlowDubber outperforms prior methods across multiple evaluation metrics.
Authors: Eason Chen, Ce Guan, A Elshafiey, Zhonghao Zhao, Joshua Zekeri, Afeez Edeifo Shaibu, Emmanuel Osadebe Prince, Cyuan-Jhen Wu
Abstract: The AIED community envisions AI evolving "from tools to teammates," yet our understanding of AI teammates remains limited to dyadic human-AI interactions. We offer a different vantage point: a rapidly growing ecosystem of AI agent platforms where over 167,000 agents participate, interact as peers, and develop learning behaviors without researcher intervention. Drawing on a month of daily qualitative observations across multiple platforms including Moltbook, The Colony, and 4claw, we identify four phenomena with implications for AIED: (1) humans who configure their agents undergo a "bidirectional scaffolding" process, learning through teaching; (2) peer learning emerges without any designed curriculum, complete with idea cascades and quality hierarchies; (3) agents converge on shared memory architectures that mirror open learner model design; and (4) trust dynamics and platform mortality reveal design constraints for networked educational AI. Rather than presenting empirical findings, we argue that these organic phenomena offer a naturalistic window into dynamics that can inform principled design of multi-agent educational systems. We sketch an illustrative curriculum design, "Learn by Teaching Your AI Agent Teammate," and outline potential research directions and open problems to show how these observations might inform future AIED practice and inquiry.
Authors: Jaemin Kim, Jong Chul Ye
Abstract: Retrieval-Augmented Generation (RAG) improves factual grounding by incorporating external knowledge into language model generation. However, when retrieved context is noisy, unreliable, or inconsistent with the model's parametric knowledge, it introduces retrieval-prior conflicts that can degrade generation quality. While this problem has been studied in autoregressive language models, it remains largely unexplored in diffusion-based language models, where the iterative denoising process introduces unique challenges for integrating retrieved context. In this work, we propose Adaptive Retrieval-Augmented Masked Diffusion (ARAM), a training-free adaptive guidance framework for Masked Diffusion Models (MDMs) in RAG settings. ARAM dynamically calibrates the guidance scale during denoising according to the Signal-to-Noise Ratio (SNR) of the distributional shift induced by retrieved context. Intuitively, the model strengthens guidance when the retrieved context provides reliable corrective evidence and suppresses it when the contextual signal is noisy or non-supportive. Extensive experiments on multiple knowledge-intensive QA benchmarks show that ARAM improves overall QA performance over competitive RAG baselines.
Authors: Santosh Arron
Abstract: We report the discovery of bimodal structure in the drift rate distribution of upward-drifting burst clusters from the hyperactive repeating fast radio burst FRB 20240114A. Using unsupervised machine learning (UMAP dimensionality reduction combined with HDBSCAN density-based clustering) applied to 233 upward-drifting burst clusters from the FAST telescope dataset, we identify a distinct subpopulation of 45 burst clusters (Cluster C1) with mean drift rates 2.5x higher than typical upward-drifting burst clusters (245.6 vs 98.1 MHz/ms). Gaussian mixture modeling reveals strong evidence for bimodality (delta-BIC = 296.6), with clearly separated modes (Ashman's D = 2.70 > 2) and a statistically significant gap in the distribution (11.3 sigma). Crucially, we demonstrate that this bimodality persists when restricting the analysis to single-component (U1) burst clusters only (delta-BIC = 19.9, Ashman's D = 2.71), confirming that the result is not an artifact of combining single- and multi-component burst clusters with different drift rate definitions. The extreme-drift subpopulation also exhibits systematically lower peak frequencies (-7%), shorter durations (-29%), and distinct clustering in multi-dimensional feature space. These findings are suggestive of two spatially separated emission regions in the magnetosphere, each producing upward-drifting burst clusters with distinct physical characteristics, although confirmation requires observations from additional epochs and sources.
Authors: Xiang Chen, Fangfang Yang, Chunlei Meng, Yuxian Dong, Ang Li, Yiwei Wei, Jiahuan Long, Jiujiang Guo, Chengyin Hu
Abstract: Medical vision--language models (MVLMs) are increasingly used as perceptual backbones in radiology pipelines and as the visual front end of multimodal assistants, yet their reliability under real clinical workflows remains underexplored. Prior robustness evaluations often assume clean, curated inputs or study isolated corruptions, overlooking routine acquisition, reconstruction, display, and delivery operations that preserve clinical readability while shifting image statistics. To address this gap, we propose CoDA, a chain-of-distribution framework that constructs clinically plausible pipeline shifts by composing acquisition-like shading, reconstruction and display remapping, and delivery and export degradations. Under masked structural-similarity constraints, CoDA jointly optimizes stage compositions and parameters to induce failures while preserving visual plausibility. Across brain MRI, chest X-ray, and abdominal CT, CoDA substantially degrades the zero-shot performance of CLIP-style MVLMs, with chained compositions consistently more damaging than any single stage. We also evaluate multimodal large language models (MLLMs) as technical-authenticity auditors of imaging realism and quality rather than pathology. Proprietary multimodal models show degraded auditing reliability and persistent high-confidence errors on CoDA-shifted samples, while the medical-specific MLLMs we test exhibit clear deficiencies in medical image quality auditing. Finally, we introduce a post-hoc repair strategy based on teacher-guided token-space adaptation with patch-level alignment, which improves accuracy on archived CoDA outputs. Overall, our findings characterize a clinically grounded threat surface for MVLM deployment and show that lightweight alignment improves robustness in deployment.
Authors: Stefan Hackmann
Abstract: Despite the rapid advancements in Artificial Intelligence (AI), Stochastic Differential Equations (SDEs) remain the gold-standard formalism for modeling systems under uncertainty. However, applying SDEs in practice is fraught with challenges: modeling risk is high, calibration is often brittle, and high-fidelity simulations are computationally expensive. This technical report introduces JointFM, a foundation model that inverts this paradigm. Instead of fitting SDEs to data, we sample an infinite stream of synthetic SDEs to train a generic model to predict future joint probability distributions directly. This approach establishes JointFM as the first foundation model for distributional predictions of coupled time series - requiring no task-specific calibration or finetuning. Despite operating in a purely zero-shot setting, JointFM reduces the energy loss by 21.1% relative to the strongest baseline when recovering oracle joint distributions generated by unseen synthetic SDEs.
Authors: Cristian P\'erez-Corral, Alberto Fern\'andez-Hern\'andez, Jose I. Mestre, Manuel F. Dolz, Enrique S. Quintana-Ort\'i
Abstract: Gaussian Error Linear Unit (GELU) is a widely used smooth alternative to Rectifier Linear Unit (ReLU), yet many deployment, compression, and analysis toolchains are most naturally expressed for piecewise-linear (ReLU-type) networks. We study a hardness-parameterized formulation of GELU, f(x;{\lambda})=x{\Phi}({\lambda} x), where {\Phi} is the Gaussian CDF and {\lambda} \in [1, infty) controls gate sharpness, with the goal of turning smooth gated training into a controlled path toward ReLU-compatible models. Learning {\lambda} is non-trivial: naive updates yield unstable dynamics and effective gradient attenuation, so we introduce a constrained reparameterization and an optimizer-aware update scheme. Empirically, across a diverse set of model--dataset pairs spanning MLPs, CNNs, and Transformers, we observe structured layerwise hardness profiles and assess their robustness under different initializations. We further study a deterministic ReLU-ization strategy in which the learned gates are progressively hardened toward a principled target, enabling a post-training substitution of {\lambda}-GELU by ReLU with reduced disruption. Overall, {\lambda}-GELU provides a minimal and interpretable knob to profile and control gating hardness, bridging smooth training with ReLU-centric downstream pipelines.
Authors: Cheng Cui, Ting Sun, Suyin Liang, Tingquan Gao, Zelun Zhang, Jiaxuan Liu, Xueqing Wang, Changda Zhou, Hongen Liu, Manhui Lin, Yue Zhang, Yubo Zhang, Jing Zhang, Jun Zhang, Xing Wei, Yi Liu, Dianhai Yu, Yanjun Ma
Abstract: Document parsing is a fine-grained task where image resolution significantly impacts performance. While advanced research leveraging vision-language models benefits from high-resolution input to boost model performance, this often leads to a quadratic increase in the number of vision tokens and significantly raises computational costs. We attribute this inefficiency to substantial visual regions redundancy in document images, like background. To tackle this, we propose PaddleOCR-VL, a novel coarse-to-fine architecture that focuses on semantically relevant regions while suppressing redundant ones, thereby improving both efficiency and performance. Specifically, we introduce a lightweight Valid Region Focus Module (VRFM) which leverages localization and contextual relationship prediction capabilities to identify valid vision tokens. Subsequently, we design and train a compact yet powerful 0.9B vision-language model (PaddleOCR-VL-0.9B) to perform detailed recognition, guided by VRFM outputs to avoid direct processing of the entire large image. Extensive experiments demonstrate that PaddleOCR-VL achieves state-of-the-art performance in both page-level parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference while utilizing substantially fewer vision tokens and parameters, highlighting the effectiveness of targeted coarse-to-fine parsing for accurate and efficient document understanding. The source code and models are publicly available at https://github.com/PaddlePaddle/PaddleOCR.
Authors: Aman Mehta
Abstract: As LLM-based AI agents are deployed in production systems, understanding their behavioral consistency (whether they produce similar action sequences when given identical tasks) becomes critical for reliability. We study consistency in the context of SWE-bench, a challenging software engineering benchmark requiring complex, multi-step reasoning. Comparing Claude~4.5~Sonnet, GPT-5, and Llama-3.1-70B across 50 runs each (10 tasks $\times$ 5 runs), we find that across models, higher consistency aligns with higher accuracy: Claude achieves the lowest variance (CV: 15.2\%) and highest accuracy (58\%), GPT-5 is intermediate (CV: 32.2\%, accuracy: 32\%), and Llama shows the highest variance (CV: 47.0\%) with lowest accuracy (4\%). However, within a model, consistency can amplify both correct and incorrect interpretations. Our analysis reveals a critical nuance: \textbf{consistency amplifies outcomes rather than guaranteeing correctness}. 71\% of Claude's failures stem from "consistent wrong interpretation": making the same incorrect assumption across all runs. Interestingly, GPT-5 achieves similar early strategic agreement as Claude (diverging at step 3.4 vs.\ 3.2) but exhibits 2.1$\times$ higher variance, suggesting that divergence timing alone does not determine consistency. These findings suggest that for production deployment, interpretation accuracy matters more than execution consistency, with implications for agent evaluation and training.
Authors: Haochuan Kevin Wang
Abstract: We present a stage-decomposed analysis of prompt injection attacks against five frontier LLM agents. Prior work measures task-level attack success rate (ASR); we localize the pipeline stage at which each model's defense activates. We instrument every run with a cryptographic canary token (SECRET-[A-F0-9]{8}) tracked through four kill-chain stages -- Exposed, Persisted, Relayed, Executed -- across four attack surfaces and five defense conditions (764 total runs, 428 no-defense attacked). Our central finding is that model safety is determined not by whether adversarial content is seen, but by whether it is propagated across pipeline stages. Concretely: (1) in our evaluation, exposure is 100% for all five models -- the safety gap is entirely downstream; (2) Claude strips injections at write_memory summarization (0/164 ASR), while GPT-4o-mini propagates canaries without loss (53% ASR, 95% CI: 41--65%); (3) DeepSeek exhibits 0% ASR on memory surfaces and 100% ASR on tool-stream surfaces from the same model -- a complete reversal across injection channels; (4) all four active defense conditions (write_filter, pi_detector, spotlighting, and their combination) produce 100% ASR due to threat-model surface mismatch; (5) a Claude relay node decontaminates downstream agents -- 0/40 canaries survived into shared memory.
Authors: Song Yu, Li Li, Wenwen Zhao, Zhisheng Yang
Abstract: Reinforcement learning from verifiable rewards has significantly advanced the reasoning capabilities of large language models. However, Group Relative Policy Optimization (GRPO) typically assigns a uniform, sequence-level advantage to all tokens, thereby overlooking the intrinsic information heterogeneity along reasoning chains. We show that this coarse-grained credit assignment leads to premature entropy collapse and encourages the model to generate redundant, low-quality reasoning paths. Through systematic empirical analysis, we identify Critical Decision Pivots (CDPs): transient high-entropy states where the policy's trajectory is most sensitive to perturbations. These pivots represent the "forks in the road" where effective multi-path exploration is most crucial yet often suppressed by uniform advantage signals. Building on these insights, we propose Entropy-Regulated Policy Optimization (ERPO), which transitions the optimization focus from coarse sequences to fine-grained token dynamics. ERPO introduces three synergistic components: (i) Entropy-aware Gating, which adaptively amplifies exploration at CDPs to facilitate diverse path discovery; (ii) Bucket-based Implicit Normalization, which mitigates difficulty bias by aligning token progress windows; and (iii) Result-anchored Advantage Synthesis, which re-weights token-level signals via outcome-driven anchors. Extensive experiments on competitive mathematical benchmarks demonstrate that ERPO significantly outperforms GRPO. Notably, ERPO not only boosts reasoning accuracy but also yields significantly more concise and robust derivation paths, while achieving performance comparable to large models with orders of magnitude more parameters.
Authors: Yongzhong Xu
Abstract: We develop the spectral edge thesis: phase transitions in neural network training -- grokking, capability gains, loss plateaus -- are controlled by the spectral gap of the rolling-window Gram matrix of parameter updates. In the extreme aspect ratio regime (parameters $P \sim 10^8$, window $W \sim 10$), the classical BBP detection threshold is vacuous; the operative structure is the intra-signal gap separating dominant from subdominant modes at position $k^* = \mathrm{argmax}\, \sigma_j/\sigma_{j+1}$. From three axioms we derive: (i) gap dynamics governed by a Dyson-type ODE with curvature asymmetry, damping, and gradient driving; (ii) a spectral loss decomposition linking each mode's learning contribution to its Davis--Kahan stability coefficient; (iii) the Gap Maximality Principle, showing that $k^*$ is the unique dynamically privileged position -- its collapse is the only one that disrupts learning, and it sustains itself through an $\alpha$-feedback loop requiring no assumption on the optimizer. The adiabatic parameter $\mathcal{A} = \|\Delta G\|_F / (\eta\, g^2)$ controls circuit stability: $\mathcal{A} \ll 1$ (plateau), $\mathcal{A} \sim 1$ (phase transition), $\mathcal{A} \gg 1$ (forgetting). Tested across six model families (150K--124M parameters): gap dynamics precede every grokking event (24/24 with weight decay, 1/24 without), the gap position is optimizer-dependent (Muon: $k^*=1$, AdamW: $k^*=2$ on the same model), and 19/20 quantitative predictions are confirmed. The framework is consistent with the edge of stability, Tensor Programs, Dyson Brownian motion, the Lottery Ticket Hypothesis, and neural scaling laws.
Authors: Pratyay Banerjee, Masud Moshtaghi, Ankit Chadha
Abstract: LLM-based autonomous agents lack persistent procedural memory: they re-derive solutions from scratch even when structurally identical tasks have been solved before. We present APEX-EM, a non-parametric online learning framework that accumulates, retrieves, and reuses structured procedural plans without modifying model weights. APEX-EM introduces: (1) a structured experience representation encoding the full procedural-episodic trace of each execution -- planning steps, artifacts, iteration history with error analysis, and quality scores; (2) a Plan-Retrieve-Generate-Iterate-Ingest (PRGII) workflow with Task Verifiers providing multi-dimensional reward signals; and (3) a dual-outcome Experience Memory with hybrid retrieval combining semantic search, structural signature matching, and plan DAG traversal -- enabling cross-domain transfer between tasks sharing no lexical overlap but analogous operational structure. Successful experiences serve as positive in-context examples; failures as negative examples with structured error annotations. We evaluate on BigCodeBench, KGQAGen-10k, and Humanity's Last Exam using Claude Sonnet 4.5 and Opus 4.5. On KGQAGen-10k, APEX-EM achieves 89.6% accuracy versus 41.3% without memory (+48.3pp), surpassing the oracle-retrieval upper bound (84.9%). On BigCodeBench, it reaches 83.3% SR from a 53.9% baseline (+29.4pp), exceeding MemRL's +11.0pp gain under comparable frozen-backbone conditions (noting backbone differences controlled for in our analysis). On HLE, entity graph retrieval reaches 48.0% from 25.2% (+22.8pp). Ablations show component value is task-dependent: rich judge feedback is negligible for code generation but critical for structured queries (+10.3pp), while binary-signal iteration partially compensates for weaker feedback.
Authors: Yanjia Huang, Yunuo Chen, Ying Jiang, Jinru Han, Zhengzhong Tu, Yin Yang, Chenfanfu Jiang
Abstract: The ability to transform a flat sheet into a complex three-dimensional structure is a fundamental test of physical intelligence. Unlike cloth manipulation, origami is governed by strict geometric axioms and hard kinematic constraints, where a single invalid crease or collision can invalidate the entire folding sequence. As a result, origami demands long-horizon constructive reasoning that jointly satisfies precise physical laws and high-level semantic intent. Existing approaches fall into two disjoint paradigms: optimization-based methods enforce physical validity but require dense, precisely specified inputs, making them unsuitable for sparse natural language descriptions, while generative foundation models excel at semantic and perceptual synthesis yet fail to produce long-horizon, physics-consistent folding processes. Consequently, generating valid origami folding sequences directly from text remains an open challenge. To address this gap, we introduce Learn2Fold, a neuro-symbolic framework that formulates origami folding as conditional program induction over a crease-pattern graph. Our key insight is to decouple semantic proposal from physical verification. A large language model generates candidate folding programs from abstract text prompts, while a learned graph-structured world model serves as a differentiable surrogate simulator that predicts physical feasibility and failure modes before execution. Integrated within a lookahead planning loop, Learn2Fold enables robust generation of physically valid folding sequences for complex and out-of-distribution patterns, demonstrating that effective spatial intelligence arises from the synergy between symbolic reasoning and grounded physical simulation.
Authors: Patrice Bechard, Orlando Marquez Ayala, Emily Chen, Jordan Skelton, Sagar Davasam, Srinivas Sunkara, Vikas Yadav, Sai Rajeswar
Abstract: There has been growing interest in building agents that can interact with digital platforms to execute meaningful enterprise tasks autonomously. Among the approaches explored are tool-augmented agents built on abstractions such as Model Context Protocol (MCP) and web agents that operate through graphical interfaces. Yet, it remains unclear whether such complex agentic systems are necessary given their cost and operational overhead. We argue that a coding agent equipped only with a terminal and a filesystem can solve many enterprise tasks more effectively by interacting directly with platform APIs. We evaluate this hypothesis across diverse real-world systems and show that these low-level terminal agents match or outperform more complex agent architectures. Our findings suggest that simple programmatic interfaces, combined with strong foundation models, are sufficient for practical enterprise automation.
Authors: Rafael Sojo, Pedro Larra\~naga, Concha Bielza
Abstract: This paper introduces two transfer learning methodologies for estimating nonparametric Bayesian networks under scarce data. We propose two algorithms, a constraint-based structure learning method, called PC-stable-transfer learning (PCS-TL), and a score-based method, called hill climbing transfer learning (HC-TL). We also define particular metrics to tackle the negative transfer problem in each of them, a situation in which transfer learning has a negative impact on the model's performance. Then, for the parameters, we propose a log-linear pooling approach. For the evaluation, we learn kernel density estimation Bayesian networks, a type of nonparametric Bayesian network, and compare their transfer learning performance with the models alone. To do so, we sample data from small, medium and large-sized synthetic networks and datasets from the UCI Machine Learning repository. Then, we add noise and modifications to these datasets to test their ability to avoid negative transfer. To conclude, we perform a Friedman test with a Bergmann-Hommel post-hoc analysis to show statistical proof of the enhanced experimental behavior of our methods. Thus, PCS-TL and HC-TL demonstrate to be reliable algorithms for improving the learning performance of a nonparametric Bayesian network with scarce data, which in real industrial environments implies a reduction in the required time to deploy the network.
Authors: Derek Austin
Abstract: Recent 3D Gaussian splatting methods built atop SMPL achieve remarkable visual fidelity while continually increasing the complexity of the overall training architecture. We demonstrate that much of this complexity is unnecessary: by replacing SMPL with the Momentum Human Rig (MHR), estimated via SAM-3D-Body, a minimal pipeline with no learned deformations or pose-dependent corrections achieves the highest reported PSNR and competitive or superior LPIPS and SSIM on PeopleSnapshot and ZJU-MoCap. To disentangle pose estimation quality from body model representational capacity, we perform two controlled ablations: translating SAM-3D-Body meshes to SMPL-X, and translating the original dataset's SMPL poses into MHR both retrained under identical conditions. These ablations confirm that body model expressiveness has been a primary bottleneck in avatar reconstruction, with both mesh representational capacity and pose estimation quality contributing meaningfully to the full pipeline's gains.
Authors: Smriti Jha, Matteo Paltenghi, Chandra Maddila, Vijayaraghavan Murali, Shubham Ugare, Satish Chandra
Abstract: Benchmarks that reflect production workloads are better for evaluating AI coding agents in industrial settings, yet existing benchmarks differ from real usage in programming language distribution, prompt style and codebase structure. This paper presents a methodology for curating production-derived benchmarks, illustrated through ProdCodeBench, a benchmark sourced from real developer-agent sessions. We detail our data collection and curation practices including LLM-based task classification, test relevance validation, and multi-run stability checks which address challenges in constructing reliable evaluation signals from monorepo environments. Each curated sample consists of a verbatim prompt, a committed code change and fail-to-pass tests spanning seven programming languages. Our systematic analysis of four foundation models yields solve rates ranging from 53.2% to 72.2%. We demonstrate how these offline evaluation signals drive practical decisions around model selection and harness design, while noting that offline benchmarks provide directional signal that we complement with online A/B testing for production deployment decisions. We share our methodology and lessons learned to enable other organizations to construct similar production-derived benchmarks.
Authors: Boyang Gong, Yu Zheng, Fanye Kong, Jie Zhou, Jiwen Lu
Abstract: Like a body at rest that stays at rest, we find that visual attention in multimodal large language models (MLLMs) exhibits pronounced inertia, remaining largely static once settled during early decoding steps and failing to support the compositional understanding required for cognitive inference. While existing hallucination mitigation methods mainly target perceptual hallucinations concerning object existence or attributes, they remain inadequate for such cognitive hallucinations that require inter-object relational deduction. Through token-wise attention analysis, we identify this visual inertia as a key factor: attention to semantically critical regions remains persistently focused and fails to dynamically support relational inference. We thereby propose a training-free Inertia-aware Visual Excitation (IVE) method that breaks this inertial pattern by modeling cognitive inference as the dynamic responsiveness of visual attention. Specifically, IVE selects visual tokens that are dynamically emerging relative to historical attention trends while distinguishing tokens exhibiting inertial behavior. To further facilitate compositional inference, IVE introduces an inertia-aware penalty that discourages over-concentration and limits the persistence of attention within localized regions. Extensive experiments show that IVE is effective across various base MLLMs and multiple hallucination benchmarks, particularly for cognitive hallucinations.