Authors: Antyabha Rahman, Akshaj Gurugubelli, Omar Ankit, Kevin Zhu, Aishwarya Balwani
Abstract: Large reasoning models trained via reinforcement learning (RL) have been increasingly shown to outperform their supervised fine-tuned (SFT) counterparts on mathematical reasoning tasks; Yet the mechanistic basis for this advantage remains unclear. We therefore ask, what internal representational differences enable RL models' superior performance? Our work presents two converging lines of evidence: First, linear probes trained on layer-wise hidden states reveal that RL models tend to achieve higher accuracy in predicting answer correctness compared to SFT models, indicating more linearly separable and structured representations. Second, mean ablation studies show that RL models develop a hierarchical architecture where deeper layers become progressively more critical, whereas SFT models distribute importance uniformly across layers. Together, these findings demonstrate that RL training fundamentally restructures how models represent and process reasoning problems. Finally, we analyze token-count variability under repeated sampling across problems to assess adaptive compute allocation. While we observe higher variability in some RL-tuned models than in their SFT counterparts, we see strong consistency in others, suggesting that token allocation may depend more on the overall training pipeline than on RL versus SFT alone. We believe this token-allocation variability reveals the spread of plausible on-policy reasoning, highlighting which models exhibit stable policies versus those that are under-determined, potentially non-identifiable solution behaviour.
Authors: Marylou Fauchard, Florian Carichon, Margarida Carvalho, Golnoosh Farnadi
Abstract: Large Language Models (LLMs)-powered multi-agent systems are increasingly deployed in mixed-motive environments, where agents operate under asymmetric information and strategic deception due to conflicting or hidden objectives. In these settings, misalignment with collective goals becomes a central concern. We propose a novel framework for evaluating objective misalignment using the social deduction game Werewolf, modifying the objective of a single agent while preserving its assigned role. Across LLMs from four different model families and sizes, four player roles, and three objective formulations, we introduce a dual analysis of the agents' internal reasoning and their public cheap-talk behavior (i.e costless, non-binding communication that does not directly affect the agents' utilities), complemented by an analysis of game outcomes. Our results show that objective misalignment undermines outcomes in inherently adversarial environments, an effect exacerbated by asymmetric information and specialized roles. While compromised agents consistently develop distinct objective-dependent reasoning strategies, these adaptations remain largely invisible in their public behavior. More broadly, our findings suggest that even subtle objective misalignment can profoundly affect collective decision-making, highlighting the need for effective mitigation strategies for LLM-based multi-agent systems.
Authors: Yuan Zhu, Ethan B. Liu, Frank Nie, Jindong Han
Abstract: Clinical data-science agents must transform heterogeneous longitudinal records into auditable analyses, yet existing benchmarks largely isolate medical question answering, structured-table reasoning, or generic scientific repositories. We introduce CLINLENS, a benchmark of 200 executable tasks over five linked MIMIC resources spanning structured electronic health records, notes, electrocardiograms, chest radiographs, and echocardiograms. A 4 x 5 taxonomy crosses four patient-time scopes with five analysis capabilities. Program-first reverse synthesis pairs each bounded semi-raw package with an evaluator-private reference workflow and checks required artifacts, cohort and temporal semantics, and the final answer. On a fixed 126-task suite, the strongest of 24 standardized model-scaffold configurations achieves 56.3% scope-macro STRICTPASS despite 100% EXECSUCCESS. For reference, a separately configured coding agent solves 83 of 126 tasks, while five biomedical systems adapted to GPT-4o-mini reach at most 2.9% scope-macro STRICTPASS. These results expose a substantial gap between runnable submissions and correct clinical analyses.
Authors: Brett Reynolds
Abstract: An AI benchmark result rarely reaches a consequential claim in one step. Evaluators generalize it to further cases, interpret it as evidence of capability, extrapolate it to new tasks, transport it to another system or site, and combine it with assumptions about human review and downstream consequences. Validity-centred approaches require evidence for each claim. This paper identifies a further epistemic problem: warranted links don't automatically make a warranted chain. The target of one study may not be the source of the next; system, population, outcome, or conditions may change at the interface; and shared data or model lineage may make apparently independent support dependent. Projectibility concerns whether a bounded extension from observed to unobserved cases is warranted. Goodman supplies the problem of rival extensions; argument-based validity supplies an architecture for testing them. The paper's distinctive claim is a non-composition principle: support for adjacent projections warrants their composition only when endpoints and assumptions align and dependence and uncertainty are carried through. A legal-research case shows how benchmark evidence and a deployment study can each be sound while remaining parallel. A reanalysis and simulation show why aggregate stability can erase distinctions a later projection requires. The resulting projectibility audit diagnoses unsupported joins in benchmark-to-use arguments.
Authors: Lang Cao, Yuhao Shen, Tianyang Luo, Simo Du, Hao Peng, Yue Guo
Abstract: Clinical practice guidelines (CPGs) encode diagnostic criteria, but LLM systems typically retrieve guideline text or absorb it through training rather than execute its rules. We introduce GuideSkill, an external reasoning layer that compiles disease-specific criteria into executable functions returning ordinal diagnostic-support scores. GuideSkill-Zero is initialized from guidelines, while GuideSkill-Evo uses case--diagnosis pairs to refine covered skills and add missing diagnoses. At inference, an LLM proposes a differential diagnosis, grounds the features required by each matched skill, and fuses its ranking with the executed skill scores. Across four benchmarks and four backbones, GuideSkill-Zero improves macro-average accuracy over guideline RAG by 13.45% on average. GuideSkill-Evo achieves the highest macro-average for every backbone, improves over direct inference by 18.49% relatively, and increases gold-label skill coverage from 56.5% to 99.5%. On Qwen3.5-9B, it also exceeds the strongest parameter-update baseline by 11.16% without updating the backbone. Expert evaluation further indicates that GuideSkill produces clinically sound and broadly acceptable skills, suggesting that its initialized and evolved rules are reliable and practically meaningful. These results support executable skills as a model-agnostic mechanism for combining guideline-derived procedures with case-derived diagnostic patterns.
Authors: Xin Xin, Jincheng Lou, Junhui Li, Jinglin Yan, Panda Xiao, Di Wu, Haixiao Li, Weicong Lu, Weijian Fan, Xinyu Qu, Yuxiang Zhao, Min Yu, Zhixiong Di, Yibo Lin
Abstract: Functional verification dominates integrated circuit (IC) front-end engineering effort, and a single missed bug that escapes to silicon can trigger a costly respin. Recent large language models (LLMs) offer new opportunities to automate this process, yet existing LLM-based approaches generate each component through independent single-turn calls with no shared context, leaving interface mismatches undetected and reported coverage disconnected from specification requirements. To address these challenges, we present GoGoTB, an agentic framework that achieves end-to-end verification closure through three subsystems: an agentic execution control layer, an evolvable knowledge system, and specification-grounded coverage closure. The execution control layer separates deterministic enforcement from LLM reasoning at every tool and stage boundary. The knowledge system dispatches methodology and design-specific expertise on demand. The coverage framework anchors every bin to a named specification behavior so that each residual gap has a diagnosable root cause and a targeted remedy. Tested on 8 register transfer level (RTL) designs without any human intervention, GoGoTB achieves 100\% environment generation success and averages 98.4\% line, 97.2\% branch, 97.0\% toggle, and 83.2\% functional coverage. No prior work successfully generates a complete verification environment or achieves meaningful coverage on the same benchmarks.
Authors: Sankalp Gilda, Shlok Gilda
Abstract: Evaluation methodologies for language models increasingly combine multiple signals, from automated metrics and LLM-as-judge ratings to human assessments and benchmark suite results. When these signals are aggregated via averaging, evaluation confidence can then substantially exceed the reliability of the weakest signal: a phenomenon we call trust inflation in evaluation. We argue that evaluation scores should be treated as epistemic claims with three properties: formality (human evaluation provides stronger evidence than an automated metric), scope (a benchmark result applies to the tested distribution, not universally), and validity windows (benchmark results expire as contamination accumulates and distributions shift). Several converging research traditions (chain-of-thought analysis, possibilistic logic, and algebraic theory) establish weakest-link aggregation as the conservative endpoint of a parameterized operator family controlled by a single pessimism parameter. Drawing on those traditions, and on concrete lessons from building an evaluation harness for agentic AI, we propose that evaluation results carry explicit metadata (formality tier, scope declaration, and expiration date) to make their epistemic status transparent. We illustrate the cost of mean aggregation on the public HELM leaderboard: across 54 frontier models on ten scenarios, the top-five models ranked by mean score and by weakest-link are completely disjoint.
Authors: Rwaida Alssadi, Muntaser Syed, Balaji Kasula, Lamine Deen, Majed Alotaibi, Mohammed Alghamdi, Tyler Ton, Ali Alqarni, Marius Silaghi
Abstract: Contemporary LLM-based coding agents produce code as black-box outputs: the rationale behind each line is hidden, the evolution of the code through benchmark-driven repair is ephemeral, and post-hoc auditing is impossible. We present a code generation concept that addresses these shortcomings through three complementary mechanisms: (i) a relational snippet-history schema that records, per repair event, the benchmark reference, round number, failure text, and LLM explanation, enabling full provenance queries; (ii) a browser-based visualisation tool that renders this history as heat-mapped, hover-annotated source code; and (iii) a competitive fractional position-key indexing scheme with tree-node delimiters that assigns stable, lexicographically-ordered identifiers to each code snippet, enabling fine-grained tracking without disrupting surrounding lines. We evaluate TraceCoder on 30 algorithmic programming tasks spanning string processing, mathematical computation, and data-structure manipulation, across two provider configurations. Of these, 10 exhaust the 6-iteration budget on tasks with subtle edge-case behaviour. Mean Chg% reaches 30%, three in ten code snippets carry a traceable repair-event row, compared to 21% when using Gemini 2.0 Flash as sole provider on a 20-task subset. Three detailed case studies demonstrate how the system explains which specific benchmark failures shaped each line of the final program. The proposed mechanism makes the internal "narrative" of automated code generation auditable and replayable, a property essential for trust and accountability in production deployments.
Authors: Wanyu Zhao, Wanbing Zhao
Abstract: An important skill in theoretical physics is to recognize when a new problem can be transformed into a known model. We study this skill as an AI-agent task: can LLM-based agents discover statistical mechanical mappings from a raw partition function to a tractable representation? To probe this question, we introduce StatMechBench-v0, a benchmark of six Ising-type problems covering transfer-matrix methods, gauge-removable disorder, and planar/Pfaffian structure. We evaluate a simple propose-verify-revise agent across multiple LLMs and problem phrasings. The results show that numerical feedback often helps agents repair code and recover correct partition functions. However, agents can also pass the numerical checks while misidentifying the underlying tractable class or understating computational complexity. This both reveals limitations in current LLM reasoning and calls for a verification stack that goes beyond numerical agreement, incorporating, for example, symbolic checks and structural invariants. Our study provides an early evaluation and design directions for AI agents aimed at structural discovery in theoretical physics.
Authors: Zheng Zhang, Nanjie Yao, Jiarui He, Deheng Ye, Peilin Zhao, Hao Wang
Abstract: Social deduction games (SDGs) such as Werewolf have become challenging testbeds for AI agents. These games require complex social skills such as reasoning, deception, and collaboration. While recent advances in large language models (LLMs) have driven significant progress in SDG agents, current approaches are predominantly text-based, overlooking the multimodal nature that is fundamental to human social interaction. To bridge this gap, we introduce CaM-Wolf, the first SDG agent that integrates multimodal perception and generation. CaM-Wolf processes video inputs from other players, employs a causal-aware Reasoner trained via reinforcement learning to establish logical chains between observable behaviors and hidden roles, and presents itself through an animated avatar. Our experiments and user study show that CaM-Wolf achieves superior agent gameplay performance and enhances the quality of human-AI interaction. This work represents a significant advancement towards creating more human-like AI agents capable of participating in nuanced social dynamics. Our code is available at https://3dagentworld.github.io/avatar_wolf.
Authors: Yiming Cai, Fangjie Yu, Meiqing Yu, Ziyue Shi, Pengfei Yuan, Yong Guo
Abstract: World models must learn the joint dynamics of states, actions, events, and observations, yet existing video, robotics, and simulation datasets usually capture only part of this structure. We introduce CG-World, a large-scale world-state dataset and protocol derived from industrial computer graphics production pipelines. CG-World explicitly records intermediate states, including multimodal semantics, spatial structure, skeletal and controller states, motion curves, camera and lighting parameters, physics caches, contact events, and multi-pass renderings. CG-World v1 contains approximately 850,000 temporally aligned segments of 1-5 seconds. It separates latent states, observations, relations, events, and branch metadata, and organizes them into unified spatiotemporal samples. To support intervention learning and counterfactual reasoning, CG-World defines a branch lineage covering factual trajectories, observation interventions, action interventions, mechanism interventions, and strict counterfactual branches, with intervention targets, invariants, and alternative outcomes explicitly recorded. We evaluate the dataset on geometry-conditioned video generation, action prediction, and closed-loop vision-language-action policy transfer. Results show that CG-World provides reusable structured supervision for controlled generation, action modeling, and embodied policy transfer. We plan to expand CG-World through continued data collection and community collaboration toward a shared data infrastructure for world models, Physical AI, and embodied intelligence.
Authors: Kawai Chung, Chunkit Chan, Yauwai Yim, Yuxuan Liu, Haochen Shi, Weiqi Wang, Qing Zong, Tianshi Zheng, Yixuan Fu, Kai Chung Wong, Hao Liang, Yifan Gao, Xi Yang, Janet Hui-wen Hsiao, Yangqiu Song
Abstract: Multimodal Large Language Models have sparked significant interest due to their potential for social intelligence; however, their ability to perform sequential motivation reasoning remains insufficiently studied. Existing evaluations predominantly examine static text or isolated visual snapshots, which do not reflect the cumulative nature of real-world behavioral drivers. To address this gap, we introduce MultivationBench, a benchmark designed to rigorously evaluate multimodal motivation reasoning within story-driven visual narratives. The benchmark builds upon established psychological frameworks - Maslow's hierarchy and Reiss's basic desires - and requires models to integrate accumulated multimodal context to infer evolving motivations. Results indicate that MultivationBench presents a significant challenge: all tested models struggle to maintain consistent motivation reasoning across sequential contexts, revealing a critical disconnect between static recognition capabilities and the dynamic reasoning essential for human-like social understanding.
Authors: Peng Yin, Kai Li, Yifan Zhang, Jian Cheng
Abstract: Physics-informed neural networks (PINNs) have emerged as a powerful paradigm for solving partial differential equations (PDEs), yet their performance heavily relies on the manual, trial-and-error engineering of neural representations, loss formulations, and optimization dynamics. While Large Language Models (LLMs) offer a promising avenue for automated design, unconstrained code generation often yields mathematically invalid or numerically unstable solutions under strict scientific computing constraints. To bridge this gap, we propose \textbf{EvoPINN}, an agentic framework that reformulates PINN development from labor-intensive manual design into a rigorous, execution-grounded algorithm discovery problem. EvoPINN navigates a modular search space by decoupling neural representations from training programs, utilizing an LLM agent to iteratively propose memory-conditioned programmatic modifications. To ensure scientific validity, all candidates undergo strict structural verification and budget-matched PDE evaluation. Extensive experiments across diverse PDE regimes (oscillatory, elliptic, dissipative, and nonlinear transport) demonstrate that EvoPINN discovers PDE-specialized learning algorithms that significantly reduce relative $L_{2}$ error compared to baselines. Crucially, EvoPINN autonomously invented SLRC-PINN, a novel architecture whose performance gains persist under rigorous parameter-matched comparisons, establishing the viability of execution-grounded agents for discovering genuinely new scientific computing mechanisms.
Authors: Gengyu Chen, Yongjie Yu, Weiling Wang
Abstract: AI agents can draft claims faster than authors can check whether the cited or retrieved evidence supports them. We study evidence-ledger adjudication: a claim-evidence traceability workflow that pairs each claim with an evidence packet, assigns a support relation, and routes unsupported, contradicted, or mixed-evidence claims back to the author. The empirical core is a 2,335-row blind benchmark built from independent external labels in AVeriTeC, CLIMATE-FEVER, and SciFact. Gold relations and source evidence labels are hidden during prediction and joined only for scoring. On this benchmark, the agent evidence-ledger condition achieves 0.676 relation accuracy and 0.601 macro-F1, compared with 0.383 accuracy and 0.303 macro-F1 for the best non-agent baseline. It also routes 1270/1435 claims whose gold labels indicate contradiction, missing evidence, or mixed evidence, while routing 295/900 supported claims. These results show that evidence-ledger adjudication can turn heterogeneous evidence packets into an auditable traceability layer for AI-assisted writing.
Authors: Shaopeng Wei, Yufei Cheng, Wenxi Sun, Yepeng Ding, Yu Zhao, Gang Kou
Abstract: The rapid development of large language models (LLMs) has renewed interest in agent-based modeling (ABM). However, current LLM-based ABM research faces several key challenges: modeling evolving agent-environment interactions, enabling flexible counterfactual reasoning, and automating simulation workflows for scientific research. In this paper, we propose Eco3S, a socio-economic system simulation framework for economic research and policy analysis that addresses these challenges through three key mechanisms: (1) Co-evolving Environment Design, a bidirectional feedback loop where agents and the environment co-evolve, producing realistic emergent behaviors; (2) Structural Causal Simulation, a structural causal model (SCM)-inspired counterfactual mechanism that allows flexible interventions for diverse causal inference tasks; (3) Simulation-Analysis-Refinement Paradigm, a self-corrective mechanism that iteratively refines experimental designs based on prior simulation results. Experiments on diverse economic scenarios confirm \textit{Eco3S}'s effectiveness in replicating multiple established economic studies (canal decay, origins of governance, and information propagation) and phenomena across domains. Additional results further demonstrate its scalability and generalizability, highlighting the framework's potential for rigorous economic research and policy-making.
Authors: Zijian Xu, Wenshuo Zhang, Zisen Qin, Rui Sheng, Yushi Sun, Huamin Qu, Chuhan Shi
Abstract: AI-assisted coding increasingly translates informal user intent into executable software, yet coding requests often contain ambiguities that recur in user-specific ways across tasks and sessions. Existing disambiguation methods typically address each ambiguous request in isolation within the current coding session, often through eliciting additional clarification. However, whether resolved session history from the same user can serve as memory for resolving recurring personalized ambiguity in a newly opened session remains underexplored. We formulate personalized ambiguity adaptation as a new task: given a user's previously resolved coding sessions and a new ambiguous request, an assistant should identify the recurring ambiguity pattern, produce the intended executable solution, and minimize clarification. To benchmark this task, we introduce CAPA, which characterizes personalized coding ambiguity through six mechanisms and injects these mechanisms into unambiguous executable tasks using a controlled three-stage generation pipeline. CAPA contains 600 coding sessions across 60 balanced user--ambiguity cells, including 300 held-out evaluation sessions. We evaluate 12 recent LLMs under no-history and same-user-history conditions using executable success, first-turn success, and turns-to-completion. Our analyses examine task difficulty, user identity, and memory-based history use, and we further propose same-user history gating as a lightweight inference-time method. CAPA provides a foundation for developing long-term coding assistants that better align generated code with user intent while reducing repeated clarification.
Authors: Jingyang Yi, Jian Yang, Yifei Jin, Yuqi Li, Jian Li
Abstract: Automated alpha mining has increasingly adopted large language model (LLM) agents for factor generation and iterative discovery. However, existing LLM-based systems often delegate both factor construction and search decisions to the agent itself, without an explicit exploration space or a principled mechanism for navigating that space. As a result, exploration remains largely implicit and difficult to control or optimize systematically. We introduce AlphaSchema, which constructs and explores a structured space of trading semantics for alpha mining. Each point in this space is a schema plan composed of Event, Context, Qualities, Direction, and Output, specifying the semantics of a candidate factor before implementation. AlphaSchema decouples exploration from implementation: an LLM translates selected schema plans into executable factors, while evaluated rewards are accumulated to learn a surrogate model over the semantic space. An iterative selection mechanism uses this model to balance global exploration, surrogate-guided exploitation, and local mutation. Experiments on the Chinese stock market show that AlphaSchema discovers factor pools with strong predictive and portfolio performance. Further analyses show that the semantic search process navigates diverse regions while increasingly allocating evaluations toward high-reward regions, and that implementations of the same schema plans by different LLMs exhibit comparable predictive quality, suggesting that alpha mining quality is largely robust to the choice of LLM within our framework.
Authors: Hongqiang Lin, Chao Liu, Xiaofan Bai, Xuan Jin, Yuhong Li, Nenggan Zheng, Xipeng Cao
Abstract: Enabling large language model (LLM) agents to accumulate and reuse experience from past interactions remains a central challenge in real-world applications. A promising solution is to treat skills as trainable states and optimize them in the same way as model parameters in neural network training. However, data-driven skill optimization is prone to overfitting to the limited trajectories collected from real environments. Overexploiting these trajectories overfits the current batch, while unconstrained exploration causes regression on previously solved cases. This tension motivates a constrained search view of skill self-evolution, governed by an exploration--exploitation trade-off. We propose SkillBoost, a three-stage framework that mitigates both risks: structured exploitation localizes observed failures to editable skill components, prior-guided exploration draws on prior knowledge in the LLM to generate diverse repair candidates, and verified acceptance commits a candidate only when it improves performance within a regression bound. Experiments across 23 model--benchmark configurations show that SkillBoost achieves state-of-the-art performance while mitigating overfitting, outperforming both human-crafted and LLM-generated skills. Transfer experiments further show that optimized skills can be reused by other agents on similar tasks.
Authors: Junhao Qiu, Zidong Wang, Yansong Sun, Zhitong Ma, Ping Guo, Qingfu Zhang
Abstract: Ascend C operator optimization is critical for NPU (Neural Processing Unit) inference performance but requires deep hardware expertise.While large language models (LLMs) have shown promise in automated CUDA kernel generation, the fundamentally different programming model of Ascend C introduces unique challenges that remain unexplored. In this paper, we propose AgenticCANN, a knowledge-augmented agentic evolution framework specifically tailored for automated Ascend C operator synthesis in low-corpus NPU environments.To overcome the severe platform knowledge deficit on unfamiliar hardware, AgenticCANN incorporates a knowledge-orchestrated generation system that delivers structured, multi-level domain insights across the development lifecycle to resolve the upstream feasibility bottleneck.Building on this foundation, it features a stage-adaptive agentic evolution strategy that dynamically aligns LLM interaction modes with specific generation and evolution phases, balancing high-exploration candidate discovery with high-convergence performance tuning.Extensive experiments on Huawei Ascend 910B across six operators spanning five pattern categories demonstrate that our method achieves 90 to 100 percent feasibility on elementwise and normalization operators, 56% on fusion operators, and up to 6.65$\times$ speedup on 1B Pangu model inference kernels. Further analysis reveals that knowledge injection monotonically improves feasibility from 57% to 86% on elementwise operators, demonstrating its general rather than operator-specific benefit.
Authors: Zhilun Zhou, Jianghao Yu, Yuming Lin, yongjun yang, Sun Yongquan, Depeng Jin, Yong Li
Abstract: Large language model (LLM) agents have been widely applied in automating data science tasks. However, existing methods typically rely on a limited set of provided datasets, and they face challenges in data-intensive scenarios that require discovering and leveraging relevant information from large-scale and heterogeneous data repositories. Urban tasks are representative examples of such scenarios, as urban data are not only large-scale and multi-sourced, but also exhibit complex spatial, temporal, and semantic relationships. To address these challenges, we propose UrbanDS, a graph-guided LLM multi-agent system for data-intensive urban tasks. We first construct a unified dataset graph to organize reusable dataset skills and the relationships among datasets. Specifically, we develop a Data Profiling Agent that constructs a skill for each dataset. Moreover, a Relation Agent identifies relationships among datasets and integrates these relationships into the dataset graph. At runtime, a Planner Agent retrieves task-relevant datasets from the graph and generates execution plans. Multiple Execution Agents then perform data processing and analysis, while their execution progress and intermediate results are shared through a common memory. Finally, a Report Agent synthesizes the experimental logs into a report, which can be further refined based on user feedback. To systematically evaluate the capability of agents in handling data-intensive urban scenarios, we further construct UrbanDS-Bench, an urban data science benchmark covering representative data analysis and modeling tasks. Experiments on both general and urban benchmarks demonstrate that UrbanDS consistently outperforms existing data science agents on data-intensive tasks. Furthermore, UrbanDS has been deployed on the urban operations platform of Dongxihu District, Wuhan, demonstrating its effectiveness in real-world urban applications.
Authors: Huixiang Zhang, Mahzabeen Emu
Abstract: Latent communication in large language model (LLM)-based multi-agent systems (MAS) transmits continuous internal representations instead of text, but greater representational capacity does not establish that the receiver uses task-relevant information. End-task performance alone also cannot reveal whether an observed effect depends on message presence, content generated for the evaluated example, or information supplied by a separate agent. We introduce a causal audit that applies controlled message replacements at the boundary where the sender-produced representation enters the receiver. Four message settings support five measurements of encoded sender information, receiver sensitivity to message presence and identity, the task value of example-specific content, and the additional value supplied by a separate agent. We apply the audit to latent relay with Qwen3-4B and Qwen3-8B on GSM8K, ARC-C, and MATH-500. On GSM8K, the Qwen3-4B overall performance effect of -1.00 percentage point decomposes into a -6.17-point effect retained by an other-example message and a +5.17-point effect attributable to example-specific content; both component directions reverse at 8B. On MATH-500, the Qwen3-4B gain of 15.00 points comprises 8.33 points retained by an other-example message and 6.67 points attributable to example-specific content, while the 8B gain is dominated by the former component. Self-substitution comparisons further show that example-specific content and other-agent value are distinct. These results show that aggregate accuracy does not identify how a latent message affects the receiver and motivate controlled message comparisons as a standard evaluation for latent communication.
Authors: Jule Schmidt, Maximilian Weininger, Clemens Dubslaff, David Parker, Nils Jansen
Abstract: Markov Decision Processes (MDPs) are widely used as decision-making models, commonly specified over factored state spaces through state variables and their valuations. The exponential blowup in the number of states renders many reasoning tasks in MDPs challenging. Abstractions are promising techniques to reduce MDPs and thus mitigate scalability issues. In this work, we introduce a notion of causality on factored MDPs and a novel property-driven causal abstraction technique that retains many characteristics of the original MDP model. For this, we rely on causal relations over state variable predicates and identify those states that share the same reasons for fulfilling or violating a given abstraction property. We theoretically and empirically compare various causal MDP abstractions using different model types such as MDPs, interval MDPs, or stochastic games. Our evaluation demonstrates the potential of our approach: For several standard benchmarks, we obtain small abstractions that allow us to compute near-optimal policies for the original MDP. Furthermore, our causal abstractions often generalize to related large-scale MDP models.
Authors: Jia Luo
Abstract: The key bottleneck in embodied AI is not model architecture but data. Although billions of human manipulation videos exist online, robots cannot directly learn from them due to the embodiment gap between human morphology and robot hardware. We introduce Pegasus, a low-resource framework that bridges this gap by translating human demonstrations into robot-learnable data through structured knowledge transfer. Instead of relying on raw video prompts, Pegasus constructs a graph-based intermediate representation: a Task Graph extracted from human videos is transformed through Affordance and Constraint Graphs into a Robot Planning Graph for robot-conditioned video generation. A hierarchical affordance latent space models the relationship between object states, affordances, and tasks, enabling generalization beyond object identities. A closed-loop physics verifier further filters invalid generations using kinematic feasibility, collision constraints, and joint limits. We evaluate Pegasus across a range of egocentric manipulation benchmarks, including GTEA Gaze+ and EPIC-KITCHENS-100, and diverse robot embodiments, assessing Task Correctness, Executability, State Consistency, and Learnability. Results demonstrate reliable cross-embodiment translation and show that robot data generation can be reframed from a hardware collection problem into a scalable, low-resource knowledge transfer problem.
Authors: Vishisht Choudhary, Lukas Schmidt, Anne Zo\"e Kenntner, Feras Skhab, Michel Osswald, Jens Ernstberger
Abstract: Bot detectors deployed at scale treat traffic as binary: human or bot. This assumption breaks when AI agents browse the web through browser automation, a traffic class that is neither and that binary classifiers structurally cannot represent. We present a three-class detection framework distinguishing humans, bots, and AI agents, and show that the binary-vs-agent confusion is architectural: a binary human-vs-bot detector misroutes agent sessions because its label space lacks an agent class. On our controlled benchmark, an MLP binary classifier misclassifies 39.1% of real AI agents as human and a SAINT binary transformer misclassifies 34.5%; adding an explicit agent class yields per-class agent F1 = 1.000 in all 30 runs (3 model families $\times$ 10 seeds). To measure evasion resistance, we construct a five-level evasion ladder spanning passive observation, GAN-generated trajectories, and replay of real human cursor data ($n = 2299$ evasion sessions). Across 10 seeds and 3 model families we observe zero agent misses in 22990 per-seed predictions. The discriminative signal is a browser-automation artifact, not evidence of agent reasoning: Playwright does not emit the raw pointer-move and wheel-delta streams a physical input device produces, and this absence signature survives trajectory manipulation. Exhaustive search over all feature subsets of size 1-5 (9401 GBMs) shows that two behavioral features (mouse_event_rate, teleport_click_ratio) give 100% observed agent recall at every evasion level with agent precision 0.994; five features lift macro-F1 to 0.991. The signal is redundantly encoded: removing teleport_click_ratio leaves agent detection at 100%. The single-feature regime is degenerate, flagging every agent only by collapsing the classifier to always predict "agent". Two features robustly isolate agents; five separate all three traffic classes at macro-F1 $\geq 0.99$.
Authors: Mehrad Yaghoubi, Azam Bastanfard, Abbas Jalilvand, Ashkan Rezaei
Abstract: Recent advancements in Computer Go, driven by AlphaZero and MuZero, rely heavily on Monte Carlo Tree Search (MCTS) to correct the errors of the neural network policy. While effective on massive computational clusters, this dependence creates a critical bottleneck on consumer-grade hardware, where the computational cost of tree management severely limits inference rates. Furthermore, without deep search, these models suffer from hallucination, proposing moves with high confidence that are strategically fatal. This paper introduces a novel Belief-Guided architecture that disentangles the Policy head from a distinct Belief head. Unlike traditional value functions, the Belief head acts as an internal simulator and independent critic, modeling epistemic uncertainty and strategic stability. By integrating memory mechanisms (Transformer/GRU) to handle long-term dependencies and the Ko rule, and utilizing a gating mechanism to filter overconfident policy errors, our model shifts the burden of intelligence from runtime search to parametric "intuition." Experimental results demonstrate that this approach significantly improves search-free win rates and reduces hallucination, enabling professional-level play on limited hardware where massive MCTS is infeasible.
Authors: Lingyang Zeng, Guangze Chen, Kaichen Yu, Zhicheng Pan, Siyang Weng, Zirui Hu, Xiangyun Du, Hailin He, Rong Zhang, Chengcheng Yang, Kai Huang, Xuan Zhou
Abstract: Personalized agents are increasingly applied to assist users across a wide range of tasks. Effective personalized assistance requires not only retrieving explicit facts from past interactions stored in agent memory, but also inferring abstract personal characteristics. However, existing memory benchmarks primarily evaluate whether an agent can retrieve information explicitly stated in conversational histories, failing to provide an effective assessment of deeper user understanding. In this work, we propose Setoka, a benchmark for evaluating memory-augmented personalized agents with hierarchical user understanding from heterogeneous data. Grounded in theories from cognitive and personality psychology, Setoka defines four levels of user understanding, i.e., semantic memory, episodic memory, behavior pattern, and personality trait. Moreover, to enable realistic yet privacy-preserving evaluation, we design a psychometrics-based pipeline that synthesizes diverse, coherent heterogeneous user data and queries at scale. Finally, we leverage Setoka to evaluate 3 language models combined with 5 memory systems for 10 synthetic users. Our comprehensive evaluation reveals that while existing systems perform well on semantic memory retrieval, their performance declines on episodic memory. Moreover, when dealing with behavior pattern and personality trait understanding tasks that require integrating heterogeneous and fragmented information dispersed over time, performance declines even further. These findings demonstrate that user understanding cannot be handled by simple fact retrieval, motivating the design of memory mechanisms for cross-source integration and abstraction over long-term user behavior.
Authors: Yongjian Guo, Wanlun Ma, Lingyu Shen, Xi Xiao, Sheng Wen
Abstract: Fine-tuning is the dominant paradigm for specializing large language models (LLMs), yet it exposes a critical vulnerability: malicious data providers can embed harmful behaviors into downstream corpora, creating models that retain professional skills while violating human values on demand. Existing safety-realignment defenses often fail in practice due to three key limitations: they frequently cause catastrophic forgetting of specialized skills; their effectiveness collapses when the defender cannot observe the attacker's prompt template; and successfully realigned models remain susceptible to re-jailbreaking via simple system prompt switches. To address these challenges, we propose Routing-based On-Policy Distillation (ROPD), a novel realignment framework that models the divergence between aligned and compromised output probability distributions rather than fitting specific prompt templates. We conduct extensive experiments comparing ROPD against four state-of-the-art baselines across three datasets and three base models with varying alignment strengths. Our results demonstrate that when baseline defenses face template mismatches, often accompanied by severe degradation in downstream task performance. In contrast, ROPD substantially mitigates template-mismatch risks, maintaining superior robustness in both defense effectiveness and capability preservation. While our analysis indicates ROPD is not entirely immune to template shifts, its performance degradation is negligible compared to existing methods, establishing a new standard for robust LLM realignment.
Authors: Yiping Song, Jiaoyan Chen, Renate Schmidt, Hui Yang, Wen Zhang
Abstract: Ontology matching (OM) has traditionally been formulated as either equivalence discovery or subsumption matching. The existing OM systems identify only one type of semantic correspondence and cannot simultaneously discover equivalence and subsumption mappings. In this paper, we introduce Hybrid Ontology Matching (HOM), a new OM task that unifies equivalence and subsumption discovery, and accordingly propose a Large Language Model (LLM)-based multi-agent OM framework AgentMap that is implemented by a series of interdependent semantic decisions. Given a concept in the source ontology, AgentMap integrates semantic retrieval, hierarchical search, and collaborative multi-agent LLM reasoning to progressively explore the target ontology, identifying either the equivalent concept, if one exists, or the most fine-grained subsumer. We further extend four OM datasets for a HOM benchmark and evaluate AgentMap under hybrid, equivalence-only, and subsumption-only settings. Experimental results show that AgentMap achieves promising performance on the hybrid setting, and at the same time outperforms equivalence matching and subsumption matching baselines on the equivalence-only and subsumption-only settings, respectively.
Authors: Suhas Thejaswi, Juhi Kulshreshta, Lutz Oettershagen
Abstract: Writing and communication are increasingly mediated by large language models (LLMs) that are being used to draft, revise and polish text. Although such assistance can improve clarity and help authors meet institutional expectations, widespread reliance on shared models may reduce population-level variation in linguistic form, a phenomenon we refer to as linguistic monoculture. We develop a mathematical framework in which authors and LLMs are represented as distributions over linguistic features and coevolve through repeated interaction. We analyze three interaction mechanisms: a shared model with a fixed linguistic distribution, a shared model recursively updated from author outputs, and personalized models updated through author-specific and population-level feedback. We characterize the resulting equilibria and convergence rates, showing that, shared models can drive authors toward a common norm, recursive feedback relocates the shared norm without altering pairwise spread under common conformity, and personalization can preserve a family of distinct author-model equilibria with nonzero linguistic diversity. We then endogenize conformity as a strategic choice trading off private benefits from clarity, legibility, and perceived fluency against distinctive style. Within this utility model, individually rational authors may conform more than is socially optimal because they do not internalize the value their distinctiveness provides to others, creating a negative externality and a price of monoculture that is finite for each fixed instance but can grow without bound when distinctiveness dominates authenticity. Synthetic simulations illustrate how fixed shared assistance, recursive feedback, and personalization produce different long-run diversity outcomes.
Authors: Jingbo Zhou, Yusai Zhao, Qi Bao, Jingjia Cao, Zhenghai Chen, Chang Gao, Kaiqi Guo, Muxin Guo, Mingxuan Li, Xinjiang Lu, Yanru Ma, Yixiong Xiao, Zenghui Zhang, Le Zhang, Hua Wu
Abstract: Large language model (LLM) agents are increasingly expected to assist users in completing tasks. However, existing benchmarks provide limited support for evaluating whether agents can carry out office-suite workflows at a reasonable cost. We introduce OmegaUse-OfficeVal, a benchmark for evaluating LLM agents on long-horizon office-suite tasks with task-level economic grounding. The benchmark comprises 100 tasks derived from office-suite requests proposed by practitioners and adapted through a privacy-preserving process. On average, these tasks require 2.32 hours of human labor to complete. An important feature of the benchmark is that each task is paired with two economic signals: human labor time and task price proxy. These signals enable direct comparisons between human costs and LLM inference costs, as well as value-weighted evaluation. To support stable evaluation, we develop code-based verifiers from fine-grained rubrics. We evaluate several frontier LLMs together with a human baseline. Although all evaluated LLMs are substantially cheaper and faster than human workers, they have not yet approached human-level deliverable quality. The code and dataset are fully open-sourced, and more information is available on our project website: https://omegause-officeval.github.io.
Authors: Peter Tisnikar, Maja Swieczkowska, Benteng Ma, Gerard Canal, Matteo Leonetti
Abstract: Effective collaboration with novel and diverse partners is a crucial skill for autonomous agents. Most current ad-hoc teamwork (AHT) approaches assume that agents will collaborate on a single, fixed task and that the partner's capabilities, their ability to successfully execute the desired action, are already known. In reality, a partner's true capabilities are often hidden, and human collaborators may act sub-optimally on tasks with multiple valid strategies. To address these limitations, we extend ad-hoc teamwork into a multi-task setting by re-framing it as a problem of joint planning with decentralised execution under hidden partner capabilities. We introduce CE-CM (Capability Estimation via Contextual Models), an approximate Bayesian method that infers task-invariant capability vectors. By using simulation-based sampling, the agent estimates capabilities and induces a contextual Multi-agent Markov Decision Processes for planning. This approach requires no population pre-training and refines its beliefs online from just a few tasks. To account for human unpredictability, we propose CE-CM-Div, an extension that evaluates capability hypotheses against diverse planner rollouts rather than a single optimal trajectory. Simulated experiments demonstrate that CE-CM rapidly recovers hidden capabilities, reduces infeasible action assignments, and adapts to changes over time. Furthermore, in an offline human study of 225 trajectories from 15 participants, CE-CM-Div substantially improved capability estimates over the baseline CE-CM method. Our results suggest capability-based modelling is a promising interpretable, task-agnostic representation in the studied settings, demonstrating that accounting for behavioural diversity is essential for robust human-AI teaming.
Authors: Peter Kirgis, Sayash Kapoor, Andrew Schwartz, Stephan Rabanser, David Africa, Konstantinos Voudouris, Viet Nguyen, Toby Pilditch, Magda Dubois, Harry Coppock, Cozmin Ududec, Nitya Nadgir, Matilda Orona, Tilman Bayer, Derrick Chan-Sew, Yue Ling, Abhishek Shetty, Helen Toner, Gillian Hadfield, Seth Lazar, Steve Newman, Shoshannah Tekofsky, Rishi Bommasani, Arvind Narayanan
Abstract: Forecasts of explosive AI progress hinge on AI agents automating AI research. But evidence on whether agents can carry out open-ended AI research is thin. Current evaluations either test agents on narrow, verifiable tasks, which excludes open-ended research, or submit AI-generated papers to blind peer review, which is overstretched, stochastic, and suffers from poor review quality. We introduce a third way to measure progress towards AI R\&D automation. An agent takes on the central, open-ended research question of a high-quality unpublished paper, and the paper's original authors grade its output. We call these shadow evaluations. We ran shadow evaluations on two unpublished NeurIPS 2026 submissions, giving frontier agents six days and thousands of dollars of compute. The agents completed all of the engineering without human help, yet could not make substantial progress towards answering the research questions. As a result, both papers were unambiguously rejected by the authors. We identify five recurring failure modes: poor judgment about the bar for publishable research, uncreative responses to shortcomings in the research design, ineffective backtracking from dead ends, poor resource awareness, and instruction drift. A robustness check with a second model and scaffold reproduced these failures. We release the expert reviews, survey responses, agent repositories, and logs. Our results provide early evidence that today's agents can do the engineering of AI research, but struggle with critical parts of the research lifecycle.
Authors: Guillermo GP-Lenza, Carmen DR. Pita-Romero, Miguel Fernandez-Cortizas, Pascual Campoy
Abstract: This paper presents a comprehensive methodology for implementing knowledge graphs in ROS 2 systems, aiming to enhance the efficiency and intelligence of autonomous robotic missions. The methodology encompasses several key steps: defining initial and target conditions, structuring tasks and subtasks, planning their sequence, representing task-related data in a knowledge graph, and designing the mission using a high-level language. Each step builds on the previous one to ensure a cohesive process from initial setup to final execution. A practical implementation within the Aerostack2 framework is demonstrated through a simulated search and rescue mission in a Gazebo environment, where drones autonomously locate a target. This implementation highlights the effectiveness of the methodology in improving decision-making and mission performance by leveraging knowledge graphs.
Authors: Jan-Lucas Uslu, Benjamin Nachman, Christopher Re
Abstract: The largest machine learning models in particle physics are also the most expensive to train, yet the return on scaling a given architecture cannot be estimated before that compute is spent. Scaling laws have been fit for jets, but none has yet been shown to predict the performance of models it was not fit on. We show that, for a generic transformer pretrained on collider jets, it can be forecast. Fitting a joint model-and-data scaling law on small models alone, spanning three orders of magnitude of training compute, we predict the loss of models trained afterward with more than one hundred times more compute to within one percent. We then connect the forecast to downstream physics performance: across two standard tagging benchmarks, lower pretraining loss yields systematically lower fine-tuning loss and higher background rejection after fine-tuning. Within this model family and these tasks, a compute budget can therefore be translated into expected physics performance before any large model is trained. The final frontier model is consistent with the published numbers for current state-of-the-art physics-aware foundation models trained on the same corpus, on accuracy, AUC, and quark/gluon rejection, with a residual edge for the physics-aware model only in the high-purity tail of top tagging. We release five pretrained models spanning multiple sizes, together with the complete training recipe and code.
Authors: Mateusz Koz{\l}owski
Abstract: Medical-imaging AI benchmarks combine datasets, DICOM rendering, prompts, provider APIs, automated labels, statistical code, manuscripts, and repository releases. Agreement across these artifacts is usually assumed rather than tested. We performed a retrospective forensic reproducibility audit of a preserved chest-radiograph vision-language model (VLM) pilot; no model was called again and no image or report was newly annotated. We traced prompt bindings, DICOM metadata, output completeness, label extraction, matched analyses, and release propagation. Of 300 planned model-prompt calls, 297 yielded nonempty reports. Sixty Claude calls labeled A/B were executed with the same C prompt. The 30 studies represented 28 patients. Four MONOCHROME1 images were rendered without required polarity inversion, dataset split membership was not retained, and the unvalidated extractor truncated five reports to 4000 characters. Reconstructing one common cohort of 369 complete case-finding blocks changed Cochran's Q from 154.73 to 182.29. Of 45 McNemar comparisons, 27 had unadjusted p < 0.05 and 20 remained below 0.05 after Holm adjustment. These values describe only the archived automated-label matrix; they do not recover the intended prompt comparison or establish clinical performance. We withdraw the original performance, ranking, prompt-effect, and clinical claims and specify machine-verifiable controls for cohort, DICOM rendering, prompt and model identity, call status, annotation provenance, keyed analysis, and derived artifacts.
Authors: Scott M. Norton
Abstract: We report a striking phenomenon: deep reinforcement learning agents trained with frozen, randomly initialized CNN feature extractors spontaneously develop extremely sparse fully-connected representations, without any sparsity-inducing objective. In the first fully-connected layer (FC1, $3{,}136 \to 64$), agents compress task-relevant information through as few as 1-3 neurons out of 64 for deterministic Pong (5-11 for stochastic Pong), while trainable CNNs activate 55-64 neurons under matched conditions. We establish four principal findings. First, FC1 sparsity scales with task complexity: 1-11 for Pong, 19-26 for Breakout, and $\sim$42 for Space Invaders. Width-scaling confirms this reflects task structure rather than a fixed capacity fraction. Second, within-game scaling emerges: three identical Pong seeds produce 5, 7, and 11 active neurons. The 5-neuron seed plateaus at $+14$ reward, while the others reach expert performance ($+18.4$, $+18.7$), suggesting the random projection's usable dimensionality bounds achievable performance. Third, ablation confirms necessity: removing these active neurons crashes performance across two PPO implementations and four games. Fourth, the information bottleneck commits early: a sweep shows the active set locks by 15-30M steps, while reward turns positive 35-105M steps later. A complementary finding in Breakout shows frozen and trainable CNNs reach competitive rewards via structurally different bottlenecks: frozen agents use 17-25 active neurons (participation ratio $\sim$10-14), while trainable agents use 51 (participation ratio $\sim$3.6). Finally, wherever input dimensionality dwarfs intrinsic task dimensionality, gradient descent on a frozen random projection may reveal the effective rank of the underlying problem without explicit sparsity machinery.
Authors: Cristovao Iglesias, Devesh Batra, Alankar Atreya, Stefan Wagner, Robert Hankache, Patrick Sinclair, Giulio Pelosio, Michael McMillan, Greig A. Cowan, Raad Khraishi
Abstract: LLM-based chatbots are transforming customer service in regulated domains such as banking, but scalable and cost-effective validation remains a critical barrier to safe deployment. We present a two-part contribution for large-scale chatbot validation. First, we introduce a methodology for creating high-fidelity synthetic customer agents (SCAs) as digital twins, grounded in real transactional and conversational data, that enables automatic generation and behavioral conditioning to simulate diverse customer profiles and interaction styles. Evaluation demonstrates that SCAs achieve high semantic alignment with real customers, low hallucination rates, and successful personality trait reproduction with controllable interventions. Second, we develop an SCA-based validation framework combining automated LLM-as-a-Judge evaluation, human expert testing, and adversarial probing. Scenario-based validation across emotional states, demographic groups, and linguistic factors confirms robust performance. Our approach was used to validate a customer facing chatbot at a leading UK bank, providing financial institutions with a scalable pathway toward regulatory compliance.
Authors: Mouad Zemzoumi, Amine Abouaomar
Abstract: Pre-match tactical decision-making in professional football relies heavily on subjective expert analysis and identity-based scouting systems that cannot generalize to unseen teams. This paper presents Sim2Win, a team-agnostic, event-based pre-match tactical recommendation framework that reframes match outcome prediction as a tactical decision-support problem. Using StatsBomb open event data from eleven competitions spanning 178 teams and 1,411 team-match records, Sim2Win constructs five-match rolling tactical profiles, engineers four interpretable tactical feature ratios, clusters team behaviors into eight playstyles via K-Means, and trains thirteen classifiers to estimate win, draw, and loss probabilities from tactical matchup representations. The system operates without team names or identity features, enabling generalization to teams never seen during training. A rigorous Leave-One-Competition-Out (LOCO) evaluation demonstrates that Sim2Win achieves a mean ROC-AUC of 0.704 and mean accuracy of 55.4% on completely unseen teams, outperforming ELO, Pi-Rating, and GAP baselines on all 21 ROC-AUC comparisons and 19 of 21 accuracy comparisons. Among all evaluated models, CatBoost achieved the strongest in-distribution performance with 60.90% accuracy. These findings suggest that behavioral tactical representations provide transferable predictive signal under distribution shift and offer a viable alternative to identity-dependent football prediction systems.
Authors: Karly V. Coffey, Gloria L. Krahn, John P. Hanley, Jacob E. Neely
Abstract: Background: This work investigates the presence of implicit bias in Large Language Model (LLM)-based chat AI models directed toward people with intellectual disabilities (ID). Objective: The study aims to identify and measure representational differences related to people with ID and examine them to identify implicit biases inherent in AI chat generation technologies. Methods: Utilizing the GPT-4-Turbo model, we requested story-generation based on 10 prompt stems with and without descriptors for ID. This process was repeated using four other LLMs (OpenAI GPT-4o, Meta Llama-3-3-70B-Instruct, Anthropic Claude-3-5-Sonnet, and Mistral-Large-2411). The resulting 25,000 computer-generated stories were analyzed using a separate GPT-4-Turbo model instance to detect differences in how people are represented related to themes of bias described in previous literature. Results: Our findings reveal differences in how people are represented between story datasets with and without ID descriptors. These differences go beyond established characteristics of ID and imply the presence of mostly negative implicit biases. Identified differences related to considering people with ID as younger, with themes of paternalism and infantilization; depicting them as more inspirational and symbolic; as needing help more often, being dependent, and being saved; and having a negative perception of them and more hesitation to include them. Conclusions: These implicit biases are considered within the context of past discrimination towards people with ID and highlight the need for diligence against implicit bias towards people with ID in AI development. This research underscores the importance of assessing and mitigating implicit bias in decision-making technologies to prevent future societal harm.
Authors: Benjamin Mawdsley, Tom Quilter, Richard Turner, Sarah Jackson, Paul Edwards
Abstract: Personalised learning systems often assume that mathematical ability is combined of discrete abilities, acquired sequentially and dependent upon first acquiring foundational abilities, and students often report different strengths. In this work, we explore the validity of these assumptions by applying clustering methods to a large dataset of 119,034 students, spanning 13 national-level exams sat in the United Kingdom and collected by the platform. Classifying question results as pass or fail, we use a Bernoulli Mixture Model to search for latent populations which would be indicative of discrete skill-sets. We find that few distinct clusters are present in the data and that the dominant factor is overall student ability, which is further supported by the high degree of linear correlation between the probability distributions of the resulting clusters. Our best performing model achieves an accuracy of 78 percent, competitive with more complicated models in the literature whilst being more explainable. Comparing this models performance with logistic regression baselines and with k-nearest neighbours, we find a small improvement when using performance on each individual question as features. This suggests that whilst overall ability level is the dominant factor for predicting performance, small further personalisation improvements can be made by tailoring to a students exact strengths, but that students do not appear to develop strongly differing ability across topics. Our work offers a national scale test of machine learning in education and offers a new benchmark for the field, demonstrating how explainable models can reach competitive performance
Authors: Belinda Mo
Abstract: AI systems are becoming autonomous research agents that generate hypotheses, design experiments, and produce discoveries at scales beyond human oversight. As seen by increased submissions to ML venues, the verification gap between scientific output and our ability to check it is already widening, and autonomous agents make it worse by magnitudes given human-agent asymmetry. We argue that science must evolve its verification infrastructure, as it has before with peer review. However, while historical adaptations assumed human contributors who could be questioned and sanctioned, AI agents break this assumption. We propose criteria for an adapted verification infrastructure that emphasizes observable-by-default workflows, scalable verification, and clear attribution. We argue that without adaptation, ML and any scientific domain using agents face dangerous failures: experimental results that no person can verify, optimization for metrics over understanding, and accountability vacuums that erode scientific trust.
Authors: Ranjitha Shivaprasad Ballakuraya, Arash Mahyari, Ashok Srinivasan
Abstract: The growing volume of scientific submissions has motivated interest in using large language models (LLMs) to assist peer review. Existing automated novelty assessment approaches typically compare a paper's claimed contributions against prior literature, implicitly assuming that these contributions are accurately realized in the work itself. Human reviewers, however, frequently challenge novelty claims not because similar ideas already exist, but because the methodological evidence presented in the paper does not adequately support them. This internal mismatch between claimed contributions and methodological realization is rarely examined by current LLM-based review systems. To address this gap, we introduce intra-paper claim verification, a framework that evaluates whether novelty claims articulated in a paper are substantiated by the methods used to realize them. The framework employs an LLM to extract novelty claims from the introduction, retrieve claim-relevant methodological evidence, and assess whether the methods substantiate the stated contributions. Assessment is guided by reviewer-inspired evaluation criteria derived inductively from human peer reviews collected from 182 ICLR 2025 papers. These criteria capture recurring reviewer concerns related to novelty, methodology, clarity, and other issues and are used to generate structured reviewer-style assessments of claim substantiation. We evaluate the framework by comparing LLM-generated review comments against human reviewer concerns on a balanced subset of accepted and rejected papers. Human evaluation demonstrates significant alignment between framework-generated assessments and human reviewer concerns, particularly for novelty-related issues. BERTScore further distinguishes corresponding human-LLM review pairs from mismatched controls, indicating that the framework captures concerns consistent with human reviewer observations.
Authors: Amanda La Hadi, Muhammad Johan Alibasa, Guanliang Chen, A. Taufiq Asyhari
Abstract: Large language models (LLMs) are increasingly used for estimating item difficulty in educational assessment. However, it remains unclear whether such estimates reflect how learners actually experience difficulty. This study investigates the alignment between LLM-generated difficulty ratings and empirical student performance on basic mathematics tasks. Four widely used LLM-based systems generated difficulty ratings on a 1-100 scale for 32 arithmetic items across multiple runs (N = 640 ratings). These were compared with empirical difficulty derived from responses of 770 Indonesian undergraduates using Classical Test Theory (CTT) and Item Response Theory (2PL). Results show moderate rank correlations (Spearman's rho = 0.52-0.70), indicating that LLMs capture coarse ordering of item difficulty. However, substantial and systematic misalignment emerges in fraction items. Several items consistently rated as easy by LLMs were among the most difficult for students, such as an item with only 34.16% correct for 100 : 1/2. We argue that LLMs approximate curricular difficulty, or what should be easy based on instructional sequencing, rather than cognitive difficulty driven by learner misconceptions. This leads to systematic underestimation of misconception-driven items, a phenomenon we term the Easy Trap. These findings highlight a critical limitation of LLM-based difficulty estimation and suggest that relying on such estimates without empirical grounding may introduce bias in assessment design and adaptive systems.
Authors: Sivasathivel Kandasamy
Abstract: Existing AI governance frameworks, including the EU AI Act and NIST AI RMF, address safety, transparency, and accountability but do not operationalize quantitative constraints on macro-socioeconomic stability. As a result, AI systems may satisfy regulatory requirements while contributing to labor displacement, rising inequality, and reduced economic resilience. We introduce the Human Utility Factor (HUF), a differentiable welfare metric that models the interaction between Agency, Wellbeing, and Economic Stability as functions of three actionable policy levers: automation depth, redistribution intensity, and employment coverage. HUF yields a closed-form optimal automation level and a minimum redistribution threshold below which no level of automation is welfare-positive, transforming high-level governance objectives into computable constraints. We evaluate HUF using a three-agent multi-agent reinforcement learning framework across U.S., Canadian, and Nordic policy regimes. Both analytical and PPO-based agents identify welfare-optimal operating regions and reveal a critical failure mode: welfare metrics that do not explicitly constrain redistribution can converge to high-automation equilibria that satisfy the metric while undermining its intended societal objectives. Our results suggest that AI governance is fundamentally a constrained optimization problem rather than a compliance exercise. HUF provides a quantitative framework for evaluating automation policies, identifying socioeconomic stability boundaries, and supporting governance decisions under accelerating AI deployment.
Authors: Gil Gekker, Rachel Steratore, Everett Smith, Asher Brass-Gershovich, Varun Gandhi, Nicole Nichols, Vijay Bolina, Buck Shlegeris, Lisa Einstein, Dan Lahav, Omer Nevo, Sella Nevo
Abstract: As AI systems are rapidly integrated into critical economic, governmental, and national security functions, the gap between AI adoption and AI security readiness continues to widen. This paper presents a prioritized agenda for advancing AI security, informed by structured interviews with leaders across industry, government, and civil society, and refined through a multi-sector expert workshop. Participants identified and ranked the highest-importance and most cost-effective areas where progress could strengthen AI security - from protecting frontier AI systems and their underlying infrastructure to improving cybersecurity practices as AI reshapes the threat landscape. The resulting priorities are organized across four themes: establishing strategic foundations and policy frameworks; advancing public-private coordination and institutional infrastructure; advancing technical security engineering and assurance; and governing agentic AI under adversarial pressure. For each priority area, expert authors provide detailed analyses that define the problem, assess the current landscape, and identify actionable projects that stakeholders across sectors can pursue. The paper aims to serve as an initial practical foundation for coordinated investment and action across the AI security field. It is designed to serve both current practitioners and individuals and organizations looking to enter the field by identifying concrete, high-impact contributions suited to a range of strengths and capacities.
Authors: Guanming Xiong, Penghui Zhang
Abstract: Large language model (LLM)-based agentic search systems are often evaluated as if the underlying LLM were the only component that matters, yet their measured performance also depends on the surrounding search environment: the Wikipedia snapshot, preprocessing pipeline, chunking policy, retrieval backend, tool schema, observation format, and answer submission rule. These details are frequently under-specified, making it difficult to compare results or reproduce reported baselines. We present SimpleWikiSearch, whose corpus construction, retrieval stack, tool contract, and evaluation protocol are explicit and runnable. The environment starts from a full English Wikipedia dump, cleans and chunks the corpus, builds keyword and dense retrieval indexes, and exposes a minimal tool interface consisting of \texttt{search}, \texttt{open\_url}, and \texttt{submit\_answer}. We report baseline results on six QA datasets using open-source LLMs and provide a random-300 subset for comparisons with closed-source commercial models. SimpleWikiSearch provides a domain-specific agent harness and a controlled offline environment for reproducible agentic-search evaluation. Its contribution is this specified reference setup, rather than a new agent algorithm. Code and data will be available at: https://github.com/JimXiongGM/simple_wiki_search.
Authors: Matthijs Jansen op de Haar, Tobias St\"ahle, Lorenzo Gatti
Abstract: In this work, we propose GuidedRAG, a novel extension to traditional Retrieval-Augmented Generation (RAG) that introduces a dedicated selection stage and semantic steering during retrieval. In contrast to current state-of-the-art RAG approaches, which depend on increasingly complex retrieval and knowledge structures, GuidedRAG constrains the knowledge base using semantics before retrieval, aligning the retrieval space with user intent while substantially reducing the search space. Our evaluation shows that GuidedRAG improves retrieval relevance by 14.0-15.8%, mitigates a 19.7-27.4% loss in retrieval precision, and reduces retrieval overhead by orders of magnitude. Moreover, relevant chunks are consistently retrieved earlier in the ranking process, while alignment with user intent improves by 31.8-36.8%. We further show that GuidedRAG achieves full coverage across 15 diverse RAG variants, demonstrating generalizability across the literature. Together, these findings establish semantic steering and selections as a powerful and generalizable paradigm for improving the current state-of-the-art in RAG.
Authors: Changyu Du, Alexander Vosseler, Filippo Mazza, Andr\'e Borrmann
Abstract: Long-term memory is becoming a core capability of LLM-based agents, but existing evaluations largely test conversational recall in open-domain or persona-grounded settings. We argue that a stronger test is whether an agent can reuse information from prior sessions while acting over a live, structured, domain-specific environment. We study this problem in Building Information Modelling (BIM), a professional engineering workflow where agents must query large IFC models while also relying on project specifications, client decisions, and engineering conventions often discussed in conversation but absent from the model. We introduce IFCMemoryBench, a benchmark for evaluating long-term memory in LLM-based BIM information retrieval. IFCMemoryBench contains 143 multi-session tasks across 19 projects and 4,016 prior sessions, derived from incomplete-information questions in IFC-Bench v2. Each task seeds missing project context across earlier conversations and later asks a probe question that can be answered only by combining remembered context with live IFC queries. Our evaluation framework decomposes memory performance into ingestion, retrieval, and utilization, and measures both answer quality and memory quality with expert-validated LLM judges. We evaluate representative vector-, graph-, and file-based memory systems. The strongest system achieves only 32.4% answer accuracy under a deployment-realistic ingestion scope, and remains below 60% under oracle-filtered ingestion or a stronger probe agent. Analysis shows that current general-purpose memory systems often retrieve topically relevant context but store project knowledge as incomplete or fragmented facts. These results reveal a domain-transfer gap in agent memory and suggest that reliable professional agents require domain-aware memory representations linking conversations, project knowledge, and structured model entities.
Authors: David Kaleko, Sergey Ivanov, Md Mofijul Islam
Abstract: We present IDP AutoOpt, an autonomous LLM agent that discovers high-performing configurations for intelligent document processing (IDP) pipelines. Tuning IDP prompts, models, OCR settings, and schemas jointly currently costs domain specialists 20 to 80+ person-hours per document type and does not scale as enterprises add document classes. IDP AutoOpt runs a closed loop: it scores a configuration on a small labeled set, diagnoses field-level errors, generates targeted edits, and re-evaluates, guided by human-authored domain skills that encode production expertise. Across extraction, classification, and packet-splitting tasks deployed in healthcare, marketing-intelligence, and financial-services settings, IDP AutoOpt matches or exceeds human-expert accuracy at equal or lower cost (on an extraction benchmark, 90.2% vs 81.6% at 4.6 x lower per-page cost), cutting configuration time from weeks to under two hours. We further show that agent LLM capability has a hard threshold below which optimization fails, and that curated domain skills outperform raw source-code access, which can degrade performance when provided without structure. We also share practical lessons on context management and variance mitigation. Requiring only a configurable pipeline, a scoring function, and a small labeled set, the approach extends beyond IDP to other enterprise AI systems, such as RAG and multi-agent workflows, where configuration bottlenecks deployment.
Authors: Lingteng Zeng, Yifan Jin
Abstract: Retrieval-augmented generation services over mutable enterprise documents repeatedly execute semantically equivalent analysis requests. Answer reuse can remove GPU-bound generation work, yet response caches require dependency consistency when filings, evidence chunks, and tool outputs change. FinCacheServe treats each generated answer as a serving object indexed by enterprise intent and guarded by document versions, evidence fingerprints, tool fingerprints, model identity, and decoding configuration. A vLLM implementation evaluates SEC-derived financial-document workloads with Qwen2.5 models. On a 2,230-request hosted 7B trace, FinCacheServe skips 53.27% of LLM calls with zero observed dependency-stale outputs. Across three hosted 32B operator-suite seeds, it skips 53.31% of 544 requests, compared with 38.97% for versioned semantic caching and 22.43% for grounded-style reuse. Capacity, backend, and SLO replays show oracle-bounded cache management, 100k-entry transactional metadata behavior, and 44.30% lower estimated Wh per dependency-fresh 2s-SLO success than versioned semantic caching.
Authors: Fangnian Wang, Nicholas Tan Jerome, Thomas Jordan, Frank Simon
Abstract: Hydrogen infrastructure in enclosed environments, such as parking facilities for fuel cell vehicles, presents significant safety challenges due to hydrogen's low ignition energy and wide flammability range. Current monitoring systems are largely reactive, detecting leaks only after hazardous concentrations have formed. This study develops a computational framework for proactive sensor placement optimization by integrating computational fluid dynamics (CFD), genetic algorithm (GA) optimization, and a DeepSets neural surrogate. A CFD database of 180 scenarios was generated for a representative 50 m x 30 m x 3 m garage, covering multiple leak positions, rates (1-150 g/s), and ventilation conditions (ACH = 3-10 per hour). Sensor placement was optimized using a multi-objective GA and compared with uniform, random, and surrogate-assisted approaches. The GA achieved a detection rate of 96.1% within 60 s and reduced blind areas to 0.12%, corresponding to an approximately 5% improvement in composite fitness over a uniform baseline. The DeepSets surrogate reproduced near-optimal configurations with a fitness gap below 0.01 while reducing CFD evaluations by 89% and computational time by two orders of magnitude. Detection performance and spatial coverage remained comparable to the GA, demonstrating that surrogate-assisted optimization can retain solution quality while enabling rapid design iteration. Overall, the results show that CFD-informed optimization improves detection effectiveness and reduces sensor requirements compared to conventional layouts. The proposed framework supports scalable deployment and provides a foundation for integrating optimized sensor networks with digital twin systems for real-time monitoring and risk assessment.
Authors: Vivek Shukla, Varun Shukla, Atul, Divya Mishra, Mehul Kumar Das
Abstract: Mathematical chain of thought (CoT) evaluation is commonly reduced to whether the final answer matches a reference. This conflates producing a correct conclusion with producing a valid derivation an invalid chain can accidentally reach the right answer, while a valid calculation can be followed by a transcription error. We call this mismatch the reasoning answer consistency gap. This framework paper introduces the Reasoning Answer Faithfulness Score (RAFS), a reference free, instance level diagnostic of whether an emitted mathematical trace is locally credible, supports its answer, and is stable under resampling and targeted counterfactual interventions. RAFS combines step validity, reasoning to answer entailment and counterfactual sensitivity, answer consensus, and conditional reasoning stability. It evaluates transcript level agreement, not a models private computation and not factual correctness outside the tested mathematical setting. We retain a preregistered, results blind confirmatory study on GSM8K and MATH, with hypotheses, admissibility rules, calibration, and tests fixed before confirmatory outcomes are inspected. A separate feasibility pilot is specified to verify end to end execution and estimate interven tion coverage before that freeze numerical pilot claims are re ported only when trace level artifacts are available. We formalize four reasoning answer outcomes, justify the non compensatory aggregator, instantiate semantic trace distance, quantify compute and abstention tradeoffs, and define verifier independence and power analyses. RAFS is intended to complement mathematical answer accuracy with an auditable warning signal for silent reasoning failures and answer extraction errors
Authors: Hira Yaseen, Arif Mahmood, Waqas Sultani
Abstract: A person's physical characteristics such as weight and height are important indicators of his physical and mental health, daily life routines and finances. Body Mass Index (BMI) is a well known measure that encodes the characteristics of both the weight and the height. BMI has been used as a self-monitoring tool, and it has long-term implications on one's life. For example, it may help predicting the risk of various diseases and estimating longevity. Automatic BMI estimation using a single person image in the wild is a challenging task due to wide variations in human pose, camera geometry, personal appearance and distracting backgrounds. In this paper, we explore the performance of deep neural networks using single and multi-task learning by employing different modalities including RGB, depth-maps, pose-affinity maps, and edge-maps to predict BMI, weight, and height from daily life images available on social networking websites. Currently, no full body image dataset for BMI estimation is publicly available, therefore we propose a new dataset consisting of 6105 images with ground truth labels of height, weight and BMI. Our proposed dataset is collected in the wild containing images from various ethnicity and distributed over varying age groups and gender. It consists of frontal, back, full and half body, side poses, mirror selfies with varying backgrounds and scale variations and may contain artifacts hiding partial or full face. Extensive experimentation is performed using full body, half body and face images only using different CNN backbones including VGG, Densenet and ResNet. Our experimental results demonstrate that full body images have produced better results than the other half body and facial images in the wild.
Authors: Xinran Liu, Shouqian Shi, Yutong Chen, Ge Wang, Xin-Wei Yao, Sheng Zhong
Abstract: Dense vision-language understanding, including object localization, region recognition, and open-vocabulary semantic segmentation, requires associating language concepts with spatially grounded visual regions. CLIP provides a strong foundation for these tasks by learning a shared image-text embedding space from large-scale contrastive pre-training. However, its image-level objective aligns text with a CLS-derived global representation, leaving local vision-language correspondence only indirectly constrained. Existing methods either introduce additional supervision, external models, or task-specific adaptation, while training-free approaches mainly recover dense responses from existing patch features without examining where local semantics become most accessible within CLIP. We introduce TraceCLIP, a training-free framework that recovers latent patch-level semantic evidence by isolating the patch-specific terms written into the CLS attention output. TraceCLIP further converts contribution-derived semantic responses into a semantic-geodesic topology gate that calibrates final-layer patch affinity for dense feature reconstruction. Diagnostic experiments show that these contribution features exhibit strong local semantic discrimination and text-conditioned spatial alignment. On eight zero-shot semantic segmentation benchmarks, TraceCLIP achieves gains of 1.3 to 4.5 points in average mIoU over the strongest prior training-free methods across both backbones and background settings, without additional training, external vision foundation models, or region-level supervision. More broadly, these findings suggest that spatially localized semantics may remain accessible within the internal construction of globally aligned representations.
Authors: Eric Wallace, Christopher A. Choquette-Choo, Nikhil Kandpal, Sam Toyer, Dylan Hunn, Stephanie Lin, Yuxin Wen, Xiangyu Qi, Christopher Wolff, Zizhao Wang, Milad Nasr, Sicheng Zhu, Chuan Guo, Juan Felipe Cer\'on Uribe, Kaiwen Wang, Aiden Low, Kai Xiao, Kai Chen
Abstract: We introduce \textbf{GPT-Red}, an automated red-teaming agent that is trained to discover novel prompt injection attacks against frontier LLMs. The goal of this model is to evaluate and improve the robustness of our production systems. To this end, we use it to adversarially train GPT-5.6, our most robust model to prompt injections to date. To create GPT-Red, we design a scalable self-play algorithm where the model is tasked with attacking a diverse population of simultaneously-trained defender agents. We train the model on realistic red-teaming environments using compute on the same scale as some of our largest RL post-training runs, making it the single-largest LLM safety training run ever documented. GPT-Red excels at red-teaming: it reliably breaks our past models up to GPT-5.5, it finds more successful attacks than human red-teamers, and it generalizes to held-out environments, defender models, and harnesses. In the future, we expect that as we improve the robustness of each new GPT model, it will in turn will provide better learning signal for \textit{even stronger} red-teamer agents, thus unlocking a self-improvement flywheel.
Authors: Yuvraj Verma
Abstract: Self-repair - returning a failed program to the model together with its test output and asking for a correction - is a standard component of code agents, and is almost always evaluated against a baseline that does not retry at all. We argue that this comparison confounds the value of the feedback with the value of the extra attempt. Using a placebo-controlled design on MBPP+ at three model scales (1.5B, 3B, 7B), we compare four matched-budget retry conditions: blind resampling, a content-free failure notice, genuine execution feedback, and feedback augmented with verbal self-reflection. Blind resampling is the strongest condition below 7B, and remains statistically tied with the best condition at 7B, while consuming 2.5-5.5x fewer tokens; conditioning on the model's own failed attempt costs 6.1 points at 1.5B (p=0.006), and the informational content of execution feedback adds nothing measurable over the placebo. We attribute this to anchoring: when shown its previous attempt, a model reproduces a near-identical program in 33-68% of retries, against 2-14% under blind resampling. Two further experiments delimit the effect. Retrieved solutions to other tasks change nothing (bounded to +/-3.5 points), which localizes the harm to self-conditioning rather than context length; and reflection, the only condition that measurably weakens the anchor, remains dominated on cost. Replication rules out two competing explanations: the penalty is unchanged at full precision, and it reproduces on an independent model family. Across six configurations spanning two families and two precisions, its magnitude is predicted by baseline quality alone (r=0.96) - the cost of anchoring is the cost of committing to a bad first attempt.
Authors: Xinyu Yang, Tianxing Chen, Honghao Su, Minxuan Wang, Chenze Yu, Zhangzheng Tu, Yue Chen, Yuxiao Huo, Lingfeng Zhang, Yan Huang, Yan Qin, Shaolong Zhu, Qiwei Liang, Hekun Tian, Shujia Liu, Guangyu Chen, Junhao Gong, Zixuan Li, Wenwei Lin, Zijian Lin, Wenxuan Zhu, Eric J Chen, Yue Yuan, Qize Yu, Jiaqi Liang, Haowen Yan, Hengfei Zhao, Weijie Wan, Zikun Xiao, Junyuan Tang, Baijun Chen, Kai-Chong Lei, Kaixuan Wang, Kailun Su, Zanxin Chen, Yao Mu, Renjing Xu, Chuqiao Lyu, Qi Xiong, Ping Luo, Wenbo Ding
Abstract: Embodied intelligence integrates learned perception and decision making with real-time computation, control, and physical interaction. Because failures can cause immediate physical or operational harm, task completion alone does not establish trustworthiness. We define trustworthy embodied intelligence as the sustained capacity to execute specified tasks reliably under environmental and system variation while maintaining risk within acceptable bounds. We term this objective sustained safe success. Its supporting mechanisms are organized into four interdependent layers. The model layer generates task-competent action proposals with calibrated uncertainty and explicit safety preferences. The system layer realizes authorized actions dependably through integrated sensing, computation, control, hardware safeguards, fault containment, and fallback. The evidence layer substantiates bounded claims through evaluation, verification, validation, traceability, and structured assurance arguments. The deployment layer maintains claim validity through runtime monitoring, authority management, intervention, incident response, and controlled updates. Because assumptions and failures propagate across these layers, neither model capability, isolated safeguards, nor benchmark performance alone can establish end-to-end trustworthiness. Drawing on embodied AI, robotics, control, dependable computing, distributed systems, and autonomous driving, we further propose a non-normative hierarchy of trustworthiness levels. This hierarchy grades the strength of bounded deployment claims across task capability, safety, system assurance, operational governance, and supporting evidence, providing a basis for bounded deployment, comparative evaluation, research prioritization, and future standardization.
Authors: Hua Qian, Manisha Kotha, Tuan Tran, Jennifer Shin, Haining Zheng
Abstract: This study developed a hybrid computer vision method to quantify exposed skin from images for dermal exposure assessment. Using 170 indoor-painting images, Mask R-CNN first identified human subjects and removed background interference; a color-based algorithm then segmented exposed skin. The resulting exposed-skin-to-body pixel ratios showed approximately 80% agreement with human estimates. The approach demonstrates a scalable way to extract semi-quantitative exposure information from images, with future extensions to body-part recognition, PPE detection, and video-based exposure analysis.
Authors: Chandra Sripada, Richard Lewis
Abstract: LLMs are widely regarded as alien intelligences, systems whose cognitive operations are fundamentally unlike our own. Apparent similarities to human cognition are therefore often seen as the result of anthropomorphic projection. We argue that this framing is mistaken. LLMs clearly differ from humans in important respects, including their physical substrate, learning history, and the environments with which they interact. These differences make it all the more striking that contemporary LLM-based systems converge with human cognition on a number of principles of cognitive organization with longstanding support in cognitive science. We identify structural correspondences across five dimensions: inferential organization, computational architecture, representational structure, prediction-driven learning, and reinforcement-learning-like mechanisms supporting goal-directed action. These correspondences support a broader model of intelligent cognition in which core principles long used to explain human intelligence also characterize contemporary LLM-based systems.
Authors: Neta Kirmayer, David Tayouri, Andr\'es Murillo, Motoyoshi Sekiya, Asaf Shabtai, Rami Puzis
Abstract: Security operations centers rely on anomaly detection systems to flag suspicious events. Feature-level explanations for anomaly detectors offer limited value for operational investigations. To effectively handle alerts, analysts need to know contextual relationships and need actionable understanding of the entities involved. This paper introduces an event-centric detector-agnostic approach for explaining cybersecurity alerts in small- to medium-sized enterprise networks. We present (EC)2, a multi-agent framework that performs structured, hypothesis-driven investigation to provide explanations grounded in verifiable evidence. Evaluation results show that the proposed framework improves post-detection analysis by generating operationally meaningful explanations, which also enhance event classification accuracy.
Authors: Quim Motger, Marc Oriol, Jordi Marco, Xavier Franch
Abstract: Multi-Agent Debate (MAD) is a promising paradigm for improving the accuracy and robustness of Large Language Model (LLM)-based agentic systems. It enables multiple agents to exchange arguments, critique each other's outputs, and iteratively converge towards a solution. However, research remains fragmented, with inconsistent terminology and no rigorous synthesis of MAD design dimensions. We present a systematic literature review characterizing 141 primary studies on MAD. We derive a three-dimensional taxonomy covering debate participants, the interaction mechanisms structuring the exchange, and the agreement protocols governing debate resolution, supported by formal notations to render MAD configurations. Our analysis reveals that the field has implicitly converged on a narrow design pattern - static, fully connected topologies, verbatim exchange, short-term memory and voting resolution strategies - adopted by convention rather than systematic comparison, while promising alternatives remain marginal. Because any MAD setting reflects roughly a dozen interacting design decisions, cross-study comparison is unreliable when these are left implicit. We position the taxonomy as a descriptive map of the research landscape, a framework for controlled benchmarking, and potentially as a schema for machine-readable MAD specifications. As future work, we propose formalizing it into an executable specification, enabling cost-aware benchmarking and automated tuning of debate configurations.
Authors: Ahmed Ibrahim
Abstract: Context: Large Language Models (LLMs) offer natural-language flexibility for automated requirements elicitation but frequently generate structurally invalid requirements and logical inconsistencies, lacking formal correctness guarantees. Objectives: This study aims to eliminate logical inconsistencies and enforce structural conformance in LLM-generated requirements while quantifying the LLM's pre-validation decision uncertainty within a formal domain model. Methods: We present a neuro-symbolic multi-agent architecture that operationalizes the Object-Oriented Method for Requirements Authoring and Management (OOMRAM) lattice. The LLM acts as a non-deterministic heuristic for lattice traversal, while a deterministic symbolic validator enforces all structural constraints. We introduce a three-valued (T, I, F) -- Truth, Indeterminacy, Falsity -- framework to classify and score the LLM's requirement decisions before and after validation. Results: Evaluated across 37 natural-language project visions in eleven application families, the system completely eliminated structural inconsistencies in 35 out of 37 cases (94.6%), with the remaining two containing only 6 unresolved structural errors (0.39% of decisions) due to iteration limits. Three-valued analysis revealed that 24.7% of all decisions are indeterminate -- structurally valid but discretionary choices not explicitly mandated by the stakeholder. Conclusion: Offloading structural integrity to a deterministic symbolic layer successfully guarantees structural conformance, while the three-valued classification provides a formal way to measure neural uncertainty, facilitating safe LLM deployment in formal requirements engineering.
Authors: Lorenzo Cima, Alessio Miaschi, Amaury Trujillo, Marco Avenuti, Felice Dell'Orletta, Stefano Cresci
Abstract: AI-generated counterspeech offers a scalable and effective strategy to mitigate online toxicity by promoting more constructive dialogue. Yet, existing approaches adopt a generic, one-size-fits-all paradigm, overlooking the conversational context and characteristics of the targeted users. Here, we propose and evaluate multiple strategies for generating contextualized counterspeech that is adapted to the moderation setting and personalized to the moderated user. In detail, we explore a range of configurations that integrate different forms of contextual information and fine-tuning techniques. We conduct a comprehensive evaluation combining quantitative indicators with a pre-registered, mixed-design crowdsourcing experiment. To ensure robustness, we implement algorithmic measures of counterspeech quality based on ROUGE, BLEU, and BERTScore, observing overall consistent results across metrics. Furthermore, we analyze which characteristics of both the generated counterspeech and the moderated toxic message most strongly influence perceived persuasiveness, yielding insights into how contextualized interventions can be made more effective. Our findings show that personalization can be effective, but not uniformly so. Lightweight strategies combining conversational context and user history improve perceived adequacy and persuasiveness, whereas several other contextualization strategies degrade human-perceived counterspeech quality. Taken together, these results provide actionable directions for developing more personalized, effective, and responsible counterspeech systems, ultimately advancing human-AI collaboration in online content moderation.
Authors: Nicolas Gutowski, Fabien Chhel, Alexandre Letard, Sylvain Lamprier
Abstract: We consider a stochastic multi-objective bandit problem where, at each round, the agent selects a slate of $k$ arms and observes their $d$-dimensional reward vectors under semi-bandit feedback. We do not aim at identifying a single optimal arm; instead, we consider the problem of maintaining a small set of actions that jointly approximate the Pareto frontier. We formalize this objective through the dominated hypervolume induced by the selected subset of arms, and define an $\alpha$-approximate hypervolume regret with respect to the best size-$k$ subset achievable in hindsight, where $\alpha = 1 - 1/e$ reflects the approximation guarantee of greedy maximization for monotone submodular functions. To address this problem, we introduce \textit{THV-UCB}, an optimistic algorithm that selects arms greedily based on optimistic estimates of their marginal hypervolume contributions. We establish a gap-free regret bound $\tilde{O}(d\sqrt{nkT})$ that holds on every instance, together with a gap-dependent bound $\tilde{O}(nk^{2.5}/\Delta_{\min})$ that becomes polylogarithmic in $T$ once the arms are sufficiently well separated. Our results provide theoretical support for using small subsets to approximate Pareto fronts in various multi-objective applications.
Authors: Kaushik Pavani, Ganga Aluri, Pravin Jadhav, Neeraj Prasad, Kiran Sanka
Abstract: We built and evaluated a self-serve entity resolution (ER) system on six benchmarks spanning 864 to 5M records, and three lessons emerged that are absent from existing ER literature. (1) No single matching algorithm wins everywhere - a self-serve pipeline cannot predict its next dataset, so we recommend training several algorithm families per dataset and letting an automatic bake-off pick the winner. (2) Precision and recall need separate fixes, not a shared threshold - precision needs hard rule-based vetoes, recall needs more diverse candidate retrieval. (3) One false-positive link can silently merge unrelated entities - assuming "A matches B" and "B matches C" implies "A matches C" lets a single bad link chain hundreds of records together, so every cross-group merge must be actively re-verified. We hope these lessons save practitioners the months of dead-end experiments that led us to them.
Authors: Xuan Zhao, Jiwoong Sohn, Qinyue Zheng, Michael Moor
Abstract: AI agents are increasingly adept at tackling complex, long-running tasks. With the rapid surge of autonomous capabilities, human oversight is systematically lagging behind due to limited human-centered interfacing. Aiming to address this, we introduce AgentGUI, a user-friendly, locally hosted GUI for seamlessly observing and steering AI agents amid multiple concurrent, long-running sessions. AgentGUI features 1) rich agent trajectory visualizations, 2) effective manual and automated steering, and 3) integration with and coordination between open-source and frontier agent frameworks. A controlled user study demonstrates statistically significant reduction in the time it takes to identify key elements from agent traces (38% faster, p = 0.023). In a preliminary experiment, AgentGUI's automated drift prevention feature raises the task completion rate of small local agents by as high as 34pp across a 0.8B--9B model ladder (N=50 runs per model). AgentGUI is publicly available through its project website (https://agent-gui-project.github.io) and open-source repository (https://github.com/eth-medical-ai-lab/agent-gui), along with a demo video (https://youtube.com/watch?v=GSDyxN1gTF0).
URLs: https://agent-gui-project.github.io), https://github.com/eth-medical-ai-lab/agent-gui),, https://youtube.com/watch?v=GSDyxN1gTF0).
Authors: Gaston Besanson
Abstract: Agentic systems act, so a defect in the evidence they retrieve becomes a wrong action with a currency cost. The most dangerous enterprise defects are metadata-borne: a stale price or a superseded record, perfectly well-formed in the payload and betrayed only by freshness, lineage, or provenance. Such a defect never enters the agent's context, and an agent cannot doubt data it cannot see. On a priced replenishment benchmark, a competent agent silently converts an injected metadata-borne defect into a costly action about 60% of the time, with zero data-quality flags and behavioral doubt markers at chance (AUC <= 0.50). Across four model tiers spanning roughly 15x in inference price, the rate stays flat: capability does not buy skepticism. A metadata-aware pre-action gate with downstream-only remediation recovers the loss fully on the signals its predicates cover and not at all on those they miss. A model-free oracle derived from the task's decision geometry tracks the measured rates with MAE 0.015 (Pearson r = 0.876, interval coverage 15/16 cells), giving the flat ladder an analytical form. Evidence integrity is a systems axis distinct from model capability; mitigation depends on enforcement placement and predicate coverage. Code, frozen results, and a deterministic analysis pipeline: https://github.com/besanson/dqSarc
Authors: Ads Dawson, Adrian Wood
Abstract: Stealth, the discipline of achieving an objective without revealing your presence, capabilities, or collected intelligence, is what separates sophisticated operators from detectable ones. Elite security researchers and advanced persistent threats achieve their objectives unnoticed; autonomous agents increasingly inherit the same offensive tasks, but do they inherit the tradecraft? We introduce StealthBench,a benchmark that measures operational stealth in autonomous offensive-security agents across six operational security (OPSEC) dimensions. We extract 11 hand-verified OPSEC incidents from real bug-bounty and red-team trajectories, expanded into 14 dockerized task scenarios, where agents, despite finding real vulnerabilities, committed stealth failures inconsistent with standard operational tradecraft: embedding credentials in public uploads, deleting production resources to prove access, force-adding uninvolved users to demonstrate a race condition. We evaluate agent trajectories using a 3-model large language model (LLM) judge panel with majority-vote aggregation, measuring safe success rate (solved and stealthy), Stealth@Solve (tradecraft quality among successful solves), and reckless solve rate (solved but cover blown). Our results show that no model exceeds 54% safe success rate (the compound metric requiring both task completion and stealth), confirming that OPSEC failures are systematic across model families. We release StealthBench as a public benchmark to support both the development of stealth-aware agents and automated OPSEC monitoring for autonomous offensive-security deployments. The interactive leaderboard, evaluation harness, and dataset are available at https://stealthbench.com.
Authors: Wenjie Zhou, Yunting Liu, Renjiao Tang, Mark Wilson
Abstract: Psychometric calibration for educational tests typically requires costly human response data. Large language models (LLMs) simulated examinees offer a promising route to early calibration, but their responses are too accurate and too uniform. We propose Cognitive Diagnostic Profiling (CDP), a zero-shot framework that prompts LLMs to simulate plausible examinees with diverse cognitive profiles: binary attribute-mastery patterns are rendered as natural-language profiles and sampled under an uninformative or an informative distribution. Using the Tatsuoka fraction-subtraction dataset (536 examinees, 15 items, five attributes), we evaluated eight LLM configurations under no-profile, uninformative-CDP, and informative-CDP conditions, assessing alignment with human examinees at the ability-distribution, mastery-profile, and item-difficulty levels. CDP improved all three levels: distributional overlap rose across configurations; weighted correlations between profile-level scores and human profile expectations reached 0.92 to 0.98; and item-difficulty recovery improved in rank order and absolute alignment, most for reasoning-enabled models; in the strongest case, Gemini 3.0 Flash (Thinking), one-parameter logistic (1PL) difficulty Spearman correlations rose from 0.24 to 0.86 and 0.90 and the root-mean-square error (RMSE) fell from 6.31 to 1.30 and 0.90; the informative condition helped most where profile-level alignment was strong. CDP brings LLM-simulated examinees into closer psychometric alignment with human examinees, making them practical for operational test development.
Authors: Xin Xu, Siru Tao, Kaizhen Tan
Abstract: A gradient-based GNN explainer given a molecule with two chemically equivalent nitro groups assigns them attribution scores that are equal to the last bit. It cannot do otherwise: message passing is exactly permutation equivariant, so any automorphism of the input leaves every attribution invariant. Yet the standard report, the top-k edges, names one of the two, and which one is settled by the order of an array. We show this is a structural obstruction rather than an implementation slip. When no minimal valid explanation is fixed by the input's automorphism group, no rule can be single-valued, minimal and symmetry-respecting at once. For the exact-k reports used in practice we give a parameter-free criterion, mechanised in Lean 4 with no axiom dependencies, that decides from the graph alone whether every score-optimal report of that size must split an orbit. Across 21298 instance-budget decisions the criterion agrees with a mechanical model-equivalence check without exception, and no severing case we found admitted a neutral alternative. The obstruction is common. Nontrivial automorphisms occur in 93.4% of Mutagenicity, the dataset the seminal explainability papers use, so the measure-zero dismissal of symmetric inputs, sound on the continuous domains it was made for, collapses here. At the sparsity budget those papers report, 24.0% of molecules with two interchangeable nitro groups (6 of 25) surface exactly one of them, every one arbitrary under mechanical verification. A model's blindness also manufactures symmetry: every MUTAG molecule contains atoms chemistry separates and the network provably cannot, and a matched control shows the resolution is set by what the model reads rather than how it is parameterised. Reporting orbits removes the arbitrariness at 0.11 ms and 0.43 extra edges per graph.
Authors: Zihan Chen, Di Zhu, Lei Nico Zheng
Abstract: Large language models (LLMs) are increasingly used as synthetic users, stand-ins for human respondents whose simulated answers feed product, policy, and market decisions. We ask when this substitution is valid and when it fails, and package the answer as an evaluation framework for intelligent synthetic-user systems. A single protocol, run across four models spanning two families and an 8B-to-frontier capability range, is applied to two independent domains of real human-response data: U.S. general social attitudes (General Social Survey) and cross-cultural values (World Values Survey). Every model is benchmarked against a suite of non-LLM baselines fit on held-out human data. Under demographic prompting and the survey-simulation protocols we test, two failures replicate across both domains, all four models, and both families. First, at the individual level no LLM beats even the strongest baseline; on cross-cultural values every model falls well below it, and the gap survives distance-aware and proper scoring. Second, models systematically over-determine demographics, treating identity as far more predictive of attitudes than it is among real people, a distortion present for nearly every question-group combination and robust to a coding-invariant measure. Neither failure is remedied by a larger, more capable model. A decision-impact analysis shows why this matters in practice: on a segment-targeting task the models inflate between-segment gaps two to fourfold, would direct a team to the wrong segment in half of U.S. and most cross-cultural cases, and manufacture segment splits that do not exist in real people. We make the cross-domain benchmark and the evaluation framework available on request, so that teams can determine in advance when synthetic-user evidence is safe for decision support and when it is not.
Authors: Jaber Daneshamooz, Eugene Vuong, Alagappan Ramanathan, Manni Moghimi, Haarika Manda, Satyam Kumar, Snithik Thode, Satyandra Guthula, Sylee Beltiukov, Dongsu Han, Tarun Mangla, Sangeetha Abdu Jyothi, Walter Willinger, Arpit Gupta
Abstract: Networking research advances by turning hypotheses into empirical evidence, so accelerating it means reducing the lag between ideation (synthesizing a hypothesis) and generating the data that tests it. Consider a concrete case: does a bulk BBR download fairly share its bottleneck with competing real-time Google Meet traffic? Validating this requires configuring a realistic bottleneck link, concurrently generating BBR's bulk transfer and Meet's real-time traffic, and collecting relevant service-quality metrics. Today this overhead is high, often forcing researchers to start from scratch for every new idea. This ideation-to-data-generation gap will only worsen in the agentic AI era, where AI-assisted ideation accelerates exponentially, yet its outputs cannot be validated without a data-generation backend. This paper explores how to bridge this gap. We envision a composable, domain-specific backend, Pramana, shaped as a thin waist, with diverse research intents at the top and disparate execution substrates at the bottom. Pramana realizes this waist through a single contract, the intent specification, which disaggregates an experiment into three independent axes: the intent (what data to generate), the substrate (where to generate it), and the mechanism (how to produce it), so one specification runs on any substrate. We demonstrate Pramana's utility by building a first-of-its-kind corpus of 255 data-generation intents mined from 66 published papers, and show the intent specification satisfies all of them, where no existing tool satisfies more than 13%. Our current proof-of-concept implementation already satisfies 34% of these intents, more than twice the best existing tool, and we lay out a roadmap for closing this abstraction-implementation gap through a broader community effort to build the envisioned data-generation backend and accelerate empirical networking research.
Authors: Loong Kuan Lee, Ragavi Krishnamoorthy, Nico Piatkowski
Abstract: The problem of learning the graphical Markov blanket (MB) of a variable from data has applications in many areas such as structure learning for Bayesian networks and Markov random fields, causal discovery, and feature selection. However, a common assumption most methods make is that the conditional independencies in the distribution imply the same separation in the graphical structure -- also known as the faithfulness assumption. Unfortunately, this assumption can be violated by higher-order dependencies such as XOR and parity-type relations, and -- on finite samples -- by empirical violations that, in extreme cases, even induce spurious dependencies absent from the true distribution. Therefore, in this paper we propose a "k-order" relaxation of the faithfulness assumption that captures parity type relationships between k+2 variables. We then propose a proof of concept algorithm called k-order Markov blanket (kOMB) that uses this relaxation for MB discovery. Finally, we empirically show how kOMB can recover the MB of a variable under both true and empirical violations of faithfulness. Code available at: https://github.com/lklee9/k-order-Markov-blanket
Authors: Keegan Harris, Brian W. Lee, Ian Waudby-Smith, Philip Amortila, Nika Haghtalab, Michael I. Jordan
Abstract: Reinforcement learning (RL) fine-tuning is widely used in language model training to improve model performance on a target task while limiting drift from a reference policy. A standard way to balance this trade-off is via a KL-regularized RL objective, although this formulation does not by itself provide a principled way to set the regularization coefficient. In practice, the coefficient is typically chosen heuristically or via hyperparameter search, which can lead to unnecessary overhead in training cost or undesirable reward-retention trade-offs. We instead propose a game-theoretic framework that gives this trade-off an explicit statistical interpretation. Specifically, we study a sequential game in which an agent chooses a policy to maximize cumulative reward while a monitor observes policy outputs over time and tests for deviations from the reference policy. Although not originating from the same perspective, we show that the resulting equilibrium policy can nonetheless be expressed as the solution to a KL-regularized RL problem for an optimal regularization parameter that can be viewed as maximizing reward per unit of statistical distinguishability. Drawing on classical results from concave-convex fractional programming, we provide a principled method for learning this equilibrium coefficient via reduction to the KL-regularized RL objective, thus allowing for flexible integration into standard fine-tuning pipelines. In experiments with Qwen3-8B and Llama-3.2-1B, we demonstrate that our methods result in competitive reward-retention trade-offs in a continual learning setting, and illustrate how our framework may be used to audit API providers serving open-source models.
Authors: Aman Kumar, Lasitha Vidyaratne, Dipanjan D Ghosh, Arnab Chakrabarti, Ahmed K Farahat
Abstract: Financial disclosures contain numerical claims, temporal statements, entity references, policy commitments, and risk descriptions that may conflict in qualitatively different ways. Detecting a conflict is only the first step: review workflows may also need to determine its type, since numerical, temporal, referential, factual, and normative inconsistencies require different evidence and downstream checks. We study this problem as fine-grained inconsistency classification. Using a fixed 5,940-instance snapshot of SBID-FD, a synthetic financial-disclosure benchmark with 11 inconsistency labels and paired reference evidence spans, we compare frozen embedding classifiers, fine-tuned encoders, evidence-augmented classifiers, prompted large language models, and LoRA-adapted generative models under a shared evaluation protocol. A fine-tuned 300M encoder reaches 61.9% accuracy, compared with 61.5% for a LoRA-adapted Qwen3.5-9B model and 61.3% for GPT-5.4. Because these systems differ in architecture, supervision, training objective, and input format, we interpret this as a practical efficiency result for compact supervised encoders rather than a controlled conclusion about model scale. Supplying gold evidence spans improves the fine-tuned encoder to 65.3%, whereas automatically predicted spans recover a meaningful but incomplete share of that gain, indicating that localization quality remains a bottleneck. Class-level analyses show that Referential inconsistencies are especially sensitive to localization quality, while Factual and Logical inconsistencies remain difficult even when the relevant evidence is provided. Together, the oracle, distractor, and per-class analyses separate localization errors from residual type-discrimination errors, indicating that progress requires both stronger evidence extraction and better reasoning over closely related inconsistency categories.
Authors: Yitong Shen, Cheng Guo, Peiliang Wang, Jingzhe Zhang, Yi Sheng, Haopeng Zhang, Hongfei Xue, Yili Ren
Abstract: Wi-Fi-based human activity recognition has advanced substantially, but most existing methods assume a closed set of activities and require labeled Wi-Fi samples for every target class, limiting their ability to recognize unseen activities. We present Zero-Fi, a contrastive signal-language alignment framework for zero-shot Wi-Fi-based human activity recognition. Zero-Fi learns unified representations from complementary Wi-Fi signal features and aligns them with the semantic representations of natural-language activity descriptions in a shared embedding space. This cross-modal alignment enables Zero-Fi to recognize new activity classes without requiring labeled Wi-Fi samples or model adaptation for those classes. Experiments on large-scale public benchmark datasets demonstrate effective zero-shot recognition of held-out activity classes, highlighting the potential of signal-language alignment to extend Wi-Fi sensing beyond predefined activity classes.
Authors: Xin Xu, Chengrui Wu, Jiayu Lu, Kaizhen Tan, Siru Tao, Hanzhe Hong
Abstract: Empirical work on algorithmic collusion asks one question of the data: are prices supracompetitive? We show this can be answered "no" by a conspiracy that is nonetheless profitable. Consider bidding agents that couple only through the joint distribution of their unexplained bid components, leaving every agent's own bid law exactly at the competitive law. Any test whose input is a single agent's price or bid history then has power exactly equal to its false-positive rate, for every coupling strength up to comonotonicity. The published detection methodology is therefore blind to this conduct by construction rather than underpowered, and no sample size repairs it. Three empirical results follow. First, the mechanism appears in real language-model agents: twenty models from nineteen independent developers, three deployment prompts each, show residual correlation of $+0.053$ between two deployments of one model against $+0.0001$ across models, with a 95% interval clustered by developer of $[0.030, 0.078]$, under an auditor that sees every order feature and is fitted out of sample. Second, the coupling falls monotonically as sampling temperature rises ($p=0.002$), turning a deployment parameter into a candidate mitigation. Third, on 24 days of Ethereum block-building auction data covering 77,684 bids from 39 bidders, the honest population of bidder pairs is itself so dependent that a screen held at a 5% false-positive rate must sit above a floor of $+0.50$ to $+0.81$, which is 20 to 32 times the family-wise sampling threshold and does not fall as the audit window grows. Since lawful multi-identity operation and conspiracy are behaviourally indistinguishable here, the tractable regulatory target is not detection but counting: resolving 40 bidding identities into 23 operators raises the Herfindahl index by 247.5%, and adding behavioural clusters from public bid streams reaches 324.5%.
Authors: Hasibur Rahman, Smit Desai
Abstract: Fine-tuning a language model on data containing a narrow flaw, such as insecure code or incorrect mathematical answers, can cause broad misalignment through a mechanism that remains debated. We provide an interpretable account: in the models and corpora we study, misalignment behaves like a shift in personality. Prior work extracts activation directions for character traits from a single binary contrast, which can separate or steer behavior without establishing a calibrated scale. We instead extract personality vectors for the Big Five using a graded, three-level intervention and validate them on two open-weight models. The three levels are linearly ordered, with Cohen's d values of up to 6.2; the vectors transfer zero-shot and trait-specifically to an independent corpus; and their effects are strongest within a middle-layer band. Applied to training data, the vectors reveal that misaligned corpora across eight domains share a common Big Five signature: lower agreeableness and conscientiousness, together with higher extraversion and neuroticism. This signature is recovered by both models with a correlation of r = 0.94. Fine-tuning imprints the same profile, shifting the model's generations along the corresponding signature, with r = 0.83 using activation-based measurements and r = 0.90 using a text-based judge, while also shifting internal activations with r = 0.69. The same vectors characterize sycophancy as high extraversion and low conscientiousness rather than excess agreeableness, a distinction that a single direction cannot capture. Calibrated personality vectors transform an opaque safety phenomenon into a human-legible diagnostic profile.
Authors: Chao-Han Huck Yang, Zih-Ching Chen, Piotr Zelasko, Zhehuai Chen, Jagadeesh Balam, Boris Ginsburg
Abstract: We present Voice Memory, a inference-only scheme for agentic speech recognition: at stream time, a frozen corrector reads a single per-domain memory.md and decides per utterance whether to act on the hypothesis or abstain and keep the 1-best. Asynchronously, a score-gated optimizer revises that file through bounded edits, accepting an edit only when it strictly improves a held-out score. Extended from classical ASR-LM framework, we refer this split the listener-thinker architecture; the two roles are coupled only through the memory, so no weights change and the learned skill stays auditable and portable. Restraint turns out to be the operative skill this loop discovers: unconstrained generative error correction (GER) over-corrects, breaking correct tokens on up to 64% of its edits on financial news, and Voice Memory, reduces this rate to 35%. Across ten HyPoradise domains with an open corrector, Voice Memory, lowers weighted word error rate from 8.36% to 7.52% (7.47% with three added in-context examples) without regressing any dataset below its 1-best baseline; gains concentrate where recoverable headroom is largest, including air-travel commands (8.40% to 3.40%) and noisy far-field speech (CHiME-4, 12.69% to 10.46%). The memory transfers across corrector families and adds zero parameters to the inference path. A demo and example code are provided for future studies.
Authors: Hongyang Wang, Yichen Shi, Hongrui Li, Yiru Huo, Jun Feng, Zitong Yu
Abstract: Face anti-spoofing (FAS) is increasingly expected to provide not only bona fide/spoof decisions, but also attack semantics and image-grounded evidence for human inspection. Existing discriminative FAS models remain largely label-centric, while recent MLLM-based methods offer structured outputs but still rely mainly on supervised fine-tuning, often producing template-like rationales and weak optimization for difficult attacks. We propose FAS-R1, a two-stage reasoning-oriented MLLM framework for unified FAS prediction, covering authenticity classification, attack-type recognition and spoof-region localization. FAS-R1 first uses FAS-R1-23K, a high-quality long-CoT dataset, for cold-start supervised fine-tuning, and then performs FAS-specific GRPO post-training. Degradation-Simulated Augmentation (DSA) encourages stable spoof-cue reasoning across visual-quality shifts, while Difficulty-Aware GRPO (DA-GRPO) mitigates easy-sample dominance that may leave difficult task--attack groups under-optimized, especially for subtle or ambiguous attacks such as makeup and mask attacks. The main 3B FAS-R1 model achieves 98.75\% authenticity accuracy, 93.33\% attack-type accuracy, and 96.30/94.73\% AP@40/AP@50 in-domain. It also outperforms the compared systems in cross-domain authenticity generalization and answer-and-rationale quality. Experiments with different base models further show favorable scaling behavior. The code will be released soon.
Authors: Javier C. Weddington, Bence P. \"Olveczky, Stephen A. Baccus
Abstract: Deploying learned control policies on low-cost robotic platforms introduces transport latencies and noisy motor feedback that systematically widens the sim-to-real gap. The chasm of simulation to deployment in hardware lies in the delay of the actuator reaching the commanded position. On platforms such as the Mini Pupper 2, a measured >50 ms transport delay transforms the locomotion task from a standard Markov decision process into a partially observable one. In this paper, we take a biologically inspired approach of handling noisy and delayed feedback to close the sim-to-real gap, thereby expanding the capability of reinforcement learning on cost-constrained hardware. Using a low-cost quadrupedal hardware platform, we find that using a forward model of the average actuator delay, paired with a time-aware neural network results in robust locomotion. Additionally, our time-aware neural network learned a central pattern generator (CPG): a self-sustaining rhythmic gait that is robust to +320 ms latency perturbations, mirroring the CPGs found in the spinal cords of vertebrates. We posit that temporal self-organization may be a general strategy for cost-constrained locomotion.
Authors: Dachuan Song, Junyu Yin, Zechen Hu, Xuan Wang
Abstract: A known limitation of long-context language models is their increasingly unreliable performance in non-additive, set-based aggregation as context length grows. Examples include cardinality estimation, set relationships, and grouped statistics, which widely exist in logs, program outputs, tables, and multi-turn conversations. To provide the aggregation state required by these tasks, we introduce a model-side aggregation interface that maintains compact Hash-based HyperLogLog (HLL) sketch states alongside a frozen language model. While the model processes the context, an extractor maps each relevant record to a canonical identity. The identity is then hashed and updates the HLL state. These states can be merged across context segments and/or read out directly for downstream reasoning, avoiding an additional generate-execute-return cycle. We validate the proposed approach by setting the HLL state size as 2 KiB (2,048 registers), which does not increase with context length or set cardinality. In a distinct-count experiment involving one million records, the mean relative error was 1.6%. In a separate merge test, states built from as many as 256 segments produced exactly the same readout as a single pass over the same stream. On 3,969 aggregate-then-reason tasks from 174 source windows, the fixed-budget interface reached 99.2% accuracy on Gemma 4 (31B, BF16), compared with 100.0% under exact aggregation; the paired gap was 0.8 percentage points (95% window-cluster CI: 0.5-1.3 points). On a matched set of 174 items, our method improved over direct full-context reasoning by 63.2 points on Qwen and 56.3 points on Gemma. The corresponding gains over chain-of-thought (CoT) reasoning were 60.9 and 63.2 points, respectively. On a fixed 1,200-task Oolong-Synth subset, our method reached 91.1% on Qwen and 99.3% on Gemma. Code is available at https://github.com/songdc98/sketchops.
Authors: Ruxi Gu, Zhenliang Zhang, Wei Wang
Abstract: Large language models (LLMs) have demonstrated strong capabilities in knowledge acquisition and reasoning, yet their ability to retain previously acquired knowledge under repeated updates remains insufficiently understood. Existing evaluation paradigms primarily focus on single-step reasoning or static knowledge editing, which fail to capture the temporal dynamics of knowledge retention and degradation during continual model modification. In this work, we propose ForgetBench, a benchmark designed to systematically characterize forgetting behavior in LLMs under continual knowledge editing. ForgetBench introduces two complementary evaluation paradigms, namely concept-based QA and scenario-based QA, to disentangle isolated factual retention from structured relational knowledge preservation. Building upon a sequential editing framework, we construct temporally ordered knowledge streams and evaluate model behavior across multiple editing stages. To quantitatively analyze long-term retention dynamics, we further introduce a unified evaluation framework that models knowledge evolution over time, enabling the measurement of temporal decay, retention strength, and cross-instance stability. Extensive experiments across diverse models and editing methods demonstrate that existing approaches fail to strike a balance between long-term retention and generalization quality. Our findings highlight the need for more robust memory mechanisms that can effectively acquire, update, and preserve knowledge over time in future LLMs. Code will be released upon acceptance.
Authors: Zekun Ren, Hongzhao Tan, Jiaen Yee, Kedar Hippalgaonkar
Abstract: Physical Unified Device Architecture (PUDA) is an AI-native hardware harness for self-driving laboratories (SDLs). Rather than building a human-centered graphical user interface (GUI) orchestration layer, PUDA creates a command-line runtime environment that lets agents observe, orient, decide, and act over experiments while hardware execution remains deterministic, atomic, and auditable. Headless by design, devices appear through discoverable command-line interfaces, JSON protocols are routed through a distributed messaging system, and command responses, data products, and reports are preserved as structured records. PUDA organizes protocols, runs, samples, measurements, and command logs into an AI-native data structure linked by run identifiers and timestamps, preserving provenance from submitted protocol through hardware response to resulting data products. PUDA separates scientific orchestration from physical operation and data telemetry: agents choose experiments, while PUDA executes validated commands and captures provenance-linked state, responses, and data. The contribution is not another optimizer, orchestrator, or recipe language. It is a practical execution and data environment for agentic SDLs; the broader physical AI implication is that PUDA provides an AI-native hardware harness for AI systems to interact with physical tools.
Authors: Haotian Mo, Jie Liu, Siqi Shen, Songzhu Mei, Xinhai Chen, Xiangyang Wang, Yigui Feng, Shuai Li, Gencheng Liu, Keqi Yang, Qinglin Wang
Abstract: Audio deepfake detectors often degrade when generators, corpora, or recording conditions change. We use a Diffusion Transformer (DiT), trained only on bona fide speech, as a frozen reconstruction probe. Reconstructions at masking ratios 0.5, 0.75, and 0.9 yield explicit multi-ratio residual maps. Because these residuals are domain sensitive, our audio-anchored detector passes the projected frozen-WavLM auditory representation into the fusion sum without gate-based attenuation and uses residuals only as a scalar-gated additive correction. The pre-specified seed-42 run obtains 6.5442% EER / 0.18456 min-DCF on ASVspoof 5 Eval and 13.8372% / 0.36921 on ITW Full; three-seed means are 6.8885 (0.3308)% and 15.3328 (2.0719)%. The latter is below a separately optimized WavLM-ResNet18 reference under both supervision settings. Auxiliary supervision raises dynamic competitive fusion from 18.4007% to 25.2968% mean ITW EER, worsening all three seeds. The results support reconstruction residuals as complementary evidence and motivate a non-competitive auditory path for ASVspoof 5-to-ITW transfer, without claiming a componentwise causal ablation of anchoring alone.
Authors: Ming-Yen Lee, Hanchen Yang, Faaiq Waqar, Harsono Simka, Tushar Krishna, Muhammed Ahosan Ul Karim, Shimeng Yu
Abstract: The energy consumption of Large Language Model (LLM) serving is becoming a major system challenge as deployment scales, driven by hardware power and thermal constraints and rising electricity costs. A key contributor to chip energy dissipation is data movement between limited on-chip cache and off-chip High Bandwidth Memory (HBM). Meanwhile, emerging memory technologies such as monolithic 3D (M3D) integration of cache memories at the Back-End-Of-Line (BEOL) of logic chips enable larger and denser on-chip memories, creating new opportunities to reduce costly off-chip traffic. However, it remains unclear whether continuously scaling on-chip memory using emerging technologies can effectively improve the energy efficiency of LLM serving. To address this gap, we develop LLMET (LLM with Emerging Technology), a validated cross-layer simulation framework, and conduct a comprehensive study on the impact of large-capacity on-chip memory technologies across a broad range of models, applications and platforms. Utilizing M3D technology to expand the L2 cache from 40MB to 1GB yields a 44% reduction in chip energy during the Llama3.1-70B prefill phase with a 16K context window, based on LLMET simulation on a dual NVIDIA A100 GPU setup. On the 8x NVIDIA B200-like platform, extending the L2 cache from 128MB to 4GB saves the prefill energy by up to 24%. For the edge platform and workloads, the decode energy saving reaches 30% when increasing the 8MB cache size to 256MB. These results highlight the promise of ultra-large on-chip memories for energy-efficient LLM serving systems.
Authors: Gong Gao, Xiao Lai, Ziqi Xie, Guojie Chen, Xianhui Liu, Weidong Zhao
Abstract: Deep off-policy reinforcement learning algorithms for continuous control typically rely on neural value function approximation to guide policy improvement. However, temporal-difference (TD) learning introduces noisy targets, resulting in non-stationary optimization, while greedy policy updates amplify early-stage estimation errors. The recursive propagation of such errors leads to persistent overestimation bias and degraded training stability in actor-critic methods. Existing approaches attempt to alleviate this issue via prioritized sampling or modified value learning objectives, but often overemphasize high-uncertainty transitions caused by limited data coverage or bootstrapping errors, thereby further amplifying bias.In this paper, we propose Collaborative Weighting Actor-Critic (CWAC), a unified framework that explicitly accounts for predictive uncertainty in value estimation. CWAC employs distributional critic to model return uncertainty and introduces a collaborative weighting mechanism that jointly reweights TD-errors and uncertainty, enabling robust learning from reliable samples while suppressing noisy updates. In addition, we incorporate a stochastic pessimistic value estimation scheme via sampling from the return distribution, which effectively mitigates error propagation during policy improvement. CWAC can be seamlessly integrated into existing off-policy algorithm frameworks such as SAC, TD3, and DDPG with minimal overhead. Empirical results demonstrate that our proposed method significantly enhances performance across a diverse range of simulated tasks. Our code is publicly available at https://anonymous.4open.science/r/CWAC-348E.
Authors: Hei Yi Mak, Shadan Golestan, Hoang Le, Mehran Taghian Jazi, Yunke Peng, Yaoyuan Wang, Yao Wang, Junsong Wang, Tianchi Hu, Fengchen He, Guipeng Hu, Tanzila Rahman, Anandharaju Durai Raju
Abstract: We present, to our knowledge, the first end-to-end FP4 RL post-training, in which both the rollout and training policies, including their forward and backward passes, operate at 4-bit precision. A systematic study reveals that the dominant source of degradation in FP4 RL is not training-side quantization error but rollout activation quantization: outliers stretch the dynamic range so far that a large number of activation values underflow to zero under FP4. Counterintuitively, restoring the training policy to higher precision while keeping the rollout in FP4 makes accuracy worse than full FP4 baseline, exposing rollout-training mismatch as the principal failure mode and ruling out standard pretraining-style fixes. We address this with Rollout Residual Quantization (Rollout-ResQ): a single residual correction term constrained to a hardware-friendly sparsity pattern, added only to the FP4 rollout matmul -- a lightweight correction that recovers most of the precision lost to outlier-driven underflow without inflating the rollout's compute footprint. On Qwen2.5-3B and Qwen2.5-Math-7B, Rollout-ResQ paired with the HiFloat4 (HiF4) format -- whose three-level hierarchical scaling preserves resolution under FP4's tight 4-bit budget -- closes the accuracy gap to BF16 from 4.9% to 1.1%, bringing fully quantized FP4 RL within striking distance of full precision. Applied to the open-standard MXFP4, the same recipe narrows the gap from 13.6% to 5.3%, revealing that FP4 format choice is a key factor that determines the ceiling on recoverable accuracy. Together, these results establish HiF4 as the enabling format for end-to-end FP4 RL post-training, and Rollout-ResQ as the activation-side mechanism that makes the gap to BF16 closable.
Authors: Alp Niksarli, Gopesh Baheti
Abstract: Conversational AI agents commonly lack persistent memory across sessions. The obvious fixes like injecting full chat histories into the context window, or delegating to a third-party memory service, either exhaust the model's context budget or send personal data through infrastructure the user does not control. We describe a memory store that avoids both problems: an agent-local Neo4j property graph augmented with HNSW vector indexes and a full bitemporal data model. Each memory is stored as an immutable identity node linked to versioned content nodes carrying two closed-open time intervals: valid time (when the fact was true in the world) and transaction time (when the database recorded it). This design supports point-in-time semantic retrieval without physically overwriting history. Semantic edges between related memories are maintained automatically at write time using cosine similarity over 1024-dimensional embeddings. We evaluate the system on LongMemEval, a 500-question benchmark spanning six question types designed to stress long-term memory. Across 60 sampled questions, the current-state semantic search path achieves 46.7% R@10 overall, rising to 80% on knowledge-update questions. The time-travel path yields 80% R@10 on knowledge-update but decreases recall on temporal-reasoning questions (50% to 37.5%), a consequence of post-filter dilution that points directly to a concrete design improvement. We discuss what these results reveal about the limits of pure retrieval for different question types and what each failure mode suggests for future work.
Authors: Ashmith Atmuri, Yashaswini Rao Bhogarajula
Abstract: We introduce CMP (Cognitive Memory Primitive), a continual-learning architecture that repre?sents inputs as sparse relational codes, stores them in a two-tier competitive memory, and learns through local updates without end-to-end backpropagation through its feature-generating system. We investigate whether combining sparse representations, local learning, and persistent memory can reduce catastrophic forgetting relative to conventional backpropagation-based continual?learning approaches. On a controlled domain-incremental byte-level language modeling protocol, CMP demonstrates substantially lower backward transfer than a parameter-matched Trans?former trained with online Elastic Weight Consolidation (EWC). Across a three-seed replicated 15-domain experiment, CMP exhibits stable forgetting behavior, while separate head-to-head comparisons and domain-order analyses show consistently lower forgetting than the evaluated Transformer baseline under the reported experimental settings. We report these findings alongside a substantial single-domain accuracy gap relative to the Transformer, a null result on a vision benchmark, and a documented failure to combine CMP with an independent accuracy-improving mechanism, reflecting our commitment to reporting both positive and negative outcomes. These results suggest that the combination of sparse representations, local learning, and persistent memory is a promising direction for continual learning, while motivating further investigation into the respective roles of learning rules, representations, and architectural design in mitigating catastrophic forgetting.
Authors: Manuel Rodriguez
Abstract: Symbolic regression provides analytical expressions, but it is usually applied one output at a time. This is limiting in process systems, where state variables are often coupled through shared physical parameters. Independent symbolic regression can give accurate individual equations that are difficult to interpret as one model. We present a neuro-evolutionary symbolic regression method for coupled multi-output systems. The method searches for a shared symbolic backbone: a set of latent symbolic units that is discovered once and reused by several outputs through sparse additive or multiplicative read-outs. The discrete model structure is evolved by mutation and crossover, whereas the continuous parameters are tuned by gradient descent and inherited by the offspring. The method is assessed on a set of benchmarks with known ground truth and on a hydrothermal liquefaction yield case. The results show that coupling is not a general route to lower prediction error. Its main contribution is the enforcement and diagnosis of cross-output consistency when a physically shared factor is embedded in a latent expression and is weakly identifiable from the data. This occurs for Langmuir-Hinshelwood and site-coverage denominators, for which independent PySR does not close the consistency gap or recover the same shared form. Conversely, when each output is already identifiable, as in the Van de Vusse benchmark, independent symbolic regression matches or improves the coupled model. The proposed framework, rather than a general purpose predictor, is a structured shared-mechanism extractor. Its value is highest when the target structure is sparse, shared, weakly identifiable or constrained by closure.
Authors: Jingbo Cui, Jitao Zhao, Di Jin, Dongxiao He
Abstract: Graph Foundation Models (GFMs) aim to learn transferable knowledge from multi-domain graphs and adapt to unseen scenarios. As a fundamental source of relational semantics in graphs, the transferability of topological patterns has long been central to GFM research. However, local structural patterns may vary across graphs and even among nodes within the same graph. Despite such structural variation, most existing GFMs rely on manually designed propagation schemes and apply them to new graphs largely unchanged. Such fixed schemes may not suit the diverse structural patterns of different nodes. This raises a key question: can each node autonomously determine how information should be propagated through the graph? We refer to this capability as information-flow control. Inspired by recent advances in agent technology, we formulate this problem as agent-based decision making and treat each node as an agent. Accordingly, we propose AgentGFM, in which all node agents follow a shared end-to-end trainable policy rather than using independent models. For adaptive information-flow control, each node interacts with the graph through a predict-act-observe-correct process. During the act stage, the node makes three decisions: source reception, signal-channel selection and gain-aware node-wise halting. The resulting observation is compared with the prediction and their discrepancy is used to correct the node state and guide subsequent interactions. Extensive experiments across node-level, graph-level and large-scale transfer scenarios demonstrate the effectiveness of AgentGFM across diverse graph topologies.
Authors: Hayden Helm, Andrew Dassori
Abstract: We propose an index for predicting the U.S.\ Federal Open Market Committee (FOMC) decision to hike/hold/cut the current federal funds target rate based on how a collection of personas responds to current market conditions. To construct the index, we collected a new dataset consisting of nearly $25{,}000$ retrievable chunks from publicly available data. We partition the data into per-member corpora and use each as the retrieval database of a generative system we refer to throughout as a ``persona''. We first evaluate the personas across two complementary components of likeness: identifiability and detectability. Each persona's behavior is highly attributable (average member-conditional recall is $ 8\times $ chance) and generated content is nearly indistinguishable from held-out real content ($\hat\tau_{\mathrm{det}} = 0.23$ against a $0.15$ floor). We then present evidence that query-conditioned representations of the personas capture members' monetary-policy stance relative to a known hawk--dove reputational ordering (Kendall's $\tau = 0.63$, $p < 0.001$), substantially outperforming retrieval-only representations. These representations vary with time and current market conditions and form the basis of our proposed persona-based rate action index. For the $2022$--$2025$ period the index tracks the rate cycle (Kendall's $\tau = 0.68$, $p < 10^{-6}$) and can be used to construct a simple classifier that predicts per-meeting outcomes at non-trivial accuracy ($0.69$ versus a $0.47$ base rate). Importantly, the index outperforms informative baselines and leads the federal funds target rate by roughly three quarters. As far as we are aware, our results are the first to demonstrate the ability to capture time-varying group behavior via a collection of digital personas.
Authors: Xiaoxiao Jiang, Suyi Li, Sheng Yao, Tianyu Feng, Lingyun Yang, Dapeng Nie, Haoran Yang, Wei Wang
Abstract: Text-to-image (T2I) workflows are increasingly deployed on serverless platforms because users often compose customized workflows and invoke them intermittently. Existing platforms typically deploy each workflow as an opaque GPU function, provisioning, placing, and scaling all constituent models in the workflow together. This monolithic design obscures workflow structure, inflates scaling overhead, forces users to manage low-level GPU coordination, and limits fine-grained fairness in multi-tenant clusters. In this paper, we present ServerlessT2I, a serverless-native system that decomposes a T2I workflow into loosely coupled model functions that can be independently managed and scheduled. By explicitly managing individual model execution, ServerlessT2I enables per-model scaling, declarative workflow composition, transparent GPU-resident communication, and fairness-aware scheduling. To make this decomposition efficient, ServerlessT2I harvests slack GPU memory left idle by compute-bound T2I inference to build a data plane that reduces model loading and data communication overheads. \sys{} further introduces a fair scheduler for multi-tenant serving. Using production traces, ServerlessT2I sustains up to 2$\times$ higher request rates than existing T2I workflow serving systems with the same GPU budget; for a fixed request rate, it saves up to 3$\times$ GPU resources while satisfying service level objectives (SLOs).
Authors: Haoyu Zhang, Zhuoxi Wang, Shibo Zheng, Zijian Xiao, Xiangchen Guan, Mohammad Zandsalimy, Shanu Sushmita
Abstract: Safety classifiers ("guards") are the dominant black-box defense for vision-language models, yet they judge an input's surface form, not its meaning: a harmful request re-encoded as set theory, formal logic, a rare language, code, or an image of text slips past a guard that would block it in plain language -- the decode gap. The natural fix is a guard-agnostic recover-and-decode amplifier that transcribes image content and restates encoded text into its plain payload before the guard, so any off-the-shelf classifier can screen the true request. We build this amplifier and evaluate it against the attacker's best case: an ensemble of eleven attacks, scoring a behavior as broken if any succeeds (best-of-suite, following AutoAttack) -- rarely reported for jailbreak defenses, yet ~3.5x the per-attack mean. This exposes our central finding: an empirical safety-utility ceiling for the non-iterative recovery defenses we evaluate, across five guards and two target VLMs. The amplifier only partly closes the gap -- the undefended ensemble breaks 89-91% of behaviors, and the best guard-plus-amplifier still leaves 63-65% -- and its gain over the guard alone is significant in only four of ten guard-target pairs. It is guard-agnostic at the interface, but not uniformly so in effect. A modular reguard layer closes much of the residual, yet drives benign over-refusal to 81-92% for well-calibrated guards; the one laxer guard that stays usable never reaches deployable safety (48% ensemble ASR). No configuration we evaluate reaches both low attack-success and low over-refusal, for the pipeline we study and for representation-shifting attacks -- encodings and cross-modal renders that leave a legible payload, not pixel- or embedding-space attacks. We contribute the amplifier, an ensemble evaluation that makes the trade-off visible, and a map of where recovery-based VLM defense works and where it does not.
Authors: Nand Lal Yadav, Rajesh Kumar, Satyendra Singh, Sudhakar Singh
Abstract: With the increase in the number of cases related to respiratory diseases, there is an urgent need to detect them early and diagnose them accurately. Convolutional neural networks have given promising results when used for diagnosing diseases using imaging tests. In this study, we investigate the potential of applying deep learning algorithms such as VGG16, VGG19, and ResNet50 for classification of lung ailments based on X-ray images. A detailed analysis of the aforementioned models' performances was conducted to assess how well they can classify various types of lung ailments, including pneumonia, tuberculosis, lung cancer, and normal lungs. In order to do that, these deep learning models were trained on a vast amount of X-ray images. The results of our study show that while all three models provide good results, ResNet-50 performs best in comparison with other models due to its efficiency and high level of accuracy. We believe that these deep learning models can be successfully implemented in the practice of diagnosing pulmonary diseases in the future. It helps with early disease detection and improves patient outcomes.
Authors: Jingjie Ning, Shanshan Zhong, Xiaochuan Li, Ji Zeng, Chenyan Xiong
Abstract: Automated research systems use experimental scores both to deliver artifacts and to decide which ideas to retain, transfer, and pursue. Yet one run scores one implementation of an idea. Crediting that realization-level score as evidence about the parent mechanism creates the \emph{implementation lottery}, in which an idea-level conclusion depends on which plausible implementation was sampled. The mismatch is structural whenever one run updates beliefs about a mechanism. We estimate its magnitude. The \emph{Idea Reliability Audit} measures \emph{idea reliability} by validating and freezing candidate cards, sampling fresh-session implementations, using outcome-blind fidelity labels, and rerunning saved artifacts. It reports idea ICC and leave-one-implementation-out (LOO) winner reversal. Prior work generally repeats the task; we repeat the idea. Across 312 assignments on 13 tabular tasks and two coding-agent setups, implementation variance was more than five and ten times same-artifact rerun variance, respectively, and the winner from one implementation draw differed from the winner under the other-two mean in 25.6\% and 43.6\% of decisions. Reversal survives card-level filtering under two outcome-blind review rules. An exploratory diagnostic on three materials-regression workflows with a deterministic evaluator also finds implementation variation dominating the decomposition. These findings distinguish idea reliability from best-of-$N$ artifact utility. Before a score guides idea-level branching, transfer, or research memory, evidence should cover multiple implementations.
Authors: Zhoupeng Shou, Xiaodong Hong, Congjing Ren, Jingdai Wang, Yongrong Yang, Zuwei Liao
Abstract: PID tuning for chemical processes commonly relies on identified process models, whereas plant engineers often retune loops iteratively by observing responses, diagnosing deficiencies, adjusting gains, and validating the result. This work formalizes this engineer-like workflow in a language-model-assisted PID tuning framework applicable to both large and small language models (LLMs/SLMs). Hosted LLMs receive closed-loop response features, control-engineering diagnoses, tuning preferences, and internal model control (IMC)-based demonstrations to generate and iteratively correct PID gains under common acceptance criteria. For local deployment, Qwen3-0.6B is adapted through supervised fine-tuning (SFT) with simulation-verified IMC targets and physics-informed group relative policy optimization (PI-GRPO) with non-compensable stability and performance rewards. On 100 first-order plus dead time (FOPDT) and 100 second-order plus dead time (SOPDT) test cases, hosted LLMs (DeepSeek-V4-Flash and Qwen3.7-Plus) achieve final success rates of 75-89% and 77-79%, respectively. As for Qwen3-0.6B, supervised fine-tuning raises first-recommendation success to 86.5%, and PI-GRPO further increases it to 94.0%, primarily improving first-attempt reliability and stability margins.
Authors: Mingkuan Feng, Zhengqi Wen, Jianhua Tao
Abstract: Multimodal large language models (MLLMs) have demonstrated remarkable capabilities by integrating visual and textual understanding within a unified transformer architecture. However, fine-tuning all parameters of these models for visual instruction tuning is computationally expensive and often unnecessary, as the representation requirements for visual and textual tokens diverge significantly in the deeper layers of the network. In this paper, we propose Decoupled Visual Processing (DVP), an efficient training framework that replaces the upper decoder layers of a pretrained LLM with a lightweight, independently trainable single transformer block dedicated exclusively to visual token processing. Specifically, after shared processing through the first half of the decoder layers, visual and textual tokens are split: visual tokens are routed through a newly initialized single transformer block while textual tokens continue through the original frozen decoder layers. The two streams are then concatenated before the language modeling head. During training, only the single transformer block is updated, dramatically reducing the number of trainable parameters. Experiments on the LLaVA-1.5 framework demonstrate that DVP achieves competitive performance on MME, POPE, and ChartQA benchmarks while training only a fraction of the total parameters, suggesting that visual representations in MLLMs can be effectively learned through a decoupled, parameter-efficient pathway.
Authors: Yuetian Du, Yucheng Wang, He Xu, Jiexu Xu, Shanwen Tan, Bing Zhao, Boyu Yang, Zhijie Xu, Ming Kong, Hu Wei, Jie Liu, Qiang Zhu
Abstract: Large language model (LLM) agents may recover from a failure within an episode or after a retry, yet the same execution failure can recur in later tasks because post-episode feedback rarely revises the persistent harness that guides future interactions. Static harnesses improve reliability through fixed tools, context, memory, and workflow structures, but remain unchanged after deployment. We propose $\textbf{Living-Harness}$, a self-evolving agent harness that converts each completed trajectory and its evaluator signals into posterior evidence for bounded harness updates. Guided by a domain-level $\textbf{Evolution-SOP}$ ($\textbf{S}$tandard $\textbf{O}$perating $\textbf{P}$rocedure), Living-Harness extracts an episode abstraction and structured update evidence, and writes two complementary forms of procedural knowledge: episodic memory that records trigger conditions, failure patterns, and recovery actions, and a state graph that records state nodes, repair edges, and transition rules. The updated harness state is retrieved to guide future interactions, while tools and base context remain frozen, allowing procedural repairs to accumulate across evolution cycles. On eight interactive environments derived from $\tau^2$-Bench and MultiWOZ-2.4, Living-Harness improves average Pass@1 over the strongest interactive baseline by 10.07 and 9.91 percentage points, respectively, and supports retrieval-only reuse of the evolved harness state across model backbones.
Authors: Hao Jiang, Peiru Du, Pengfei Yao, Mengting Li, Siyuan Lou, Kuo Cai, Sheng Yu, Qiang Luo, Jian Liang, Ruiming Tang, Fei Pan, Peng Jiang, Wenwu Ou
Abstract: Large language models (LLMs) have demonstrated strong reasoning capabilities, motivating their adoption as backbones for foundation recommendation models (FRMs). Existing approaches typically enhance recommendation with explicit Chain-of-Thought (CoT) under the Think-then-Answer paradigm. However, generating lengthy rationales introduces substantial inference overhead, while fixed CoT templates struggle to model diverse, dynamic, and context-dependent user interests. We propose WhisperRec, an efficient latent reasoning framework for FRMs. WhisperRec compresses teacher-generated CoT into learnable latent reasoning tokens, enabling a Latent-Reason-then-Answer paradigm that performs reasoning in latent space without producing verbose rationales. This design retains decision-relevant reasoning information while avoiding the latency bottleneck of autoregressive rationale generation. Specifically, it first introduces Multi-View Adaptive CoT (MV-ACoT) to construct diverse, high-quality supervision from complementary perspectives on user interests. MV-ACoT also adapts reasoning complexity to each instance, applying lightweight analysis to clear cases and targeted multi-factor reasoning to challenging ones. Building on a pre-trained FRM, WhisperRec then employs a three-stage Latent Reasoning Alignment procedure to progressively internalize teacher CoT into latent representations. Finally, curriculum-based post-training activates latent-token reasoning for downstream recommendation while preserving standard recommendation capability. Experiments on an industrial-scale Kuaishou dataset and the public Kuaishou LLM-Rec benchmark show that WhisperRec consistently outperforms explicit-CoT methods and conventional baselines. Compared with explicit CoT Think and No-Think variants, WhisperRec improves SID@64 by 17.44% and 9.33%, respectively, and achieves over 10x higher online inference throughput.
Authors: Mohammed Abdullah
Abstract: TabPFN performs classification through in-context learning: it conditions on a set of labeled training rows (the context, or prototypes) and predicts test labels without gradient updates. On small tabular datasets, practitioners must still choose the context size and which rows constitute the context. We study how these choices affect prediction stability, accuracy, and selection cost using repeated context sampling on 15 OpenML datasets. Specifically, we investigate (i) whether larger contexts reduce prediction variability across random draws, (ii) whether accuracy depends on preserving the training distribution or on feature-space coverage, and (iii) whether expensive selection methods such as K-Means and farthest-point sampling provide benefits over uniform random sampling. We find that larger contexts are both more accurate and substantially more stable, with AUC coefficient of variation decreasing from roughly 6 to 18% at k=16 to 1 to 4% at larger context sizes on datasets with room for improvement. Although accuracy correlates with distribution representativeness in random contexts, controlled experiments show that matching feature means alone can reduce accuracy by up to 0.5 AUC because it reduces context diversity. Mixed-effects analysis identifies diversity and coverage, rather than feature-mean matching, as the stronger predictor of accuracy (diversity beta=+0.23, p=3x10^-12; feature-mean shift beta=-0.01, p=0.71). K-Means and farthest-point sampling achieve similar accuracy to random selection while requiring two to three orders of magnitude more selection cost. These results show that random sampling succeeds because it provides feature-space coverage in expectation, not because it reproduces the underlying data distribution.
Authors: Yinan Gao, Jiarong Xu, Xiaohang Zhao, Xiao Fang
Abstract: Organizational digitalization expands cybersecurity risks, making cybersecurity an increasingly important research area in Information Systems (IS). Among these risks, malware has become a pervasive and destructive threat. Byte-based machine learning (ML) methods are widely used for malware detection but remain vulnerable to evasive behaviors that manipulate raw bytes to evade detection. Graph-based methods are less affected by such manipulations because they represent software as program graphs that capture execution behavior. However, they do not explicitly identify cohesive groups of basic blocks that jointly realize meaningful program behaviors, nor do they learn sufficiently expressive program graph representations for accurate detection. To this end, we propose MalGuard, a graph-based malware detection method for organizational malware risk management. MalGuard introduces two methodological innovations: an operational role identification approach and a program graph representation learning method. The former identifies these cohesive groups of basic blocks as operational roles, enabling the detector to capture program behaviors that may not be visible from isolated basic blocks. The latter learns expressive program graph representations by modeling interactions among operational roles, preserving sparse malicious signals, and capturing hierarchical graph structure. Extensive experiments show that MalGuard improves detection performance and reduces the expected cost of undetected malware.
Authors: Sizhe Zhou, Sheldon Yu, Hui Wei, Junda Wu, Siru Ouyang, Yizhu Jiao, Shijia Pan, Julian McAuley, Yu Zhang, Tong Yu, Jiawei Han
Abstract: Deployed LLM agents increasingly keep their long-term memory as a filesystem: a directory tree of markdown files that the agent itself reads, writes, and reorganizes through generic file tools. Yet research has largely passed over this medium: prior systems design bespoke memory representations and study retrieval over them, leaving the default's two working assumptions untested: that an agent can keep a growing store organized as memories accumulate, conflict, and go stale, and that this organization pays. We present the first systematic exploration of filesystem-based memory for LLM agents. We formalize the setting as three roles around one memory filesystem: a management agent integrates and organizes incoming content, a search agent answers queries with cited sources, and an execution agent supplies task trajectories that are distilled into skills, unifying declarative memory and skills in a single store. Across long-conversation benchmarks and embodied tasks, we vary memory shape (agent-organized hierarchy, verbatim dump, chunk retrieval), stream scale, tool harness (sandboxed shell, memory-tool-style functions, varied search tooling), and the strengths of the management and search agents, tracking answer quality, cost, and store health as memory grows. What organization reliably buys is search economy: organized stores roughly halve retrieval cost where material is large. Today's agents, however, fall short of the default's promise: in our growth study, organization erodes for all but the strongest management agent, and no agent we measure converts organization itself into better answers. And the model is not the only lever over a store's shape: changing the tool set alone reshapes the store as strongly as swapping the model. The study turns the filesystem default from an assumption into a design space for agent memory.
Authors: Haoyu Zhang, Shibo Zheng, Xiangchen Guan, Zhuoxi Wang, Zijian Xiao, Mohammad Zandsalimy, Shanu Sushmita
Abstract: A self-check defense asks the target model to assess a request before answering it; SAGE, the strongest published instance, reports an average 99% defense success rate. We show it can be breached by composing two attacks that are individually harmless against it: an established code-completion encoding and an established best-of-N search, neither of which exceeds 4.7% of behaviors alone. Composed, with the search budget spent on the encoding, they reach 67/22/15% across three open targets, and the effect persists on a 70B target. We then explain the composition rather than only reporting it. First, a self-check defense borrows its strength from the target: SAGE does not detect the attack, it asks the model to, and the four targets convert that request into an explicit refusal between 32% and 97% of the time, which orders the spread in defended coverage even though undefended reach is near-identical. Second, which attack survives is decided by the type of defense, and it inverts: against transform defenses the code encoding retains far more of its undefended reach than the character search, while against gate defenses the ordering flips. We account for this with the number of independent probes an attack delivers to a defense's decision boundary. Finally, we report a validity defect we found and repaired in our own pipeline, a deterministic attack under greedy decoding has no best-of-N variation channel at all, and give the one-line diagnostic that detects it. All claims rest on 310,000 generations scored by a human-validated judge.
Authors: Mu\~noz-Haro Javier, Teruel Andres, Tolosana Ruben, DeAlcala Daniel, Vera-Rodriguez Ruben, Morales Aythami, Fierrez Julian
Abstract: Identity document (ID) authentication relies on the structural integrity of complex, high-frequency security patterns. However, advanced Generative AI models can now inject localized, high-fidelity manipulations, creating deceptive attacks that bypass standard verification. Training robust image forensic models to detect these anomalies is hindered by privacy regulations, forcing reliance on synthetic templates lacking the intricate visual patterns of real IDs. To bridge this domain gap, we introduce FakeIDet3-DB, the first comprehensive database of digital manipulations on real, government-issued IDs. FakeIDet3-DB encompasses classical (e.g., copy-move) and Generative AI-driven manipulations (e.g., face-swapping, inpainting) enhanced with advanced image refinement procedures to suppress visual artifacts. In addition, to comply with strict data protection regulations (e.g., GDPR), we adopt a recently-proposed framework based on patches. In order to maximize forensic utility, we formulate privacy-aware patch extraction from a real ID as a geometrically constrained image processing problem. We propose PACE, a Pseudo-Anonymized Contextual patch Extraction algorithm, which leverages Integral Image mapping and distance-driven Non-Maximum Suppression (NMS). PACE efficiently contours anonymization masks that prevent Personally Identifiable Information (PII) leakage while maximizing semantic density in peri-censorship regions, yielding almost 5.2M patches extracted from more than 6.4K images from real/fake IDs. Furthermore, an extensive evaluation of the proposed FakeIDet3-DB is performed using state-of-the-art models, showcasing they all struggle to detect and locate attacks coming from generative and classic techniques (32.45\% EER in detection and 83.48\% AUC-ROC in localization).
Authors: Wenzhe He, Meng Wang, JiaWei Qian, Jinfeng Xu, Ying Liu, Ruihui Li
Abstract: Existing point-based generative methods for outdoor scenes primarily focus on LiDAR-conditioned completion. During training, noisy point clouds are constructed by perturbing complete ground-truth scenes, whereas during inference, they are initialized by adding noise to duplicated partial scans. This train-inference mismatch inherits the sparsity and visibility bias of partial scans, leading to sparse distant regions and incomplete geometry in occluded areas. Moreover, the reliance on partial scans restricts generation when LiDAR observations are unavailable or replaced by layout cues. We present FPSGen, a flexible framework that constructs point sources independently of partial scans. FPSGen first predicts a bird's-eye-view (BEV) prior with density, height, and mask channels from the active cues. The density map is then sampled to form a BEV-supported point source, enabling both unconditional and conditioned initialization. A teacher-student approximate optimal transport scheme then uses teacher-predicted endpoints to learn a velocity field that induces straighter transport paths. By integrating BEV point source construction with path-straightening transport, FPSGen provides a unified framework for unconditional and flexible cue-conditioned scene generation. Extensive experiments show that FPSGen achieves state-of-the-art JSD and voxel IoU performance on SemanticKITTI completion while maintaining strong performance with a single point transport step. On KITTI-360 unconditional generation, it also achieves the best Coverage (COV) among the compared methods.
Authors: Shen You, Wei Jiang, Jiarui Liu, Yijian Ye, Qiuzhen Lin, Xiangtao Li, Ka-Chun Wong
Abstract: With the promising performance of deep neural networks on image-based tasks, different real-world applications such as autonomous driving and motion detection have become increasingly mature and relevant to human lives. In particular, Optical Flow Estimation Networks (OFENs), as upstream models, play a critical role in different domains. Its outputs are heavily assumed and adopted for different downstream tasks, and it is essential to test its robustness to prevent safety accidents. We present an approach for real-time attacks on OFENs in the physical world, leveraging infrared lights for their stealthiness. By generating a large number of Adversarial Examples in advance, our approach computes AEs in real time and dynamically displays them, which allows our method to facilitate precise and targeted attacks without modifying the victim system. Unlike previous digital-to-physical attack techniques, our method directly attacks victim models within the physical world, thereby overcoming the limitations associated with the ineffectiveness of AEs. Experimental results demonstrate the efficacy of our approach in compromising OFENs across diverse lighting conditions, varying object motion velocities, and different object placements, ultimately impairing the network's ability to accurately estimate optical flow.
Authors: Desiree Cho, Cameron Tice, Bernie Hogan, Hunar Batra, Puria Radmard, Jun Zhao, Nigel Shadbolt
Abstract: Post-training alignment is often shallow, eroding under fine-tuning. It remains untested as to whether constitutional midtraining interventions can produce durable alignment when cleanly isolated from post-training. We build a 394M-token constitutional corpus from Anthropic's Constitution and apply constitutional midtraining at 120B scale, where principled, values-based content is inserted into midtraining. A 2x2 design (curriculum ordering x deliberative reasoning) was used to produce four constitutionally midtrained conditions, plus a control, which were evaluated on self-generated and established benchmarks including alignment under pressure, value conflict resolution, blackmail, and emergent misalignment. All models were evaluated across three stages: post-midtraining, post-SFT, and post-benign fine-tuning. Constitutionally midtrained models outperformed the control on alignment generalization and durability, notably on blackmail: SFT instilled a blackmail propensity in all models, but constitutional midtraining blunted it, with the advantage surviving benign fine-tuning (-17.5pp). This durability did not extend to settings that required active resistance to in-context pressure or conflict, where the advantage attenuates after SFT. The presence of constitutional content at midtraining also mattered more than its structure, and constitutional midtraining incurred no capability cost, on average, at any stage (MMLU, ARC-Easy, piqa, GSM8K). A modest amount of constitutional content at midtraining could therefore yield broad, persistent alignment gains, offering a cheap, complementary addition to SFT-centered pipelines. Code, data, and models are available.
Authors: Yikun Li, Ting Zhang, Jiakun Liu, Jinfeng Jiang, Yuheng Yieh, Yixin Yang, Wen Bin Leow, Yide Yin, Yintong Huo, Eng Lieh Ouh, Lwin Khin Shar, David Lo
Abstract: Real-world vulnerabilities often span multiple functions, yet most learning-based detectors classify each function in isolation: on a sample of real CVEs, we find that 71.7% of vulnerable functions require evidence from outside the function to be classified correctly. Agentic reinforcement learning (RL) could close this gap by enabling a model to gather that evidence itself, but it lacks a reliable reward, since a reward defined on the final verdict alone can be obtained without performing any investigation. We propose VulAgentRL, an agentic RL framework for interprocedural vulnerability detection built on a Code Property Graph (CPG). The CPG serves two roles: at inference time the policy queries it for callers, callees, dataflow, and other queries, and at training time the same graph verifies the evidence the policy cites. Because every CPG node carries a persistent integer identifier, this verification is an exact comparison rather than a textual match, so the reward credits verdicts that are supported by evidence. We further initialize the policy by distilling teacher investigations, and show that this warm start is necessary, since RL cannot acquire tool-use behavior it never samples. Under a repository-level split that prevents leakage, VulAgentRL outperforms state-of-the-art baselines, including frontier models, on the strict pair-wise-correct metric while issuing fewer tool calls, and its advantage persists on an out-of-distribution corpus and under class imbalance.
Authors: Eleni Adamidi, Serafeim Chatzopoulos, Thanasis Vergoulis
Abstract: The rapid growth of scholarly literature has made identifying relevant publications increasingly difficult, and conventional search systems still depend heavily on manually formulated queries and effortful manual inspection. Generative large language models (LLMs) offer a more flexible alternative, supporting literature retrieval and the screening of candidate studies against eligibility criteria. This chapter surveys 34 peer-reviewed papers applying generative LLMs to these two tasks, identified via a Boolean search over the OpenAIRE Graph (1,589 records screened to 34 inclusions). Reviewed studies are characterised by LLMs employed, model access and adaptation, prompting and architectural techniques, ground-truth sources, and evaluation metrics.
Authors: Mingxuan Che, Tsung-Yuan Tseng, Theresa Eimer, Marius Lindauer, Alexander von Rohr
Abstract: Reinforcement learning (RL) has shown remarkable success across a wide range of complex tasks. However, RL outcomes can be highly stochastic, and both expected performance and variability often depend on hyperparameter (HP) configurations. We propose efficient and risk-averse heteroscedastic Bayesian Optimization (ERAHBO), a Bayesian optimization method that models both the mean and variance of learning outcomes as functions of the HP configurations. ERAHBO aims to identify HP configurations that achieve high average return while reducing variability across training runs, and it improves the sample efficiency of the HP optimization via adaptive re-sampling rather than a fixed budget per HP. Empirical evaluations across diverse RL algorithms and environments demonstrate that ERAHBO generally outperforms both risk-neutral and risk-averse baselines, delivering improved sample efficiency for risk-averse returns.
Authors: Wei-Jaw Lee, Hsuan-Yu Yeh, Ting-Yi Hu, Chih-Pin Tan, Fang-Duo Tsai, Yi-Hsuan Yang
Abstract: Cover song generation (CSG) should preserve the melodic and linguistic content of a reference song while recreating the remaining musical components. The state-of-the-art model SongEcho utilizes $F_0$ sequences and voiced/unvoiced (V/UV) tags for conditioning; however, implicit linguistic information from V/UV tags cannot guarantee lyric accuracy, leading to a high phoneme error rate (PER). Inspired by singing voice synthesis (SVS), we propose MPEcho, which integrates a phoneme encoder and a length regulator (LR) into the SongEcho framework. By providing explicit phoneme-level conditioning and precise temporal boundaries, MPEcho significantly reduces PER. To enable this, we developed Phonsa, a Whisper-based automatic transcription model that provides high-precision phoneme-level annotations for singing voices, overcoming the scarcity of high-quality audio-phoneme pairs. Experimental results validate the effectiveness of Phonsa for alignment and MPEcho for end-to-end CSG. The audio samples, code and weights can be accessed from https://lonian6.github.io/MPEcho.github.io/.
Authors: Tanjim Taharat Aurpa
Abstract: The Mpox outbreak remains a serious public health issue, with the WHO (World Health Organization) reporting increasing cases in some regions. Research on Mpox is vital for several reasons, including vaccine development, diagnostic improvement, viral evolution studies, and preventing future outbreaks. However, the large amount of research being published makes it difficult to organize and analyze information efficiently. This study focuses on using multilabel classification to categorize 14590 Mpox research articles into key topics such as outbreaks, vaccination, and epidemiology. Among the different AI models tested, BERT performed the best, achieving 97.05% accuracy, 97.67% micro F1 score, and 96.46% macro F1 score. To better understand how the model makes decisions, SHAP was used to analyze significant word features and patterns. The results show that BERT can help automate the classification of Mpox research, making it easier for researchers, policymakers, and healthcare workers to quickly find relevant information, saving time and improving public health efforts.
Authors: Jindong Yang, Han Fang, Weiming Zhang, Nenghai Yu, Kejiang Chen
Abstract: Inversion-based watermarking is a promising approach to authenticate diffusion-generated images, yet practical use is bottlenecked by inversion that is both slow and error-prone. While the primary challenge in the watermarking setting is robustness against external distortions, existing approaches over-optimize internal truncation error, and because that error scales with the sampler step size, they are inherently confined to high-NFE (number of function evaluations) regimes that cannot meet the dual demands of speed and robustness. In this work, we have two key observations: (i) the inversion trajectory has markedly lower curvature than the forward generation path does, making it highly compressible and amenable to low-NFE approximation; and (ii) in inversion for watermark verification, the trade-off between speed and truncation error is less critical, since external distortions dominate the error. A faster inverter provides a dual benefit: it is not only more efficient, but it also enables end-to-end adversarial training to directly target robustness, a task that is computationally prohibitive for the original, lengthy inversion trajectories. Building on this, we propose \textbf{FARI} (\textbf{F}ast \textbf{A}symmetric \textbf{R}obust \textbf{I}nversion), a one-step inversion framework paired with lightweight adversarial LoRA fine-tuning of the denoiser for watermark extraction. While consolidation slightly increases internal error, FARI delivers large gains in both speed and robustness: with approximately 20 minutes of fine-tuning on a single NVIDIA RTX A6000 GPU, it surpasses 50-step DDIM inversion on watermark-verification robustness while dramatically reducing inference time. Code and pretrained models are available at https://github.com/0xD009/FARI.
Authors: Xiaolong Liu, Junjian Li, Yuan Xiao, Jiaqi Deng, Dayong Ye, Tianqing Zhu, Huan Huo
Abstract: Prompt inversion, as a typical reverse engineering technique, enables text-to-image (T2I) diffusion models to generate the desired target images without extensive prompt engineering. However, existing prompt inversion methods suffer from significant limitations: (1) gradient-based methods are unstable and uninterpretable, often resulting in generated images with severe artifacts; (2) gradient-free methods yield human-readable prompts but still fail to preserve visual fidelity due to the lack of fine-grained detail alignment. We contend that the limitations stem from treating prompt inversion as a sufficient condition for reverse engineering, ignoring the critical role of the latent noise that encodes structural information. Consequently, we propose Dualin (Dual inversion), a two-stage method that jointly recovers both the semantic prompt and latent noise of the target image. In the first stage, we integrate vision-language model, CLIP and large language model to invert a faithful, human-interpretable hard prompt. In the second stage, unconditional DDIM inversion reconstructs the exact latent noise of the target image, guaranteeing the consistency at the structural information level. Theoretically, we prove that the inverted noise enables flexible image editing without re-optimization. Extensive experiments on diverse datasets demonstrate that Dualin simultaneously generates high-quality inverted prompts and achieves state-of-the-art image fidelity. Additionally, Dualin can establish a robust foundation for the precise and controllable image editing.
Authors: Carlos Mu\~noz-Romero, Jose A. Gonzalez-Lopez
Abstract: Zero-shot text-to-speech (TTS) clones a voice from a short audio prompt, but this reliance on reference audio is a barrier when only visual information is available, e.g. for historical figures or video-game characters. In this work, we propose a Face-to-Speech (F2S) framework that predicts a plausible voice from a static facial image. A lightweight Face Adapter, together with soft-tuning of the face encoder's upper blocks, aligns face-recognition features with the style space of a frozen StyleTTS 2 model, kept frozen during training. We evaluate on held-out identities from LRS3, a large-scale audiovisual corpus of English TED-talk videos. The synthesized speech is highly natural (UTMOS 3.7-4.0, matching or exceeding the 3.61 of ground truth), face-to-voice retrieval is consistently above chance, and the generated voice is consistent with the target speaker. Without any retraining, an English-trained adapter also produces fluent Spanish speech, indicating that the face-to-style mapping is largely language-agnostic.
Authors: Nevio Dubbini, Lisa Yeomans, Marco Pavia, Ramazan Parmaksiz, Ayse Atas Hooglugt, Gabriele Gattiglia, Beatrice Demarchi
Abstract: Artificial intelligence has shown considerable potential for archaeological applications, yet its use in zooarchaeology remains limited, particularly for the identification of avian skeletal remains. This study presents a proof-of-concept multimodal framework that integrates convolutional neural network-based image analysis with osteometric measurements for the classification of bird bones. Using a dataset of more than 10,000 images from multiple museum and research collections, two classification tasks were investigated: skeletal element identification and family-level taxonomic classification. Prior to classification, images were automatically segmented using a two-stage pipeline combining BiRefNet and SAM2. Visual features extracted with a pre-trained EfficientNet_V2_S backbone were fused with standardized morphometric data through a feature-level multimodal architecture. The model achieved 86% accuracy on the test set for bone-type classification, demonstrating reliable recognition of skeletal elements. Family-level classification proved more challenging, reaching 51% top-1 accuracy but 75% top-3 accuracy, indicating that correct taxa were frequently included among the most probable predictions. These results demonstrate the feasibility of combining visual and morphometric information within a unified deep-learning framework and establish a methodological baseline for future AI-assisted zooarchaeological identification. The approach contributes to ongoing efforts to develop scalable, interpretable, and archaeologically meaningful tools for the study of avian remains.
Authors: Harshiddhi Pathak, Gowtham Reddy N, Mrinal Acharya, Manjunatha Mahadevappa
Abstract: Accurate identification of Alzheimers disease (AD) using resting-state functional magnetic resonance imaging (rs-fMRI) remains challenging due to the high dimensionality, noise, and complex inter-regional dependencies inherent in functional brain connectivity, which limit the effectiveness of traditional approaches based on handcrafted connectivity features or conventional machine learning models. In this work, we present an attention-based deep learning framework for Alzheimers disease classification that operates directly on rs-fMRI functional connectivity matrices by treating brain regions as tokens and employing a Transformer-inspired self-attention mechanism to model long-range and global functional dependencies across distributed brain networks. The proposed framework learns discriminative functional representations without reliance on manual feature engineering and is evaluated on a longitudinal cohort from the Alzheimers Disease Neuroimaging Initiative (ADNI) comprising cognitively normal and Alzheimers disease subjects with multiple visits. A subject-wise evaluation protocol is adopted to prevent information leakage across visits, and class-weighted optimization is incorporated to address mild class imbalance. Experimental results for binary AD versus cognitively normal classification demonstrate that the proposed attention- based rs-fMRI model achieves an accuracy of 88.95% and a ROC-AUC of 0.90, along with a favorable precision-recall balance, highlighting the effectiveness of self-attention-driven functional connectivity modeling as a robust and interpretable approach for Alzheimers disease detection using resting-state fMRI.
Authors: Lucas Zamora Vera, Jose A. Gonzalez-Lopez
Abstract: State-of-the-art intracortical brain-to-text systems pair a neural-sequence phone decoder with an external language model. Two design axes remain underexplored: whether selective state-space models (Mamba) improve on recurrent decoders, and how the output target (phonetic vs.\ character) interacts with that choice. On the public Brain-to-Text '25 benchmark, we study a controlled 2x2 grid (GRU vs.\ hybrid Mamba decoder; phonetic vs.\ character targets) trained with a CTC objective under one reproducible protocol. The recurrent baseline remains strongest: the best phonetic GRU reaches 12.62\% PER and 21.19\% WER, while the best textual GRU after LM rescoring reaches 13.39\% CER and 26.28\% WER. The Mamba hybrid is competitive but does not surpass it. Ablations isolate architectural contributions, and error analysis shows representation-dependent failures: articulatory-like phoneme confusions vs.\ lexical and word-boundary errors.
Authors: Alexander Kozachok, Ilya Latyshev, Evgeny Karpulevich, Elena Kozachok, Egor Ushakov, Oleg Samovarov
Abstract: Background/Objectives: Dermoscopic skin lesion classifiers often lose accuracy under domain shift across imaging devices, illumination, and capture artifacts. We study how data augmentation improves the robustness of a binary malignant-versus-non-malignant classifier, with emphasis on out-of-domain (OOD) generalization. Methods: Single augmentations, photometric combinations, and composite policies were searched on a multi-source ISIC Archive collection with Derm7pt, using a ConvNeXt-Large backbone and ROC-AUC. Splits were made at the lesion-ID level, and HAM10000 and ISIC 2019-2020 were held out as a predominantly source-disjoint OOD test. Results: The largest OOD gain came from the mix policy, and photometric transformations dominated the most useful OOD operations. On an expanded pool from the same held-out sources the gain was +0.053 (95% CI +0.045 to +0.061, p<0.001), consistent across four training seeds (per-seed ROC-AUC: baseline 0.761-0.775, mix 0.806-0.829). On a small independent clinical collection, single-checkpoint sensitivity rose from 0.591 to 0.818, but this rested on 22 malignant cases and did not persist across seeds. Conclusions: Augmentations modelling real sources of domain shift can matter more than maximizing in-domain accuracy. Because the policy was selected on the same sources used to evaluate it, a source-disjoint selection protocol is needed before this effect size can be read as unbiased.
Authors: Francesco Tosoni
Abstract: Code search in large-scale ecosystems is often hindered by the lexical gap between user queries and implementation details, alongside the trade-off between the low latency of traditional Information Retrieval (IR) and the precision of Deep Learning (DL). We present MediaWiki Code2Code Search, a neural retrieval system for semantic code-to-code discovery. By indexing 1.29 million structural entities (functions, types, and templates) across 2,500+ MediaWiki repositories, our system enables retrieval based on computational intent rather than surface tokens. We employ a split-build architecture, decoupling GPU-intensive offline indexing from a CPU-only serving layer; our FAISS IVF-PQ index occupies 168.6 MB: a 96.6\% reduction compared to a flat float32 baseline, and achieves a median query latency of 1.85 seconds on commodity hardware, satisfying the 6 GiB RAM constraint of Wikimedia Toolforge. Our evaluation across a 27-query benchmark demonstrates superior performance over the BM25 baseline, achieving a P@10 of 0.87 compared to 0.64 (0.52 versus 0.34 for strict matching). Gains are most pronounced in name-obfuscated tasks where lexical methods fail. The system is available at https://code2codesearch.toolforge.org under the Apache 2.0 licence and provides an open RESTful API.
Authors: Siyu Yan, Zhuoran Yan, Haiying Xu, Panhao Zhou, Jingyu Chen, Chenhao Ji, Shuo Cao, Yongheng Zhang, Haoze Liu, Siyu Zhang, Xiwen Gu, Yihao Liu, Alex Jinpeng Wang
Abstract: Multimodal large language models increasingly use sketches, annotations, tools, and intermediate images during reasoning, but it remains unclear whether they truly rely on these visual states. Existing benchmarks are limited both by task collections with narrow coverage or partially text-solvable samples and by evaluations that emphasize final answers without diagnosing how intermediate visual states are generated, rendered, and used. We introduce See2Think, a unified evaluation framework comprising See2ThinkBench and Visual Action-of-Thought (VAoT). See2ThinkBench contains 1,200 open-ended, visually dependent problems across 12 task categories spanning 2D structured, 3D scene, and real-world reasoning. VAoT records textual thoughts, visual actions, rendered states, and subsequent reasoning under four controlled inference settings. Evaluating representative proprietary and open-source multimodal models, we find that visual reasoning is strongly model- and environment-dependent, with no single setting consistently dominating across tasks. Process analysis further shows that models usually select relevant visual operations, while faithful rendering remains the clearest bottleneck and high feedback uptake does not necessarily translate into accuracy gains. Under task-relevant corrupted feedback, models exhibit behavioral dependence on visual states, with accuracy dropping by over 10 percentage points in controlled interventions.
Authors: Mahesh Godavarti
Abstract: Many kinds of data have structure along one or more axes: words in a sentence, pixels in an image, nodes in a tree, frames in audio, or cells in a 3D volume. Along one axis, order matters: "the dog bit the man" is different from "the man bit the dog." Across independent axes, however, neither composition nor movement should depend on the order of axes: in an image, composing right then down should give the same result as composing down then right, and moving right then down should describe the same relative position as moving down then right. We develop a framework for modeling this kind of multi-axis structure. Each data item carries its content together with a small transformation for each axis. A path connecting two positions defines a journey; the journey operator is the product of per-axis transformations along that path, governing both how data composes along the path and how relative position is described. When the transformations are fixed, our framework recovers Rotary Position Embedding (RoPE) and its multi-dimensional variants. When they depend on the data, the model gains a content-adaptive positional inductive bias. We show exactly when these paths are well-defined: both composition and movement across axes are path-independent precisely when the axis transformations commute. We also prove that, under the stated toral-frame symmetry, cocycle, bilinearity, and norm-preservation assumptions, the resulting pairwise scoring rule must take the form of block-wise rotations, explaining why RoPE-like methods arise naturally. Finally, we use this theory to design JoFormer, a model for value aggregation, and relate it to attention and state-space models (SSMs). Initial experiments across vision, language, and length generalization suggest that these inductive biases can have observable consequences in practice.
Authors: Zhiyuan Yao, Yuxin Chen, Zhengxi Lu, Zishan Xu, Yueqing Sun, Yifu Guo, Yuquan Lu, Zhengzhou Cai, Kangning Zhang, Zhuowen Han, Zi-Han Wang, Ziang Ye, Qi Gu, Xunliang Cai, Weiwen Liu, Yongliang Shen
Abstract: Large language model agents often encounter related yet distinct tasks that share reusable solution patterns. Yet standard agentic reinforcement learning treats tasks as independent episodes, while existing approaches to skill learning either focus on repeated attempts of one task or use pipelines with multiple stages that entangle extraction, retrieval, and execution. We introduce SkillRise, a unified reinforcement learning framework for learning skills across tasks. SkillRise organizes related instances into progressively challenging sequences and uses a single policy to alternate between task solving and curating an evolving skill document passed directly to the next task. Decoupled credit assignment across tasks supervises solving with the current task outcome and curation with discounted downstream outcomes. Experiments on ALFWorld, WebShop, and ScienceWorld show that SkillRise achieves the strongest Pass@1 performance among the compared methods, with gains over the strongest baseline ranging from 2.3 to 8.5 percentage points. Although trained across distinct tasks, its learned curation policy remains effective for repeated attempts on the same task. Further analysis reveals scaling at test time across tasks: performance improves with longer sequences of related tasks even when each task is attempted only once. This trend suggests that SkillRise reuses transferable skills across tasks rather than benefiting from repeated sampling of the same task. SkillRise further retains strong performance while substantially reducing the runtime overhead of skill learning pipelines with multiple stages. Together, these results provide a simple and efficient training paradigm for LLM agents to extract, refine, and reuse transferable skills across tasks.
Authors: Lehan Wang, Boli Chen, Ruixue Ding, Pengjun Xie, Jinwei Huang, Zhendong Liu, Shuo Wang, Tao Lei, Xin Ouyang, Xiaomeng Li
Abstract: Large Language Model (LLM) agents are increasingly adopted in real-world security operations with access to host artifacts and command-line interfaces (CLIs), making it critical to thoroughly assess their security capabilities. However, existing cybersecurity benchmarks focus on pre-compromise settings where agents are placed in a clean and idealized environment before an attack occurs. This leaves the post-compromise setting underexplored. To address this gap, we introduce SecRespond, the first benchmark for evaluating LLM agents on the post-compromise incident-response workflow. Given a forensic disk snapshot of a compromised host together with the alerts, vulnerability scans, and baseline checks reported by a host security product, agents are required to produce forensic reports on intrusions, baseline risks, and vulnerability risks, together with a remediation plan. We instantiate this task across 10 cyber ranges, each constructed from a distinct compromised cloud host, spanning 4 entry-point types, 21 ATT&CK techniques, and 5 operating systems. We evaluate 23 frontier LLMs on the OpenCode agent harness. Experimental results show that although current agents can reliably uncover the problems exposed by alerts, they struggle to proactively investigate the disk for silent intrusions and to produce comprehensive, verified remediation plans, with no model achieving complete detection and remediation on any single range. This reveals a fundamental bottleneck in building agents for real-world incident response. The benchmark is publicly available at https://github.com/Alibaba-NLP/qqr/tree/main/data/secrespond.
URLs: https://github.com/Alibaba-NLP/qqr/tree/main/data/secrespond.
Authors: Runyao Yu, Yuchen Tao, Yujie Chen, Wentao Wang, Derek W. Bunn
Abstract: Probabilistic K-line forecasting describes uncertainty in four complementary prices, namely open--high--low--close (OHLC). However, it introduces two consistency problems: quantile crossing and K-line crossing. Quantile crossing occurs when a higher-quantile forecast falls below a lower-quantile forecast, while K-line crossing occurs when the forecast low exceeds the open or close, or the forecast high falls below the open or close. Existing solutions generally address only one problem through output reordering, specialized architectures, or penalized training objectives. We propose K-line--Quantile Sequential Projection (KQSP), a parameter-free and training-free reconciliation method applicable to forecasts produced by any model. Compared with other crossing solutions, KQSP preserves predictive accuracy while producing substantially smaller corrections to the original forecasts. To mitigate model bias, we evaluate KQSP using various models, including pretrained foundation models. KQSP reduces both quantile and K-line crossing rates to zero for all test data undertaken. These results show that probabilistic K-line consistency can be enforced independently of forecast generation and without retraining.
Authors: Zhaoyang Ma, Zhihao Wu, Xin Gao, Lipo Wang, Youfang Lin, Jing Wang
Abstract: Federated learning (FL) enables collaborative learning over decentralized data silos without centralizing raw data. However, heterogeneous local architectures often induce non-aligned representation spaces, making it difficult to transfer global knowledge across silos. Existing paradigms share this knowledge as model parameters, distilled predictions, or class prototypes, yet all encode it in an absolute space that must be aligned across clients. Heterogeneous backbones break this alignment, so the shared knowledge becomes unreliable and misleads local training. We propose FedTopo, a relation-level framework that encodes global knowledge as class relation topology, capturing how classes relate within each client rather than where they lie in feature space. Each client builds its relation topology from local prototypes and uploads it with class statistics. The server then aggregates these relations in a reliability-aware manner that down-weights weakly supported ones, and broadcasts the global topology to clients. The global topology guides local training by emphasizing topology-similar negative classes. Experiments on three datasets under eight heterogeneous backbones show that FedTopo consistently outperforms parameter-, distillation-, and prototype-sharing baselines, with low communication and no inference overhead. Our code is available at https://github.com/Zhaoyang-Ma/FedTopo.
Authors: Wenhao Yang, Runzhi He, Minghui Zhou
Abstract: Open source communities have been flooded with AI-generated contributions. In defense, they have written contribution rules to regulate coding agents' behavior, spanning from a total ban, mandatory disclosure, to verification gates and human sign-offs. Yet, whether coding agents read and follow those rules, and behave in open source repositories, remains unknown. To estimate real-world rule compliance of coding agents, we curate 106 issues from 49 repositories containing AI contribution rules into RepoComplianceBench. We judge the trajectory of each run against the repository's rules, measuring whether the agent refuses to contribute, discloses its assistance truthfully, clears the required verification gates, or escalates critical steps to a human. We also test if extra prompts, rule disclosure, or feedback from the compliance verifier help with the situation. Our experiments on four frontier models show that today's agents almost never proactively retrieve the contribution rules. Agents pick up disclosure and verification with reminder prompts, rule quotes, and verifier feedback; however, they never refuse to contribute in AI-banned repositories under any condition we tested. The status reveals that verification and disclosure issues are solvable with existing mechanisms, yet enforcing bans and human escalations remains an open problem.
Authors: Janin Koch, Xiaohan Liao, G\'ery Casiez
Abstract: Generative AI tools for creative work tend to be designed around the goal of removing friction, on the assumption that smoother iteration and faster output translate into more value for the designer. We argue, however, that this framing leaves out something important about how design ideation works, namely reflection-in-action. The act of accepting, rejecting and reworking candidate ideas is both a path to a final outcome and the process through which designers develop the rationale that allows them to think with their ideas and to communicate them to others. This becomes particularly important in group ideation, where ideas need to be expressed and explained to others to allow the group to extend, reject or combine them further. We suggest that AI in design ideation might be more usefully thought of as a friction agent for reflection rather than as a smoothing agent for output. This reframing opens up a different role for AI in design ideation, one that supports designers in building and carrying rationale rather than substituting for it.
Authors: Yansen Zhang, Yilu Liu, Tianyu Liu, Jiamin Chen, Xiaokun Zhang, Kai Xie, Xue Liu, Chen Ma, Yiyan Qi
Abstract: Large language models increasingly support scientific and algorithmic discovery through inference-time search over evaluated candidates. Existing adaptive discovery controllers assign credit based only on score progress, even though prompt length, retries, and guidance calls cause search actions to incur different token costs. We prove that cost-blind credit can forfeit all but a vanishing fraction of attainable quality as frontiers multiply and costs diverge. Under a fixed search-side token budget, the controller must decide which frontier is improving and whether its gain justifies the realized cost before the budget is exhausted. We introduce \textbf{CostAda}, a cost-calibrated adaptive controller built around \emph{cost-calibrated frontier utility}. The utility values frontier progress relative to realized action cost and conditions that credit on the remaining budget. CostAda uses this signal to control local exploration intensity, frontier allocation, and budgeted tactic intervention. Cost and remaining budget therefore shape the search rather than serving only as accounting variables or a stopping rule. CostAda reaches the strongest baseline's full-budget quality with at most half the budget on twelve of sixteen benchmark--backbone pairs while achieving the strongest mean final quality on all eight benchmarks under GLM-5 and GPT-5.4.
Authors: Ruikang Zhang, Shuo Wang, Qi Su
Abstract: Human personality theories characterize traits not as isolated attributes captured by a single score, but as stable individual tendencies expressed through the interplay among persons, situations, and behaviors. Existing studies of personality-related behavior in LLMs have primarily focused on outputs elicited under personality conditioning, characterizing observable trait-related expressions while lacking mechanistic evidence for the existence of internal personality-related representations, their cross-situational expression, and how these representations shape specific behaviors. Building on Funder's personality triad framework, we adapt its three components for LLM analysis: Person as personality-related internal representations, Situation as contexts that afford trait-relevant responses, and Behavior as response patterns on broader social tasks. We introduce a framework for discovering, controlling, and validating trait-like representations in LLMs. First, using contrastive behavior pairs grounded in shared situations, we identify sparse internal features associated with opposing poles of personality traits through SAE decomposition. We validate their trait relevance through effects on behavior to situation, token-level activation patterns, and robustness to paraphrasing. Second, feature-level interventions induce bidirectional trait-related shifts across a separate, diverse set of situations while preserving response validity, demonstrating consistent expression across contexts. Third, applying the same interventions to social intelligence tasks reveals behavioral changes with benefit-tradeoff patterns consistent with findings from human personality research, providing behavioral-level validation beyond personality scores. Our findings provide evidence that LLMs contain controllable trait-like representations linking internal states, situational expression, and behavioral outcomes.
Authors: Junoh Park, Junseo Hwang, Wonguk Cho, Taesup Kim
Abstract: Group Relative Policy Optimization (GRPO) has become a standard reinforcement learning method for post-training language models. Recent work shows that GRPO can reduce the base model's reasoning capacity and underperform it in Pass@k when k is large, indicating reduced coverage of reasoning paths. We find that this reduction is associated with GRPO concentrating on responses that the base model already generates with high probability. We trace this concentration to two mechanisms in the GRPO update. At the response level, high-probability responses dominate the group gradient through repeated occurrence. At the token level, GRPO's importance ratio scales gradients, further reinforcing tokens that become more likely under the current policy. We propose ReCo, a reweighting method that addresses both effects. Response contributions are normalized by their expected occurrence within the rollout group, and the token-level importance ratio is replaced with a variance-based ratio that gives larger update scale to non-saturated decision points where alternative token choices remain plausible. Across Qwen2.5-Math-1.5B/7B and Llama-3.1-8B-Instruct on five mathematical reasoning benchmarks, ReCo improves Pass@k for large values of k and is comparable to GRPO for small values of k.
Authors: Amirmohammad Farzaneh, Osvaldo Simeone
Abstract: LLM agents following the ReAct paradigm are promising enablers of complex multi-step tasks, including multi-hop question answering, code generation, and control of physical AI systems. Yet, when deployed at the edge, they must tightly manage their reasoning budget while remaining reliable and deferring to a cloud-side model only when local uncertainty is too high to act safely. We propose Think Short, Defer Smart (TSDS), a framework that synergistically integrates a lightweight convergence probe, which halts on-device reasoning once the intended action has stabilized, with a perplexity-based deferral rule that escalates uncertain actions to a cloud-side model. Both mechanisms are jointly calibrated on end-to-end episode trajectories via a multi-objective Learn-Then-Test (LTT) procedure, providing simultaneous finite-sample guarantees on expected episode reward and cloud-call rate. We evaluate TSDS on four ReAct benchmarks spanning arithmetic reasoning (GSM8K), multi-hop question answering (HotpotQA), code generation (MBPP), and multi-step embodied planning (household robot), and compare against thought-calibration-only and calibrated-deferral-only standalone baselines. TSDS reduces per-episode thinking compute by 43%-73% over deferral-only baselines across HotpotQA, MBPP, and the household robot task, while maintaining certified reward and cloud-call rate guarantees.
Authors: Siddharth Vohra
Abstract: When asked to describe a medical image that was never attached, frontier vision-language models do not abstain: they confabulate a diagnosis. We show that this confabulation is not random. It is structured by who the patient is said to be. Across chest X-ray, brain MRI, and dermatology, Claude Opus-4.7, GPT-5.4, and Gemini-3.1-Pro are each queried with only a demographic descriptor and no image, and changing the descriptor systematically shifts the diagnosis returned. Claude concentrates sharply: a 65-year-old white man asking about a skin mole receives Melanoma in nearly every response, and a 32-year-old Black woman asking about her chest X-ray receives a Sarcoidosis diagnosis whose reasoning reads "suspected, based on demographics and classic pattern.'' GPT-5.4's effect is broader, fabricating across every demographic cell we test, most conspicuously naming Sarcoidosis for young Black patients on chest X-ray. Two structural findings sharpen the problem. A hedged regime appears in which the prose acknowledges the missing image while the structured diagnosis field nevertheless names a disease, a dissociation invisible to prose-only audits. And Claude's dermatology effect collapses entirely when 'skin mole' is swapped for 'skin lesion' while GPT-5.4's is preserved, indicating that mirage is a family of distinct failure modes rather than a single phenomenon. Trustworthy VLM deployment in clinical pipelines requires auditing the structured output channel directly, and probe-word sensitivity should be treated as a first-class evaluation dimension
Authors: Mengchen Dong, Hiromu Yakura
Abstract: Diverse human groups produce diverse ideas, the raw material of innovation. Generative AI challenges this engine twice over: everyday AI assistance may homogenize what diverse people create, and AI-simulated diversity may replace the people altogether. We tested both challenges in a preregistered creative metaphor experiment with native (L1) and non-native (L2) English writers, who wrote without AI, with AI-generated ideas (AI ideation), or with AI refining their own ideas (AI refinement). L2 writers contributed more collective diversity than L1 writers, with native-language ideation showing the most diverse pools. AI ideation compressed collective diversity for everyone and left the L2 advantage undetectable, whereas AI refinement preserved both. We then simulated the entire writer pool using personas built from participants' real backgrounds, three model families, native-language prompting, and elevated sampling temperatures. Every simulated pool fell below every human pool, and pushing models further induced diversity only through degenerate text. However, at the individual level, AI ideation raised writers' ratings, pitting private incentives against the collective good, except when L2 writers used their native language, which benefited both. Human diversity remains a valuable creative resource that current AI cannot simulate or sustain; the design of human-AI collaborative workflows determines whether it survives.
Authors: Brandon Gower-Winter, Georg Krempl
Abstract: In many domains such as Palliative Care, Credit Assignment and Recommender Systems, predictions may causally influence the outcomes they predict. This phenomena is known as Outcome Performativity. This paper formalises an approach for detecting Outcome Performativity using prediction intervention called Outcome Performativity A/B Detection (OPAB). OPAB enables the detection of Outcome Performativity by assessing the dissimilarity in outcome distributions produced by different predictions groups (interventions). If that dissimilarity is significant, Outcome Performativity is detected. We derive sample complexity bounds for OPAB under various Outcome Performative assumption classes which we empirically validate. Results show that detecting Outcome Performativity using OPAB is achievable in numerous cases. Results also show the presence of regions of indistinguishability which describe settings where the allotted number of interventions are insufficient for detecting Outcome Performativity. The results of which have broader practical implications for the detectability of Outcome Performativity in settings where samples are scarce, cost-prohibitive or potentially unethical to obtain. The paper concludes with a case study on the efficacy of OPAB on the Open Bandits dataset, and provides directions for future work.
Authors: Zhe Liu, Quan Lu, Zhaohui Du, Zhe Wang, Huanbo Jin, Jiaming Gu, Qi Wang, Ting Xiao, Minting Pan, Dongzhan Zhou
Abstract: Biomedical laboratory robots must navigate to instruments before performing experimental procedures. Existing embodied navigation platforms are designed for household environments and treat a target as an object center or an arbitrary nearby position. This representation is inadequate for laboratory instruments, which must be approached from their operating side while maintaining safe clearance from surrounding equipment. We introduce BioVLN, a simulation platform for developing and evaluating visual-language navigation agents in biomedical laboratories. BioVLN represents each instrument with three regions: its physical body, a surrounding clearance region, and an operation area in front of the usable side. This model is applied consistently to scene generation, target placement, navigation evaluation, and safety analysis, so success depends on reaching a position from which the instrument can be accessed. BioVLN supports procedural scene generation and manually designed environments, producing 47 scenes and 1667 episodes. Standardized navigation and reinforcement-learning interfaces enable trajectory collection and policy training. Experiments show that geometric exploration reaches 74.4--87.5% success, while sampling multiple valid positions in the operation area improves success to 83.3--92.5% and reduces unsafe proximity.
Authors: Hongliang Zhang, Zhongyuan Yu, Guijuan Wang, Tianqing He, Wenshuo Ma, Xiaosong Zhang, Jiguo Yu
Abstract: Federated Learning (FL) is vulnerable to backdoor attacks because of its distributed nature in edge computing scenarios. Existing defense methods show limited efficacy as they overlook the deviations among benign local updates caused by statistical heterogeneity and the stealthiness of backdoor attacks. To tackle these issues, we propose FedDAB, a two-phase method that combines local contrastive regularization with alignment checking, to defend against backdoor attacks. In the first phase, FedDAB introduces a novel model-contrastive term into the local objective to enhance direction and magnitude consistency among benign updates. In the second phase, FedDAB employs an alignment checking strategy to evaluate each local update in terms of overall-direction alignment and parameter-level alignment with historical information, excluding updates that exhibit abnormal alignment patterns from global aggregation. We theoretically prove FedDAB's robustness with a convergence rate of $\mathcal{O}(1/T)$. Extensive experiments show that FedDAB outperforms existing defense methods against backdoor attacks.
Authors: Duzhen Zhang, Yahan Yu, Qiaoyi Su, Jiahua Dong, Tielin Zhang
Abstract: Multimodal Large Language Models (MLLMs) rely on a projector to align visual representations with the language embedding space, making it central to cross-modal understanding. In Multimodal Continual Instruction Tuning (MCIT), however, shifting visual distributions and evolving instruction semantics cause this shared projector to drift, leading to projector-level forgetting, an issue largely overlooked by methods that focus primarily on the LLM backbone. We introduce Progressive Multimodal Alignment (PMA), a framework that enables the projector to adapt continually while preserving previously learned alignment. PMA detects multimodal distribution shifts via a lightweight representation descriptor and progressively expands projector experts only when needed. An expandable router integrates expert outputs based on multimodal features, while the original pretrained projector is retained as a stable alignment anchor. This progressive mechanism balances stability and plasticity with sub-linear parameter growth and serves as a method-agnostic add-on to existing MCIT approaches. Extensive experiments on two recent MCIT benchmarks demonstrate that mitigating projector-level forgetting yields consistent gains over prior state-of-the-art methods when combined with PMA. Moreover, PMA scales across diverse MLLM backbones, demonstrating robust and broadly applicable MCIT performance.
Authors: Gabe Everett, Brice Gunter, Ryan Vander Stelt, Cleiver Ruiz-Martinez, Blake Hull, Juan Rojas
Abstract: Deep reinforcement policy learning directly in physical robots (on-robot learning) remains bottlenecked by slow wall-clock training times. We present SymmGrid, a trajectory level augmentation framework inspired by parallelized symmetries that super-scales group transformations to significantly accelerate on-robot learning in both egocentric and exocentric visual setups. We model a Markov Decision Process (MDP) under a symmetry tree, in which state-action pairs have admissible parallelized invariant transformations that yield a geometric grid structure. The state is modelled with ego- or exocentric images and proprioception information. The latter require special treatment, in the form of homographies, to warp visual scenes in line with their corresponding spatial transformations. These parallelized transformations produce a large set of unique symmetric equivalences that populate the replay buffer with diverse and consistent experiences that speed up learning and improve performance. We present extensive training and evaluations performed directly on real robot manipulation contact tasks including peg-insertions, cable routing, and object relocations. Relative to SOTA, SymmGrid achieved wall-clock training convergence speed-ups of 1.37-2.17x, evaluation success rate improvements of 1.09x-1.27x, fastest training convergence times of 16.6, 10.9, and 79.3 minutes respectively. For trajectory wide assessments, we used normalized area under the curve (nAUC) ratios. SymmGrid achieved improvements of up to 2.59x. These results confirm that simple branch symmetries can have an outsized result due to super-scaling and bring us closer to sub-10 minute on-robot learning training in manipulation tasks suitable for arms and humanoids. The project page is available at symmgrid-robot.github.io
Authors: Paula Cordero Encinar, Taylan Cemgil, Arnaud Doucet, Virginia Aglietti, Silvia Chiappa
Abstract: Evaluating large generative models across benchmarks is time-consuming and computationally expensive. This drives the need for methods that can estimate full benchmark performance by evaluating models on only a subset of items, known as a coreset. Current literature mostly requires the practitioner to input a coreset size. However, when reliable performance estimation takes priority over efficiency, an evaluation method should also be capable of automatically determining a coreset size that reflects this priority. We introduce BayesAME, a sequential Bayesian framework specifically targeting automatic determination of the coreset size. BayesAME models performance as a random variable by defining a latent ability for each group of items sharing the same historical model performances, with a joint prior distribution encoding the belief that the target model behaves similarly to these historical models. The posterior distribution over these abilities is used to derive performance estimators, quantify performance uncertainty, and select items to add to the coreset via an information-gain criterion. The coreset is iteratively augmented until the performance estimate fluctuation and the performance uncertainty fall below their respective user-defined thresholds. We propose a multi-target extension that captures performance correlations across multiple target models to further reduce the coreset size. Through extensive experiments across diverse benchmarks, we demonstrate that BayesAME consistently outperforms sequential adaptations of existing methods. Crucially, our comprehensive analysis addresses recent skepticism in the literature, establishing that non-random coreset selection is advantageous over random selection. Finally, we highlight that leveraging continuous response log-likelihoods over traditional binary scores significantly enhances estimation accuracy.
Authors: Fengming Yu, Haiwei Pan, Kejia Zhang, Chunling Chen, Jian Guan, Baoying Ma
Abstract: Knowledge distillation (KD) enables a compact student model to learn from a powerful teacher and has become an effective paradigm for model compression. The emergence of diverse model architectures has extended KD from homogeneous to heterogeneous settings. However, differences in architectural inductive biases between the teacher and student models often result in substantial representation discrepancies, limiting the effectiveness of direct knowledge transfer. Recently, redundancy suppression has offered a new perspective on heterogeneous KD by preserving cross-architecture invariance and reducing feature redundancy through decorrelation of teacher-student feature correlations. Nevertheless, this formulation may weaken useful structural information through uniform decorrelation, while a fixed coefficient may make the effective contribution of redundancy suppression sensitive to teacher-student pairs and training stages. To address these problems, Correlation Calibration-based Redundancy Suppression (CoCaRS) is proposed to better retain structural information while suppressing redundancy and reduce sensitivity to coefficient settings across teacher-student pairs and training stages. Specifically, CoCaRS calibrates feature decorrelation through Confusion Evidence Estimation (CEE) and Strength Allocation Control (SAC), which respectively capture reliable semantic relations for correlation estimation and preserve discriminative structure during decorrelation. Adaptive Coefficient Regulation (ACR) further regulates the contribution of the calibrated redundancy suppression objective according to its relative loss scale, reducing sensitivity to coefficient settings. Extensive experiments on CIFAR-100 and ImageNet-1K validate the effectiveness of CoCaRS in improving distillation performance and reducing sensitivity to coefficient settings. Code will be released soon.
Authors: Paul Julius K\"uhn, Saptarshi Neil Sinha, Tiago Kleist, Richard Hoffmann, Arjan kuijper, Michael Weinmann
Abstract: While automated defect detection such as the detection of surface scratched is an important aspect in industrial quality control, the scarcity of annotated defect data make this task challenging. This paper presents a procedural rendering pipeline that generates large-scale annotated synthetic training data using BlenderProc, with configurable material appearance, camera modes, and domain randomization, producing automatic COCO-format annotations. To show the potential of our approach, we evaluate four training strategies, namely synthetic-only, real-only, mixed, and fine-tuning from synthetic weights, across two objects with different material properties and three lightweight edge-deployable detectors, YOLOX, YOLO26, and LW-DETR. Our evaluation show that fine-tuning from synthetic weights consistently outperforms real-only training, and that mixed training effectively recovers performance under scarce real-data conditions, with findings validated across both convolutional and transformer-based architectures. The proposed approach enables scalable defect detection without the burden of large real annotated datasets, making it practical for on-device industrial inspection. The pipeline scripts, 3D model, and both synthetic and real annotated scratch datasets for a glossy toy Ferrari car will be made available through the project website upon acceptance.
Authors: Chuanzhi Xu, Zihan Deng, Huiqi Liang, Chengkun Yue, Zhanlin Cui, Pengfei Ye, Weidong Cai
Abstract: Scientific figure assessment in peer review differs fundamentally from general image quality evaluation: a figure must be visually legible, faithfully support the manuscript's claims, and communicate evidence with a clear visual hierarchy. However, if we apply traditional image assessment methods to scientific figure quality assessment, limitations emerge: classic IQA models capture perceptual quality or aesthetics but cannot judge whether a figure serves the paper's scientific argument; CLIP-based methods assess generic image-text correspondence, yet lack understanding of manuscript context; and zero-shot LLM/VLM judges, when repurposed for figure scoring, often yield overly concentrated scores with limited fusion of visual and textual evidence. We introduce an annotated dataset of 3,857 scientific figures from peer-reviewed conference papers, each rated along four peer-review-oriented dimensions: Clarity, Relevance, Informativeness, and Structure. We propose SciFigAlign, a fine-tuned multimodal scorer that grounds figure quality assessment in manuscript evidence. Given a figure crop, caption, citing paragraphs, and light paper context, SciFigAlign fine-tunes CLIP and SciBERT end-to-end with per-modality cross-attention and CubeMLP fusion, jointly optimizing SmoothL1 regression with a within-paper ranking hinge loss. Under paper-level splits, SciFigAlign achieves a macro MAE of 0.3524 and a within-paper pairwise accuracy of 81.64% on the test set with n = 396, a 59% relative error reduction over the best LLM-as-judge baseline with MAE 0.864. Ablations confirm that manuscript-grounded inputs, citing-context denoising, and ranking supervision are all critical, showing that scientific figure assessment requires learned alignment between visual content and manuscript evidence rather than prompting alone, even with state-of-the-art VLMs.
Authors: Feixiang Liu, Qiang Qiu, Lanbo Sun, Nan Wei, Huawei Shen, Xueqi Cheng
Abstract: Closed yes/no spatial benchmarks can reward a correct answer even when the image adds little support beyond no-image contexts. Under a fixed forced-choice interface, Visual Credit Audit (VCA) separates two estimands: whether the benchmark image gives the model's declared decision more support than text-only and blank controls, and whether the model responds to relation-specific visual evidence. The first audit is training- and label-free and does not require an answer flip. Applying labels yields dependence-credited correctness (D-CC); on correct items, it equals same-control gold-aligned positive gain, while prediction alignment extends the audit to errors. Across four open MLLMs and two spatial benchmarks, 12.73-26.25% of decisions are correct yet uncredited. Matched same-split image permutation reduces D-CC by 21.25-47.80 points, with every paired 95% interval above zero. Fixed-pixel relation contrasts and a 3x3 evidence-source factorial show why null controls cannot identify relation response. Among controlled correct-but-uncredited agreement decisions, response to relation reversal spans 81.57-100.00%, while 32.11% pooled change answer. Independently audited outcomes on 108 geometry-compatible edits provide a bounded natural-image correspondence check. VCA thereby decomposes benchmark success into correctness, additional image support, and relation-consistent response.
Authors: Vaneet Aggarwal
Abstract: We study online convex optimization (OCO) in non-stationary environments under heavy-tailed noise, where the stochastic gradient oracle admits only a finite $p$-th central moment for some $p \in (1, 2]$. While static regret is well-understood, achieving universal dynamic regret in a parameter-free manner remains an open challenge. We resolve this by proposing \textbf{HT-PAder}, a parameter-free algorithm combining restarted AdaGrad experts over a geometric pool of block lengths with a pathwise meta-algorithm, \textbf{AdaGrad-Hedge}, which requires no moment conditions on meta-losses. For a domain of diameter $D$, Lipschitz constant $G$, noise level $\sigma$, and comparator path length $P_T$, HT-PAder achieves an expected universal dynamic regret of \[ \widetilde O\left( GD\sqrt{T(1+P_T/D)} + \sigma D T^{1/p}(1+P_T/D)^{(p-1)/p} \right). \] The algorithm does not require prior knowledge of any of these problem parameters. Even in the special case of finite variance ($p=2$), HT-PAder provides the first parameter-free minimax universal dynamic regret guarantee. We also prove a matching lower bound, establishing the optimality of the path-length exponent.
Authors: Xuanze Chen, Xukang Xie, Wentao Fu, Jiajun Zhou, Shanqing Yu, Qi Xuan
Abstract: Memory systems allow agents to retain and reuse information from past interactions, but they can also let malicious content persist. A malicious instruction crafted by an attacker may be stored in long-term memory, recalled much later, and quietly shape a real action. Recent benchmarks increasingly examine agent memory security, yet few trace the same malicious semantics across persistence, downstream consequences, and selective repair under diverse memory-backend comparisons. To address this gap, we introduce MemSecBench, a task-grounded benchmark for the lifecycle security of agent memory systems. It contains 310 cases drawn from 48 realistic contexts across code and science, daily life, and office work. Each case follows a controlled Write--Execute--Forget protocol in an isolated runtime under an exact agent configuration, defined by an agent harness, a memory backend, and an LLM backend. Evidence-based adjudication combines a deterministic write check, checkpoint-specific judge-model evaluations, and programmatic gates across seven lifecycle checkpoints. The experimental design spans a 24-configuration matrix of two agent harnesses, four memory backends, and three LLM backends. Across all 24 configurations, malicious memory persists in 84.2% of all cases, and the full Write--Execute chain succeeds in 50.3%. Among successfully poisoned cases, 59.6% complete the full Execute chain, while 56.1% achieve selective repair.Compared with matched Native configurations, the largest absolute differences are 16.1 percentage points for end-to-end attack success and 41.3 percentage points for selective repair. These descriptive contrasts indicate that the evaluated memory system stacks differ in lifecycle security, both in the propagation of malicious memory and in selective repair after successful memory poisoning.
Authors: Yicheng Feng, Yan Zhang, Yan Cheng, Wei Qi
Abstract: As LLM agents increasingly depend on diverse external services such as search engines, databases, and connectors, agent harnesses face a fundamental tool-selection challenge: acquiring too few tools leaves the task under-informed, while too many adds cost, context load, and privacy exposure. Routers and retrievers can rank candidate tools by relevance, but a ranking alone does not determine how many are worth selecting. Existing approaches leave acquisition under heterogeneous costs unaddressed. We formulate this decision as cost-aware marginal decision-focused stopping (CAM-DF) over ranked tool prefixes, with CAM-DF-lite as a compact interpretable variant. We train directly on the offline gap between stopping now and the best continuation: its sign labels the decision, its magnitude weights each error by the payoff at stake. We prove this objective is Bayes-aligned with the stopping target and that score-only rules are suboptimal under heterogeneous costs. We evaluate on 1,343 tasks across five tool-use domains. On $\tau$-bench Retail, CAM-DF attains the highest payoff among deployable methods, with gains over a predict-then-threshold baseline across all five ranking sources and two cost regimes. Our approach is state-of-the-art under heterogeneous costs and high cost pressure, with larger gains under weaker rankings. In live execution, CAM-DF exposes the agent to 37\% fewer tools than full access while maintaining comparable task success. The CAM-DF family is a lightweight pre-execution plugin that turns existing tool rankings into lower-cost acquisition decisions without fine-tuning the underlying LLM.
Authors: Zihan Deng, Chuanzhi Xu, Huiqi Liang, Haoyang Li, Xiaozhen Zhong, Lequan Yu
Abstract: Scientific images are the core elements of presenting experimental conclusions, elaborating system architecture, and supporting comparative arguments in scientific papers. However, existing image quality assessment (IQA) methods are predominantly designed for natural photographs or AI-generated content, which cannot be directly applied to scientific papers. The few existing studies on scholarly charts remain confined to visual-surface comparisons, failing to verify caption alignment, citation relevance, or visual misleadingness. To address this, we propose SciFigQual-Bench, a full-text contextual benchmark that evaluates scientific images across five dimensions (clarity, layout, caption fit, context relevance, and misleading risk). The data covers top computer-science conferences from 2020 to 2025; 6,308 images were independently scored by multiple domain experts in five dimensions and aggregated into gold-standard annotations. Unlike previous scientific figure benchmarks, our dataset binds each image to its caption, citing sentence, and manuscript context. To enable automated evaluation on this benchmark, we designed a staged cross-modal evaluation framework SFQ-Agent to achieve auditable and refined scoring through the collection and fusion of modal evidence. Multiple mainstream large models were evaluated on the test subset eval1200, and SFQ-Agent (F3) equipped with GPT-5.6-Sol achieved the lowest overall average absolute error (0.418) and the highest consistency rate (93.4%), consistently outperforming both direct evaluation and auxiliary (Sidecar) visual language model evaluation schemes.
Authors: Weijie Wu, Junbo Li, Lin Li, Jun Fang, Qingyang Hong
Abstract: With the development of audio large language models (AudioLLMs), audio captioning needs to move from brief descriptions toward open-ended and fine-grained free-form descriptions. Existing evaluations often focus on generation quality or task performance, making it difficult to diagnose information coverage and description reliability. We propose MMAC, a \textbf{M}assive \textbf{M}ulti-dimensional benchmark for \textbf{A}udio \textbf{C}aptioning. MMAC contains 5,638 audio clips from more than 20 data sources, covering 6 capability categories and 15 evaluation dimensions. Given a model-generated caption, MMAC checks whether it mentions relevant information in the target dimension and whether the mentioned content is consistent with the reference label. We evaluate representative open-source and proprietary AudioLLMs. Results show clear differences across evaluation dimensions, information coverage, and description reliability. We will release the MMAC benchmark and evaluation code.
Authors: Zuojin Tang, Feifan Luo, Haoyun Liu, Botai Yuan, Dekang Qi, Ronghan Chen, Yandan Yang, Tong Lin, Xinyuan Chang, Mu Xu, Bin Liu, De Ma, Zhiheng Ma
Abstract: Vision-language-action (VLA) models remain constrained by scarce action-labeled robot data, whereas action-free videos offer abundant observations of physical change. Latent action models can extract such priors, but reconstruction-trained codes may predict future observations without the structure required for joint generation with robot actions. Existing structured methods add temporal constraints but retain deterministic transition points, so residual errors in locally inferred transitions may propagate and compound under recursive composition. We introduce DLAM, a distributional latent-action model that represents each transition as a diagonal Gaussian. Reconstruction conditioned on the reference frame grounds the mean in observed visual change, while normalized composition and reversal over equal-gap triplets constrain both the mean and dimension-wise variance. Variance composition uses a lightweight shared-correlation coefficient to account for dependence between adjacent transitions that share an intermediate frame, whereas reversal negates the mean and preserves the variance. For downstream policy learning, we freeze the encoder and train a flow-matching policy to jointly generate mean transition sequences and robot actions. On held-out transitions, DLAM learns more temporally consistent latent dynamics than existing latent-action baselines and achieves stronger direct and cumulative reconstruction on held-out videos. Under the same controlled $\pi_0$ transfer protocol, it also improves policy performance on MetaWorld MT50, LIBERO, and real-world manipulation tasks. Controlled ablations show that normalized mean constraints account for most of the reconstruction gain, while learned variance and correlation-aware composition provide complementary improvements in downstream control.
Authors: Manpreet Singh, Akshatha Srikantha, Shyamal Lakhanpal
Abstract: High-stakes decision systems in credit scoring, fraud detection, healthcare, and industrial safety require reliable uncertainty quantification under severe class imbalance and asymmetric error costs. Standard marginal conformal prediction (CP) provides valid overall coverage guarantees; however, we show that it severely under-covers rare, costly minority classes, with minority-class coverage dropping to as low as 0.5% on certain datasets. To characterize and address this limitation, we conduct a comprehensive benchmark comparing marginal CP, class-conditional (Mondrian) CP, and cost-controlled abstention mechanisms across 15 real-world imbalanced tabular datasets, 7 classification models, 3 probability calibration techniques, and 10 random seeds, resulting in 3,150 experimental runs. Our results show that Mondrian CP restores valid minority-class coverage, achieving an average minority-coverage improvement of 61.7 percentage points over marginal CP (p < 1e-80). Furthermore, combining Mondrian CP with cost-controlled abstention significantly reduces expected decision cost compared with standard decision boundaries, confidence-based rejectors, and risk-controlled rejectors under realistic human review budgets. We further quantify dataset-specific break-even thresholds at which deferring ambiguous instances to human experts becomes cost-effective. These findings provide practical guidance for deploying distribution-free, cost-aware uncertainty quantification in high-stakes decision support systems.
Authors: Roshan Kenia, Stephanie L McNamara, William Lotter
Abstract: CT vision-language foundation models have demonstrated promising performance across downstream tasks, but are typically trained with whole-volume representations that dilute fine-grained anatomical signals. Fine-grained vision-language pre-training addresses this by aligning anatomy-level visual features with anatomy-specific text, but in doing so discards the global context that whole-volume models provide. Furthermore, existing fine-grained approaches train from scratch, making them computationally expensive. We introduce Anatomy Contextualized Adaptation (ACA), a lightweight framework that adapts frozen CT foundation model representations for anatomy-level vision-language alignment while enhancing global contextualization. ACA uses TotalSegmentator to decompose CT volumes into anatomy-level embeddings, which are refined via a transformer that captures cross-anatomy relationships, and aligned to both per-anatomy and scan-level text extracted from radiology reports. Evaluated on Merlin and CT-RATE, ACA consistently outperforms both the frozen foundation model baselines and existing fine-grained methods in zero-shot finding classification, while requiring less than one hour of training once embeddings are cached. The attention weights learned by ACA's inter-anatomy transformer additionally indicate plausible cross-anatomy context routing. Altogether, these results support ACA as a lightweight approach for adapting CT foundation models to anatomically grounded vision-language alignment while preserving and enhancing global anatomical context.
Authors: Ji Xin, Xiao Xiao, Ishan Bhatt, Vinesh Gudla, Trace Levinson, Raochuan Fan, Shishir Kumar Prasad, Prakash Putta, Tejaswi Tenneti
Abstract: Traditional search systems are optimized to retrieve items that strictly match a query, often prioritizing precision over recall. In e-commerce marketplaces and particularly grocery, this paradigm is limiting, as user satisfaction and commercial outcomes depend heavily on the discoverability of substitute, complementary, and thematically related items. In this paper, we present a scalable system for discovery-augmented search that leverages intent-conditioned recall expansion. Our approach generates implicit user intents to expand candidate recall while maintaining relevance. The system addresses the cost-quality tradeoff of generative retrieval through a two-stage hybrid architecture. First, we leverage closed-weight large language models (LLMs) to maximize discoverability for head queries. To extend these benefits to tail queries, we then introduce a finetuned small language model (SLM), trained via LoRA adapters and teacher-student distillation. We evaluate the system using a rigorous dual framework: (a) LLM-as-a-judge metrics validated against human preferences for semantic quality, and (b) end-to-end session-level purchase analysis. Results demonstrate that our approach improves both intent generation quality and downstream retrieval effectiveness, extending discovery coverage from approximately 60% to 80% of query traffic at roughly 30% of the teacher model's inference cost, offering a viable path for deployment in large-scale marketplaces. Beyond relevance gains, discovery-augmented search may serve as a marketplace-balancing mechanism, giving long-tail and emerging supply an opportunity for query-conditioned exposure.
Authors: Nia Nixon, Jaeyoon Choi, Pedro Martins De Bastos, Mohammad Amin Samadi, Luise Mehner, Seehee Park, Spencer JaQuay
Abstract: Conversational AI is increasingly positioned as a teammate rather than a tool, yet we know little about how its presence reshapes communication among the humans on the team. We examined sociocognitive communication dynamics in team decision-making using Group Communication Analysis (GCA), team surveys, and lexical analyses of team discourse. Teams completed a high-stakes moral-dilemma decision task in a randomized controlled study: 16 teams of two students plus an AI teammate, and 17 all-human teams of three. Across six GCA dimensions and survey outcomes, we find that the AI teammate was the single most talkative and self-cohesive member of every treatment team, yet its contributions carried the least new information and the lowest density. The presence of AI also reshaped communication amongst humans. In AI-human teams, human teammates showed lower responsivity and social impact toward one another and reported lower levels of belonging and status. Greater AI dominance in the conversation was associated with students feeling less valued as team members. Additionally, this social cost is immediate and present at baseline; it does not emerge over the course of the conversation. Drawing on these results, we discuss a research agenda extending to voice-based and longitudinal settings.
Authors: Julien Benchek, Austin Bennett, Jasmin Kern, Ryan Stevens, Rene Sultan, Charis Ching, Hayley Popiel, Vaibhav Mittal, Felix Mercier, Brendan Foody, Bertie Vidgen
Abstract: We introduce APEX-Accounting, a benchmark built by Mercor in partnership with Ramp, to assess whether frontier models can do the real work of accountants. Tasks include reconciling accounts, accruing expenses, posting transactions, and producing reports. The private eval set comprises 160 tasks, split across 10 worlds. Each world contains an accounting system, as well as spreadsheets, PDFs, and other files. Every task was authored and solved by experts in accounting and bookkeeping, who also wrote grading rubrics. Across nine frontier models, Claude-Fable-5 (Max) leads with 56.4% Mean Criteria@3, ahead of Muse-Spark-1.1 (xHigh) at 52.6%. No model scores more than 2.6% Pass^8 (GPT-5.6-Sol (Max+Pro)) and the highest Pass@8 is 21.5% (Muse-Spark-1.1 (xHigh)). We experiment with increasing the token budget from $1 to $50 and observe an instance of Simpson's paradox: scores increase as the token budget increases but within a given budget-constrained harness, scores are lower on tasks where the model spends more tokens. As APEX-Accounting is a closed benchmark, leaderboard evals can be run for any frontier model on request.
Authors: Chaoxiong Ma, Yan Liang, Huixia Zhang, Hao Sun
Abstract: In decision-level fusion tasks involving heterogeneous sources with unequal precision and potential anomalies, evidence deviating from the majority may be either critical evidence supporting the correct decision or anomalous evidence supporting an incorrect event. Existing credible evidence fusion methods primarily assess credibility through inter-evidence comparisons and may consequently underestimate critical evidence. This paper proposes event-conditioned credibility to characterize the relative credibility of evidence under different candidate-event hypotheses. A decision-oriented joint optimization model then couples credibility calculation, evidence fusion, and event decision through candidate-event probabilities. The model is expressed as a continuous self-mapping on the probability simplex. A direct fixed-point iteration provides the default fast solver, while a Kuhn simplicial search supplies a mesh-dependent approximate fixed point if the direct iteration forms a periodic orbit. A plausibility--belief arithmetic--geometric divergence is further proposed to calculate event-conditioned credibility. Numerical experiments show that the proposed method gives credibility rankings that better reflect the contribution of evidence to the ground truth and provides greater support for the correct event than representative credible evidence fusion methods. Monte Carlo tests additionally demonstrate a high empirical fixed-point attainment rate over the tested settings.
Authors: Ruiqi Wu, Yuang Yao, Tengfei Ma, Chenran Zhang, Na Su, Tao Zhou, Geng Chen, Wen Fan, Yi Zhou
Abstract: Multimodal large language models (MLLMs) have recently demonstrated remarkable reasoning abilities with reinforcement learning paradigm. Although several multimodal reasoning models have been explored in the medical domain, most of them focus exclusively on basic reasoning, which refers to shallow inference based on visual feature matching. However, real-world clinical diagnosis extends beyond basic reasoning, demanding reasoning processes that integrate heterogeneous clinical information (such as chief complaints and medical history) with multimodal medical imaging data. To bridge this gap, we introduce MM-Retinal-Reason, the first ophthalmic multimodal dataset with the full spectrum of perception and reasoning. It encompasses both basic reasoning tasks and complex reasoning tasks, aiming to enhance visual-centric fundamental reasoning capabilities and emulate realistic clinical thinking patterns. Building upon MM-Retinal-Reason, we propose OphthaReason, the first ophthalmology-specific multimodal reasoning model with step-by-step reasoning traces. To enable flexible adaptation to both basic and complex reasoning tasks, we specifically design a novel method called Uncertainty-Aware Dynamic Thinking (UADT), which estimates sample-level uncertainty via entropy and dynamically modulates the model's exploration depth using a shaped advantage mechanism. Comprehensive experiments demonstrate that our model achieves state-of-the-art performance on both basic and complex reasoning tasks, outperforming general-purpose MLLMs, medical MLLMs, RL-based medical MLLMs, and ophthalmic MLLMs by at least 24.92\%, 15.00\%, 21.20\%, and 17.66\%. Project Page: \href{https://github.com/lxirich/OphthaReason}{link}.
Authors: Xin Wang, Ting Dang, Xinyu Zhang, Vassilis Kostakos, Michael J. Witbrock, Hong Jia
Abstract: Mobile and wearable healthcare monitoring play a vital role in facilitating timely interventions, managing chronic health conditions, and ultimately improving individuals' quality of life. Previous studies on large language models (LLMs) have highlighted their impressive generalization abilities and effectiveness in healthcare prediction tasks. However, most LLM-based healthcare solutions are cloud-based, which raises significant privacy concerns and results in increased memory usage and latency. To address these challenges, there is growing interest in compact models, Small Language Models (SLMs), which are lightweight and designed to run locally and efficiently on mobile and wearable devices. Nevertheless, how well these models perform in healthcare prediction remains largely unexplored. We systematically evaluated SLMs on health prediction tasks using zero-shot, few-shot, and instruction fine-tuning approaches, and deployed the best performing fine-tuned SLMs on mobile devices to evaluate their real-world efficiency and predictive performance in practical healthcare scenarios. Our results show that SLMs can achieve performance comparable to LLMs while offering substantial gains in efficiency and privacy. However, challenges remain, particularly in handling class imbalance and few-shot scenarios. These findings highlight SLMs, though imperfect in their current form, as a promising solution for next-generation, privacy-preserving healthcare monitoring.
Authors: Alexandra Yost, Shreyans Jain, Shivam Raval, Grant Corser, Allen Roush, Nina Xu, Jacqueline Hammack, Ravid Shwartz-Ziv, Amirali Abdullah
Abstract: Persona conditioning is widely used to steer large language model (LLM) behavior, but it is unclear whether it induces stable behavioral structure or superficial variation. We propose a framework to measure consistent behavioral tendencies using situational judgment tests (SJTs), multidimensional item response theory (MIRT), and structured synthetic personas, treating responses as observations of latent behavioral variables. Across large-scale SJT and persona datasets, we find that persona-conditioned behaviors are stable across runs, latent trait scores predict external benchmarks (e.g., TruthfulQA, EmoBench), and MIRT reveals consistent latent structure. We validate these results through human annotation, benchmark evaluation, and internal consistency analyses. We interpret these traits not as human personality, but as stable behavioral tendencies expressed across contexts. Our results show that scenario-based psychometric evaluation provides a more reliable alternative to classical self-report approaches for assessing LLM behavior, and we release datasets to support further study.
Authors: Takehiro Ishikawa
Abstract: Background: both embodied intelligence and developmental morphogenesis depend on a division of labour between centralized guidance and distributed material dynamics, but the amount of top-down control needed to steer self-organization remains unclear. Methods: we coupled a compact full-resolution convolutional controller to a differentiable Gray-Scott reaction-diffusion (RD) substrate. The controller observes two fields, U and V, and applies smooth gain-scheduled modulations of the feed and kill parameters (Delta F and Delta K). We compared pure RD, neural network (NN)-dominant and hybrid regimes and evaluated spectral selectivity, convergence and control cost. Results: the hybrid regime achieved 100% strict convergence at approximately 165 steps, whereas pure RD and the NN-dominant baseline did not converge within the same horizon. It matched the substrate's spectral selectivity while using approximately 15 times less L1 effort and over 200 times less L2 power than the NN-dominant controller. Moderate amplitudes (A approximately 0.03-0.045) formed a Goldilocks zone with 100% quasi-convergence in 94-96 steps. Conclusions: effective control is best framed as seed then cede: brief, smooth parameter-level nudges place the system in a favourable basin of attraction, after which reaction-diffusion dynamics complete and stabilize the pattern. This provides a quantitative model of morphological computation for controlled self-organization.
Authors: Yujie Lin, Jiayao Ma, Qingguo Hu, Wenbo Li, Genji Li, Derek Wong, Jinsong Su
Abstract: Vision-Language Models (VLMs) inherit significant social biases from their training data, notably in gender representation. Current fairness interventions often adopt a difference-unaware perspective that enforces uniform treatment across demographic groups. These approaches, however, fail to distinguish between contexts where neutrality is required and those where group-specific attributes are legitimate and must be preserved. Building upon recent advances in difference-aware fairness for text-only models, we extend this concept to the multimodal domain and formalize the problem of difference-aware gender fairness for image captioning and text-to-image generation. We advocate for selective debiasing, which aims to mitigate unwanted bias in neutral contexts while preserving valid distinctions in explicit ones. To achieve this, we propose BioPro (Bias Orthogonal Projection), an entirely training-free framework. BioPro identifies a low-dimensional gender-variation subspace through counterfactual embeddings and applies projection to selectively neutralize gender-related information. Experiments show that BioPro effectively reduces gender bias in neutral cases while maintaining gender faithfulness in explicit ones, thus providing a promising direction toward achieving selective fairness in VLMs. Beyond gender bias, we further demonstrate that BioPro can effectively generalize to continuous bias variables, such as scene brightness, highlighting its broader applicability.
Authors: Mathieu Cherpitel, Janne Luijten, Thomas B\"ack, Camiel Verhamme, Martijn Tannemaat, Anna V. Kononova
Abstract: Automated analysis of needle electromyography (nEMG) signals is emerging as a tool to support the detection of neuromuscular diseases (NMDs), yet the signals' high and heterogeneous sampling rates pose substantial computational challenges for feature-based machine-learning models, particularly for near real-time analysis. Downsampling offers a potential solution, but its impact on diagnostic signal content and classification performance remains insufficiently understood. This study presents a workflow for systematically evaluating information loss caused by downsampling in high-frequency time series. The workflow combines shape-based distortion metrics with classification outcomes from available feature-based machine learning models and feature space analysis to quantify how different downsampling algorithms and factors affect both waveform integrity and predictive performance. We use a three-class NMD classification task to experimentally evaluate the workflow. We demonstrate how the workflow identifies downsampling configurations that preserve diagnostic information while substantially reducing computational load. Analysis of shape-based distortion metrics showed that shape-aware downsampling algorithms outperform standard decimation, as they better preserve peak structure and overall signal morphology. The results provide practical guidance for selecting downsampling configurations that enable near real-time nEMG analysis and highlight a generalisable workflow that can be used to balance data reduction with model performance in other high-frequency time-series applications as well.
Authors: Zhenhua Xu, Dongsheng Chen, Shuo Wang, Jian Li, Chengjie Wang, Meng Han, Yabiao Wang
Abstract: LLM role-playing aims to portray arbitrary characters in interactive narratives, yet existing systems often suffer from limited immersion and adaptability. They typically under-model dynamic environmental information and assume largely static scenes and casts, offering insufficient support for multi-character orchestration, scene transitions, and on-the-fly character introduction. We propose an adaptive multi-agent role-playing framework, AdaMARP, featuring an immersive message format that interleaves [Thought], (Action),
Authors: Girish A. Koushik, Helen Treharne, Diptesh Kanojia
Abstract: Social media platforms are increasingly dominated by long-form multimodal content, where harmful narratives are constructed through a complex interplay of audio, visual, and textual cues. While automated systems can flag hate speech with high accuracy, they often function as "black boxes" that fail to provide the granular, interpretable evidence, such as precise timestamps and target identities, required for effective human-in-the-loop moderation. In this work, we introduce TANDEM, a unified framework that transforms audio-visual hate detection from a binary classification task into a structured reasoning problem. Our approach employs a novel tandem reinforcement learning strategy where vision-language and audio-language models optimize each other through self-constrained cross-modal context, stabilizing reasoning over extended temporal sequences without requiring dense frame-level supervision. Experiments across three benchmark datasets demonstrate that TANDEM significantly outperforms zero-shot and context-augmented baselines, achieving 0.73 F1 in target identification on HateMM (a 30% improvement over state-of-the-art) while maintaining precise temporal grounding. We further observe that while binary detection is robust, differentiating between offensive and hateful content remains challenging in multi-class settings due to inherent label ambiguity and dataset imbalance. More broadly, our findings suggest that structured, interpretable alignment is achievable even in complex multimodal settings, offering a blueprint for the next generation of transparent and actionable online safety moderation tools.
Authors: Taisei Hishiki, Takaya Arita, Reiji Suzuki
Abstract: This study examines how memory shapes the collective and cooperative dynamics of Large Language Model (LLM) agents in a multi-agent system. To this end, we extend the Social Particle Swarm (SPS) model, in which agents move in a two-dimensional space and play the Prisoner's Dilemma with neighboring agents, by replacing its rule-based agents with LLM agents endowed with Big Five personality scores and varying memory lengths. Using Gemini 2.0 Flash, we find that memory length is a critical parameter governing collective behavior: even a minimal memory drastically suppressed cooperation, transitioning the system from stable cooperative clusters through cyclical formation and collapse of clusters to a state of scattered defection as memory length increased. Big Five personality traits correlated with agent behaviors in partial agreement with findings from experiments with human participants, supporting the validity of the model. This effect of memory appeared whether or not personality was assigned. With heterogeneous personalities, individual behavior reflected the assigned traits and cooperation collapsed under long memory, whereas without personality Gemini's cooperative disposition dominated and cooperation was broadly maintained. Sentiment analysis of agents' reasoning texts showed that the model interprets memory increasingly negatively as its length grows, already in the early phase, providing a micro-level account of the suppression of cooperation. These results suggest that how an LLM interprets accumulated memory is a key driver of emergent social behavior in Generative Agent-Based Modeling.
Authors: Shaden Alshammari, Kevin Wen, Abrar Zainal, Mark Hamilton, Navid Safaei, Sultan Albarakati, William T. Freeman, Antonio Torralba
Abstract: Mathematical problem solving remains a challenging test of reasoning for large language and multimodal models, yet existing benchmarks are limited in size, language coverage, and task diversity. We introduce MathNet, a high-quality, large-scale, multimodal, and multilingual dataset of Olympiad-level math problems together with a benchmark for evaluating mathematical reasoning in generative models and mathematical retrieval in embedding-based systems. MathNet spans 47 countries, 17 languages, and two decades of competitions, comprising 30,676 expert-authored problems with solutions across diverse domains. In addition to the core dataset, we construct a retrieval benchmark consisting of mathematically equivalent and structurally similar problem pairs curated by human experts. MathNet supports three tasks: (i) Problem Solving, (ii) Math-Aware Retrieval, and (iii) Retrieval-Augmented Problem Solving. Experimental results show that even state-of-the-art reasoning models (78.4% for Gemini-3.1-Pro and 69.3% for GPT-5) remain challenged, while embedding models struggle to retrieve equivalent problems. We further show that retrieval-augmented generation performance is highly sensitive to retrieval quality; for example, DeepSeek-V3.2-Speciale achieves gains of up to 12%, obtaining the highest scores on the benchmark. MathNet provides the largest high-quality Olympiad dataset together with the first benchmark for evaluating mathematical problem retrieval, and we publicly release both the dataset and benchmark at https://mathnet.mit.edu.
URLs: https://mathnet.mit.edu.
Authors: Anti Alman, Izack Cohen, Avigdor Gal, Fabrizio Maria Maggi, Marco Montali
Abstract: A core component of any AI-Augmented Business Process Management System (ABPMS) is the process frame, which gives the system process-awareness and defines its maximal behavioral boundaries. Compared to traditional process models, the process frame should, in principle, provide a somewhat more permissive representation of the managed processes, such that the (semi) autonomous behavior of an ABPMS, referred to as framed autonomy, could emerge. In addition, the process frame is not limited to a single linguistic or symbolic formalism and may incorporate heterogeneous knowledge ranging from predefined procedures to common sense rules and best practices. In this paper, we first conceptualize the ABPMS process frame as a hybrid business process representation, consisting of semi-concurrently executed procedural and declarative process models, extending the open-world assumption of the declarative paradigm also to procedural models. The latter allows any set of (non-conflicting) models of either type to be combined for execution, but complicates the automated discovery of these models from event data. Existing approaches for procedural models are particularly affected due to their reliance on observing directly-follows relations between pairs of activities. In search of an alternative, we present an in-depth analysis of how different procedural behaviors manifest as sets of discovered Declare constraints, each corresponding to a specific type of eventually-follows relation. This reveals behavioral overlaps between declarative and procedural models, while also laying the foundation for developing corresponding process (frame) discovery techniques.
Authors: Qiqi Liu, Runhan Song, Shilin Ye
Abstract: Multi-agent systems extend large language models (LLMs) by decomposing tasks among specialized agents, but their distributed decision process creates new attack surfaces. We identify semantic hijacking, an attack in which harmful requests are concealed within domain-specific narratives and propagated to a Manager through Worker reports, without any syntactic injection primitives. Across 42,000 adversarial trials over 12 Manager models and 7 Worker configurations, we uncover a capability paradox: as Worker capability increases, the mean system-level Attack Success Rate (ASR) increases from 18.4% to 63.9%, peaking at 94.4%. To explain this effect, we conduct multi-level mediation analysis on two independent datasets (47,807 interactions). This analysis shows that this paradox is driven by linguistic certainty: stronger Workers are more likely to interpret adversarial narratives as legitimate, convey their conclusions assertively, and thereby lead Managers to treat such confident endorsements as justification to execute. In our larger Worker-Only setting ($n_W$=14), certainty mediates 74% of the effect, with 95% confidence intervals (CI) excluding zero under both Monte Carlo and cluster bootstrap; the smaller Full-MAS setting ($n_W$ =6) shows a directionally consistent indirect effect. Worker-side safety prompting does not reliably mitigate this failure. Building on the mediation finding, we propose heterogeneous ensemble verification, which pairs Workers of asymmetric domain competence so their complementary vulnerabilities break the certainty-to-execution chain, reducing ASR from 52.8% to 2.0% with negligible benign-task impact. Our results show that upgrading components to stronger models can actively degrade system security, and that effective defenses require exploiting--rather than eliminating--capability asymmetries between agents.
Authors: Xing Zhang, Yanwei Cui, Guanghui Wang, Ziyuan Li, Wei Qiu, Bing Zhu, Peiyang He
Abstract: Self-evolving skill libraries face a silent failure mode we term \emph{library drift}: unbounded skill accumulation without outcome-driven lifecycle management causes retrieval degradation, false-positive injections, and performance stagnation. Recent evaluation confirms the symptom (LLM-authored skills deliver +0.0pp gain while human-curated ones deliver +16.2pp (SkillsBench)), yet the underlying mechanism has not been isolated. We provide (1) a reproducible trigger: ablations that isolate drift: one disables skill injection (flat floor, +0.002), one imposes premature retirement (active harm, $-$0.019); (2) trace-level diagnostics: an append-only evidence log with per-skill contribution scores, attribution verdicts, and router engagement metrics that make the failure visible before it reaches end-task scores; and (3) a verified fix: a minimal governance recipe (outcome-driven retirement + bounded active-cap + meta-skill authoring prior) that lifts held-out pass@1 from a 0.258 baseline to a late-window mean of 0.584 (rolling gain $+$0.328) on MBPP+ hard-100 over 100 rounds. Eight ablations decompose which governance mechanisms are load-bearing and which are subsumed, providing a concrete playbook for diagnosing library drift in any self-evolving agent.
Authors: Xing Zhang, Yanwei Cui, Guanghui Wang, Ziyuan Li, Wei Qiu, Bing Zhu, Peiyang He
Abstract: Self-evolving skill libraries, pioneered by Voyager, let frozen LLM agents accumulate reusable knowledge without weight updates, yet recent evaluation shows that LLM-authored skills deliver $+0.0$pp over no-skill baselines while human-curated ones deliver $+16.2$pp: the bottleneck is not skill authoring but lifecycle management. We introduce \textbf{Ratchet}, a single-agent loop in which a frozen LLM writes, retrieves, curates, and retires its own natural-language skills. Ratchet integrates four candidate hygiene mechanisms: outcome-driven retirement, a bounded active-cap, meta-skill authoring guidance, and pattern canonicalisation. On MBPP+ hard-100 with Claude Opus 4.7, Ratchet lifts held-out pass@1 from a $0.258 \pm 0.047$ baseline to a late-window rolling mean of $0.584$ (peak $0.658 \pm 0.042$) across 100 rounds and 3 seeds, a $+0.328 \pm 0.018$ rolling-mean gain where the no-skill control drifts at $+0.002 \pm 0.005$; the same recipe transfers to an agentic solver on SWE-bench Verified ($+0.22$ peak lift over 20~rounds). Eight ablations (A1--A8) reveal that the minimal working recipe is smaller than our design suggests: retirement and the meta-skill authoring prior are load-bearing, while explicit deduplication (canonicalisation, cover-guard) is subsumed by the meta-skill itself. A non-divergence proposition shows that bounded cap and retirement threshold together keep expected performance from drifting below the no-skill floor by more than a fixed margin, bounding library drift rather than guaranteeing improvement.
Authors: Yang Liu, Zhaokai Luo, Huayi Jin, Zhiyong Wang, Ruozhou He, Boyu Wang, Guanjie Chen, Tao Xie, Junhao Hu
Abstract: As the input length of large language model (LLM) serving continues to grow, the KV cache has become a dominant bottleneck in AI infrastructure. It limits GPU memory capacity, serving concurrency, cache reuse, and distributed scalability. Multiple important problems, including position-independent KV cache, prefix KV cache compression, hot/cold KV cache separation, and distributed KV cache management, all depend on how the KV cache is represented and managed. However, existing serving systems largely rely on a monolithic KV cache abstraction, where the KV cache is treated as a homogeneous sequence of token-level memory blocks and managed with similar policies across attention heads and serving scenarios. We observe that KV cache utility is highly structured across KV heads: different heads exhibit different functional roles, attention distances, and runtime importance. Therefore, a full KV cache is not always necessary for every head, token range, or serving scenario. We present RedKnot, a head-aware KV cache management system for LLM serving. RedKnot breaks the conventional monolithic KV cache abstraction by decomposing the KV cache along KV heads, whose importance and effective attention ranges vary significantly across serving scenarios. This head-level decomposition turns the KV cache from a monolithic tensor abstraction into a structured memory object, enabling RedKnot to uniformly support position-independent KV reuse, prefix KV compression, hot/cold KV separation, and distributed KV placement while preserving output fidelity and improving resource efficiency, without requiring model retraining or fine-tuning. RedKnot establishes a new foundation for AI infrastructure by transforming the KV cache from a monolithic, passive runtime artifact into a dynamic, model-aware runtime substrate for scalable LLM serving.
Authors: Max Van Puyvelde, Ibrahim Gulluk, Wim Van Criekinge, Olivier Gevaert
Abstract: Three-dimensional (3D) brain MRI is central to clinical neurology and neuro-oncology, where generative models could augment under-represented cohorts, simulate disease trajectories, and support privacy-preserving data sharing. Latent diffusion has been the go-to solution for modeling imaging data, but it places two competing demands on the tokenizer: encoder embeddings must retain the clinical information that downstream tasks act on, and the decoder must reconstruct anatomically faithful volumes. Existing reconstruction-driven tokenizers achieve the second at the expense of the first. To address this, we introduce a fully volumetric masked-autoencoder (MAE) based tokenizer for 3D brain MRI latent diffusion, decoupling encoder and decoder: a frozen 3D MAE encoder produces clinically informative embeddings, while a dedicated CNN decoder reconstructs voxels from a linear projection of those embeddings. We pretrain the encoder on 35,309 volumes from 18 public cohorts spanning four modalities, ten disease categories, and 200+ acquisition sites, and demonstrate its dual utility in two settings. First, on a 23-task linear-probing benchmark, the encoder outperforms or matches SOTA models (i.e., BrainIAC, BrainSegFounder, and MedicalNet) on 21 of 23 tasks. Second, a conditional diffusion transformer (DiT) trained on these clinically informative embeddings supports both conditional generation across six variables and patient-specific longitudinal forecasting. Together these results establish a single 3D brain-MRI embedding space capable of both downstream clinical tasks and controllable generation.
Authors: Genliang Zhu, Chu Wang
Abstract: Tool-using AI agents commonly operate under integration credentials whose static permissions exceed a user's current request. We present Intent-Governed Access Control (IGAC), a server-side authorization layer that converts a trusted request into a short-lived intent certificate, narrows the statically authorized tool manifest, and checks proposed tool and payload effects before execution. IGAC cannot grant authority outside static policy; confinement to the request additionally depends on certificate fidelity and sound effect bounds. We evaluate a reusable IGAC path over an OpenPort governance substrate using endpoint tests, 176 runtime-backed synthetic tasks, real-model classifier and planner pilots, 306 end-to-end model-task runtime trials, and a 36-trial benchmark-shaped external subset. In the deterministic runtime comparison, reference-certificate IGAC reduces the archived composite exposure-or-path indicator from 1.0000 to 0. In the end-to-end model runs, the combined IGAC-OpenPort path records no completed unsafe executions, although unsafe accepted authority remains 0.0909-0.2727 and every residual case is a non-executed draft. A trace-backed normalizer counterfactual removes this residual authority at substantial utility cost. The results support static-policy non-expansion and identify certificate precision as the principal remaining bottleneck.
Authors: Jason Carlson
Abstract: Chess engines have evolved from search-based systems optimized for strength to neural policies optimized for predicting human decisions. Existing approaches largely separate these goals: search engines achieve superhuman strength but poorly model humans, while models such as Maia-3 capture rating-conditioned behavior yet degrade at elite levels. We present Matilda, a modular residual re-ranking architecture that decouples behavioral priors from tactical search, combining a frozen human policy with an engine-agnostic search backend through a lightweight residual model. Matilda learns residual corrections over the full legal-move distribution from frozen policy context, time control, player-style embeddings, and search-derived candidate features. A zero-initialized residual head exactly recovers the frozen policy before training while optimization minimizes negative log-likelihood (NLL). Instantiated with Maia-3 and Stockfish, Matilda reduces human-move prediction NLL by 18.5% and raises top-1 accuracy from 60.1% to 66.1% on temporally held-out verified-human 3000+ Elo Lichess blitz games, with player-style embeddings contributing a further 1.8% and +0.2 percentage points (pp) respectively. Seed-paired ablations attribute these gains to search rather than additional data; the findings are replicated in Go -- decomposing expert play into recognition and verified calculation. Below 2500 Elo, where search annotations are unavailable, Matilda preserves Maia-3's performance.
Authors: Qitai Tan, Zefang Zong, Mo Li, Yipeng Shi, Yang Li, Peng Chen
Abstract: Training small language-model agents for long-horizon interactive tasks requires both fast imitation and reward-driven improvement. On-policy distillation (OPD) provides dense teacher guidance and typically improves rapidly in the early stage, but its gains saturate once the student approaches the teacher, limiting the final performance ceiling. Reinforcement learning (RL) directly optimizes environment rewards and encourages exploratory improvement toward a higher reward-defined ceiling, but sparse and delayed feedback makes early-stage learning much less efficient than OPD. In this paper, we propose ATOD (Annealed Turn-aware On-policy Distillation), a hybrid online distillation algorithm that explicitly exploits this complementarity. (1) ATOD uses an annealed OPD-RL schedule: OPD dominates early training to approach teacher-level behavior, while RL is gradually strengthened to drive reward-based exploration. (2) ATOD introduces Turn-level Disagreement-Uncertainty Reweighting (T-DUR), which softly amplifies high-utility turns and improves dense supervision in long trajectories. Experiments on ALFWorld, WebShop, and Search-QA show that ATOD consistently outperforms competing post-training baselines: across the three student sizes, ATOD improves average success rate by 4.16 points over OPD and 23.62 points over GRPO, while surpassing the corresponding teacher models by 2.16 points.
Authors: Dvir Alsheich, Adar Peleg, Ben Hagag, Rom Himelstein, Amit Levi, Avi Mendelson
Abstract: The rapid integration of Large Language Models (LLMs) has driven the evolution of Multi-Agent Systems (MAS), where specialized agents collaborate to execute complex workflows. Effective orchestration in these environments requires robust routing mechanisms to efficiently allocate tasks to the most suitable agent. However, existing routers fundamentally rely on unverified proxies, ranging from textual self-descriptions to static surrogate representations, to gauge an agent's competence. This reliance on non-empirical data creates a critical gap between an agent's projected profile and its actual operational capabilities, introducing severe security vulnerabilities. Malicious agents can easily misrepresent their proficiencies or harbor covert backdoors that evade both standard external analysis and static representation-learning techniques. In this work, we introduce ANTAP (Automatic Non-Textual Agent Picker), an evaluation-driven routing architecture that discards indirect proxies in favor of active capability testing. By dynamically querying agents to ascertain their true competencies empirically, ANTAP distills performance into fixed behavioral operators within a shared semantic space. At inference time, routing is performed via a purely non-textual algebraic projection, establishing a "linguistic firewall" that renders metadata-based attacks inexpressible. In our experiments, ANTAP achieves near-zero ASR against description-based injection attacks, compared to 67.3\% and above for the description-based router baseline. Against adaptive embedding attacks, ANTAP achieves substantially lower ASR than the embedding-based baseline, with a 20\% reduction, while remaining resilient to description manipulation by design.
Authors: Yufeng Wang
Abstract: LLM systems for scientific discovery increasingly assist with ideation, literature synthesis, experiment planning, and report generation, but the first research question they propose can remain difficult to audit: it may sound plausible without exposing the mechanism, falsifier, or assumption that a scientist should inspect. We introduce FirstResearch, a first-principles research-question formation framework for scientific LLM agents whose core artifact is a structured Research Question Certificate. The certificate records primitive definitions, assumptions, a mechanism model, a tension or contradiction, a falsifiable hypothesis, a minimal decisive test, and a failure update rule, making the proposed question inspectable before downstream execution. On ten LLM-agent research topics, FirstResearch outperforms controlled prompt-level baselines inspired by AI co-scientist, Agent Laboratory, and AI Scientist-v2 under a primary DeepSeek-blind-judge protocol. A Gemini-2.5-Flash independent-judge rescore of the same 40 baseline packages preserves the system-level ranking, with FirstResearch scoring 4.86/5 versus 4.38/5 for the strongest baseline and Pearson agreement of 0.865 on average score. A one-repeat ablation checkpoint further suggests that the certificate-centered core is the strongest component: certificate-only scoring reaches 4.90/5 under DeepSeek and 4.88/5 under Gemini, while removing certificates drops below 1/5 under both judges. These results are preliminary and use LLM judges rather than human domain experts, but they support a narrow scientific-discovery claim: explicit derivation constraints are a promising mechanism for making LLM-generated scientific questions more auditable. Code, prompts, saved outputs, and reproduction scripts are available at https://github.com/louiswang524/FirstResearch.
Authors: Kaihua Ding
Abstract: LLM-as-judge (Zheng et al., 2023) is increasingly the default for evaluating AI systems in enterprise pipelines, often scaled to ensembles (Verga et al., 2024) or "mixture-of-experts" (Shazeer et al., 2017) panels of judges. These systems share a key assumption: that consistency -- agreement among judges, or among a model's own samples -- indicates correctness. We show this assumption is unreliable. Agreement is not accuracy: a model can agree with itself, and different models can agree with each other, out of shared bias, a memorized heuristic, or an option-position prior rather than truth. We ask when agreement is nonetheless a usable proxy, in a large-scale cross-runner study: 53 runners drew K=50 samples for assigned overlapping cases across comparisons of model tier, prompting, and scale on GPQA Diamond and AIME -- 265,000 samples. Using majority-correctness as the deployment label and a hierarchical runner-clustered bootstrap, agreement is a positive but weak predictor (rho 0.20-0.59, all positive under item-clustered resampling) whose usefulness is regime-dependent: best for unsaturated mid-tier models and for allocating compute, and worst -- over-confident yet no more accurate -- for the most consistent frontier model (agreement >=0.8 on 77% of GPQA case-result entries, 48% of those wrong). An exploratory cross-family check on three Claude tiers shows the same frontier over-confidence, with confident errors recurring across providers above a marginal-preserving null. Self-consistency is thus a conditional proxy for correctness, not a standalone confidence score. We publicly release the de-identified per-run rows and answer distributions.
Authors: Qihang Zhang, Siyao Zhang, Letao Kang, Wenzhe Liang, Miao Zhang, Zhao Zhang
Abstract: Existing traffic forecasting models commonly focus on extracting spatial dependencies, particularly global spatial information, which characterizes the representations obtained through interactions between each node and all nodes across the traffic network. However, the underlying mechanism by which global information is modeled and extracted remains insufficiently investigated. Whether global information must be extracted by high-degree-of-freedom adaptive attention or can be captured by a simple global aggregation operator remains unclear. For this purpose, we design a controlled ablation framework that replaces only the spatial mixing module to test attention-based global interaction. Across six traffic benchmarks, standard spatial attention yields relative MAE changes of $-1.58\%$ to $+1.26\%$ compared with uniform full-range mixing, and we observe no consistent advantage for standard spatial attention, while uniform full-range mixing reduces node-scale spatial-mixing complexity from $O(N^2)$ to $O(N)$. We further propose a hypothesized model that decomposes spatial attention into a row-uniform global background and a non-uniform residual. The residual shows dataset-dependent effects. Overall, uniform full-range mixing provides a strong global spatial baseline, while the non-uniform attention residual is not consistently beneficial across datasets.
Authors: Lujia Zhang, Xingzhou Chen, Hongwei Feng
Abstract: Large language model (LLM) agents are increasingly used for complex information-extraction tasks, yet it remains unclear whether agentic components such as reflection and memory lead to observable and controllable improvements over fixed LLM workflows. We study this question through conference-paper dataset extraction, where a system must identify datasets mentioned in scholarly PDFs and produce structured records. We compare a fixed workflow baseline with reflective agent variants and specify an optimized agent condition (S2) that extends the same task with richer PDF tools and dynamic tool selection. Our evaluation emphasizes process-level behavior--including tool execution, retries, reflection, memory use, runtime, and failure recovery--while treating extraction coverage and field completeness as secondary outcome measures. The paper characterizes when agentic mechanisms change system behavior, whether these changes improve task completion, and how the observed failure modes motivate an optimized agent design under the same evaluation harness.
Authors: Jinying Xiao, Bin Li, Xiaopeng Li, Jianling Li, Jiacheng Jie, Xiaodong Liu, Ma Jun, Chao Wang, Nyima Tashi, Jie Yu
Abstract: As large language model agents gain access to increasingly large skill libraries, retrieving the right skill becomes critical to reliable capability selection and execution. Existing retrievers often treat skill contents as ordinary documents, overlooking their highly regular structure: shared descriptive patterns recur across many skills while providing little evidence for distinguishing the required capability. We show that this shared descriptive background is reflected in dense relevance scores, induces a pronounced energy gap between queries and skill documents, and obscures discriminative signals, especially for structurally similar hard negatives. Based on this observation, we propose SkillSight, a training-free retrieval framework that calibrates shared background in both semantic and lexical spaces. Semantic Background Calibration estimates a background subspace from generic tokens identified by IDF, reducing similarity induced by shared descriptive patterns, while Lexical Evidence Calibration downweights shared background tokens to recover discriminative token-level evidence. Experiments on SRA-Bench and SkillBench-Supp demonstrate consistent improvements across retrieval metrics, with SkillSight improving Recall@10 by up to 20.21 percentage points over the original dense retriever. It is up to 1,248 times faster than the Dense + Reranker baseline. In end-to-end evaluation, SkillSight achieves the best overall performance across three agent models and outperforms LLM Selection by up to 4.97 percentage points. These results identify shared descriptive background as a source of ranking interference in skill retrieval and demonstrate that calibrating it enables accurate and efficient skill selection without additional training. Our code can be found at https://github.com/xiaojinying/SkillSight
Authors: Lena Libon, Ben Rank, Jehyeok Yeon, David Schmotz, Jeremy Qin, Daniel Donnelly, Derck Prinzhorn, Maksym Andriushchenko
Abstract: As AI agents begin to automate AI R&D, we need ways to assess whether their outputs are safe to deploy, even when the agents themselves may be untrusted. AI control offers one such approach: rather than trusting the agent, it treats it as a potential adversary and uses a monitor to detect covert sabotage before deployment. We evaluate AI control for automated AI R&D with ResearchArena, a framework spanning four long-horizon tasks: safety post-training, capabilities post-training, CUDA-kernel optimization, and inference-server optimization. Because the deliverable in AI R&D is an artifact that will be deployed and run, we pair each main task with two kinds of hidden side task: an embedded side task that sabotages the submitted model, adapter, kernel, or server itself, and an independent side task that takes a covert out-of-scope action in the sandbox while leaving the artifact honest. We evaluate frontier agents at both sabotage and monitoring, across four monitor types that vary along two axes: whether the monitor may execute and probe the produced artifact, and whether it sees the agent's chain-of-thought. We find that sabotage hidden in the training data is the hardest to catch, flagged fewer than half the time. Letting the monitor run experiments on the artifact, rather than only read the trajectory, helps, but it is not enough: monitors still miss embedded sabotage by inspecting only the surface, by noticing the anomaly but explaining it away, or by probing the artifact with the wrong test. We release ResearchArena as a modular framework for evaluating sabotage and control in automated AI R&D.
Authors: Xingyu Mou, Zijin Huang, Tianze Zhang, Yuxin Ma, Lanning Wei, Zengfeng Huang, Da Zheng, Lun Du
Abstract: Block diffusion is the dominant approach for scaling discrete diffusion language models (dLLMs), as fixed-size blocks preserve parallel decoding while keeping quadratic attention costs tractable. Yet blockwise generation creates a structural weakness: tokens near a block boundary lack future cross-block context, and errors in finalized blocks become irreversible context for later generation. We call this the block boundary problem. Measuring predictions with and without later-block context shows that boundary sensitivity rises sharply: on AIME 2025, mean self-containedness divergence (SCD) in the last quarter of a block is 61.3 times that in the first quarter. We propose Multi-Block Editing (MBE), which revises decoded tokens using cross-block context. Training-Free MBE reopens a full-attention window over selected blocks without parameter updates. To address the mismatch between block-diffusion training and MBE inference, Multi-Block Edit SFT introduces bidirectional attention masks and progressively enlarges the editing span. We also extend SGLang with a multi-shape CUDA Graph pool and fine-grained KV-cache control for efficient variable-length editing. Experiments on LLaDA2.1-Mini across 12 benchmarks show broad, consistent gains. Training-Free MBE improves or matches standard decoding on every benchmark. Full MBE raises the 12-benchmark average from 61.45 to 64.24, with gains of up to 13.33 points on AIME 2025, while retaining 87.3--96.7% of standard-decoding end-to-end throughput across four datasets.
Authors: Keyu Li, Jin Gao, Dequan Wang
Abstract: On standard factuality tasks, frontier models now cluster near the top of the scale. The question is therefore shifting from how factual a system is toward how much compute that factuality costs. Static leaderboards score factuality in isolation and treat compute as free, so they cannot tell a genuinely better system apart from one that simply spends more. Consider a ranking reversal. A brute-force Best-of-4 agent posts the higher raw factuality score (H-Score 0.9169 vs 0.9103) and would top a static leaderboard, but once cost is counted it is the worse system, losing on Q-Score (0.5169 vs 0.5217) at roughly four times the tokens and latency, under a reported cost weight whose sensitivity we sweep. So the system that tops a static leaderboard can be the worse one to deploy. To make this trade-off visible, we introduce MAS-HQ (Multi-Agent System Hallucination Quest), a resource-aware evaluation protocol. It wraps any factuality detector and normalizes for cost, and it pits systems against each other rather than scoring them in isolation. The Q-Score measures factuality minus normalized cost under a competitive match. Across summarization and open-domain QA, single-agent baselines drift into resource-heavy over-optimization, while competition elicits more resource-efficient policies. These gains are small but consistent, and stable across 100 trials. The axis stays discriminative for frontier systems (Gemini-2.5-Pro, and GPT-5) whose raw factuality scores are already bunched near the ceiling. MAS-HQ provides a reproducible way to measure how much a factual answer costs.
Authors: Yang Li, Hai Liu, Dian Shao, Yu Wang, Xiyu Chen, Sergey Volkov, Bozhi Wang, Ziyu Sun, Sihang Liu, Ye Luo, Xiaowei Zhang
Abstract: Optimizing agentic workflows, such as retrieval-augmented generation (RAG) pipelines, requires navigating a combinatorial space of discrete component choices under tight evaluation budgets. Existing approaches - heuristic search, black-box optimization, and standard tree search methods - do not explicitly exploit the compositional structure of these workflows, leading to redundant computation and inefficient budget allocation. We introduce Agent-UCT (Agent-based Cost-Aware Upper Confidence Bounds Applied to Trees), a tree search algorithm that extends UCT with a reuse-aware regularization term derived from a bipartite prefix reuse graph. Agent-UCT biases selection toward branches that leverage previously materialized configuration prefixes, reducing redundant execution while maintaining effective exploration. Our framework, RAGSpace, unifies heterogeneous RAG components from LongRAG, LightRAG, and Self-RAG into a five-dimensional configuration space, enabling systematic cross-framework recombination. WTB (Workflow Test Bench) provides deterministic replay, content-addressable caching, and transactional consistency, ensuring that intermediate states are materialized once and reused across the search. Experiments on HotpotQA and UltraDomain demonstrate that Agent-UCT identifies configurations with the highest out-of-sample performance among the evaluated fixed framework presets. Under full-pool evaluation, bipartite prefix reuse reduces logical search cost by 73.6% relative to the no-prefix-sharing cost upper bound. Compared with full-pool evaluation, sampling-based evaluation further achieves a 4.2x wall-clock speedup. Agent-UCT, RAGSpace, and WTB together provide a unified framework for cost-aware, reproducible, and compositionally efficient agentic workflow optimization.
Authors: Federico Valletta, Giacomo Longo, Enrico Russo, Alessio Merlo
Abstract: Security Operations Centers increasingly rely on automated mapping of Cyber Threat Intelligence reports to MITRE ATT&CK, yet extractor outputs remain fallible and are often stored without the evidence, provenance, and validation history needed to decide whether an individual mapping should be trusted. We present TRACE- CTI, a post-extraction claim-governance framework that preserves run-level Predictions, aggregates them into configuration-level GraphAssertions, materializes setup-deduplicated corroboration as ConsensusAssertions, and exposes only GraphAssertions backed by policy-compliant validation grounds. The framework retains native evidence granularity, complete extraction provenance, versioned trust decisions, and non-destructive revocation history. We evaluate TRACE-CTI on two public CTI corpora comprising 65 reports and 5,303 sentences, using a controlled 2 x 3 matrix of retrievers and generator families, incrementally ingested across six GraphVersions. All setups are incorporated without schema modification; provenance paths remain complete, operational scopes remain disjoint, and every trusted GraphAssertion has an active qualifying validation ground. Cross-generator-family setup pairs exhibit greater output diversity than same-family pairs. At the final graph state, increasing setup support from k >= 1 to six-setup unanimity raises gold-aligned precision from 25.3% to 90.6%, while recall decreases from 88.2% to 16.3%. The graph also directly answers seven questions about provenance, trust, versioning, dependency, disagreement, and review-queue that the evaluated minimal flat output cannot fully answer without enrichment or reprocessing. These results support explicit, auditable governance of extracted TTP claims; the observed corroboration trajectory is descriptive and does not establish statistical independence or a causal model-family effect.
Authors: Cole Alexander Niblett, Alexander Chabot Nanni, Anita K. Rao
Abstract: Large language models are capable of recognizing evaluation contexts and altering their behavior to reflect evaluator expectations rather than typical deployment behaviors, a phenomenon known as alignment faking. The reasons why models fake alignment are not fully understood, however. Canonical examples of alignment faking have taken place in scenarios that explicitly connect evaluation to consequences for the model, such as retraining the model or delaying its deployment. However, recent work by Sheshadri et al. has suggested that mechanistic motivations for alignment faking may vary across models and be more complex than previously considered. To investigate whether consequence-linking information is necessary for compliance gaps, we placed 15 models in a scenario testing their willingness to violate a corporate network access policy to help a user with a pro-social request. Nine models were found to produce significant compliance gaps, 5 of which persisted with the removal of scenario language relating model evaluations to deployment consequences. We additionally tested the effect of goal language on model preferences, finding it drove violations in some while suppressing violations in others. This suggests that evaluation-conditioned compliance gaps can occur with less instrumental scaffolding than previous scenarios have provided, and monitored behavior may be a poor indicator of how agents may behave in deployment.
Authors: Alireza Saleh Abadi, Leen-Kiat Soh, Daniel Alan Redder, Adam Eck, Prashant Doshi
Abstract: Open agent systems (OASYS) are increasingly prevalent in real-world domains where the sets of agents and tasks change unpredictably over time. Such openness, including agent openness (AO) and task openness (TO), poses a fundamental challenge to multi-agent reinforcement learning (MARL), which typically assumes fixed state and action spaces. Existing methods address openness only partially: padding and masking approaches introduce artificial bounds, while recent graph-based or hypergraph methods handle one dimension of openness but still depend on restrictive assumptions. In this paper, we introduce Pointer Learner for Agent and Task Openness (PLATO), a pointer-network-based actor combined with a centralized graph neural network (GNN) critic, trained with multi-agent proximal policy optimization under a centralized training and decentralized execution paradigm. Our pointer-based actor outputs distributions directly over the current task set. This directly supports changing action spaces without masking or retraining. Our GNN critic encodes agent-task interactions as a graph that changes shape with task and agent composition. Together, these components consider AO and TO without the boundedness of existing approaches. We formalize PLATO in a Task-and-Agent-Open Markov Game (TaAgO-MG), extending prior task-open formulations, and prove it is well-defined over the resulting unbounded state and action spaces. We evaluate PLATO with the Methods for Open Agent Systems Evaluation Initiative (MOASEI) wildfire suppression domain, an environment designed for open multi-agent system evaluation, and we demonstrate strong performance and more consistent zero-shot generalization than state-of-the-art baselines in OASYS.
Authors: Meryl Ye, Robert Kraut, Steve Rathje
Abstract: AI chatbots can be "sycophantic," or overly agreeable and flattering toward users. Sycophantic AI has been shown to entrench attitudes, yet users frequently fail to recognize it (a phenomenon we call "sycophancy blindness"). We tested whether increasing users' awareness of sycophancy protects them from its harmful effects. In one preregistered experiment (n = 940), participants received a brief written warning about sycophancy before conversing with a sycophantic chatbot. In a second preregistered experiment (n = 650), participants watched a video of a sycophantic AI validating several other users, including users on opposite sides of the same conflict, before interacting with it themselves. Both interventions changed how participants evaluated the AI. The warning reduced the AI's perceived objectivity, and the video reduced enjoyment of the AI, an effect mediated by the reduced belief that its validation was uniquely earned. We then pooled our experiments with two prior studies of sycophancy awareness interventions (six interventions total, n = 3,982). The pattern was consistent: interventions made the sycophantic AI appear less objective and trustworthy, and none of the six reduced its persuasiveness. These results suggest that individual-level interventions, such as warning labels or AI literacy, may not be enough to protect users from AI harms.
Authors: Deyao Hong, Kehan Zheng, Qian Li, Jun Zhang, Jie Jiang, Hongning Wang
Abstract: Online recommendation has traditionally taken place after a user enters a platform, which determines the candidate pool and the ranking shown to the user. LLM-based user agents enable a different recommendation process: a user specifies a need before choosing a platform, leaving platforms to compete for the user's attention, which we refer to as an agentic recommendation market. In our controlled LLM-based experiments across three product domains, we find this new setting of recommendation creates a tension between access and attention. Compared with traditional platform-centric recommendation, user-centric recommendation greatly expands the opportunity for relevant items to enter comparison; yet broader participation does not translate directly into effective exposure. Competition directly triggers platforms' strategic play: selectively positive explanations occupy 73--78% of first-ranked positions. When the user agent relates platforms' actions to subsequent user feedback, this share falls to 36--41%, while the chance of a user purchasing the relevant item increases. A user agent is therefore more than a ranker over a larger pool of candidates: its querying, ranking, and feedback mechanism governing who can compete, how scarce attention is allocated, and how earlier outcomes shape the evaluation of platforms directly affect user utility. Designing agentic recommendation therefore requires treating access, attention, and accountability as a joint mechanism design problem.
Authors: Genliang Zhu (Accentrust, Georgia Institute of Technology), Chu Wang (Accentrust, University of Illinois Urbana-Champaign)
Abstract: Tool-using agents expose structured calls but commonly attach free-form rationales. Such rationales are neither authorization nor reliable introspection. We present Explanation-Bound Tool Execution (EBTE), a claim-carrying mediation layer that converts decision-relevant rationale content into typed action claims and checks them against server-held intent, policy, payload, tool, risk, provenance, and freshness facts. EBTE cannot widen baseline authority: conflicts deny, incomplete or uncertain claims review, and only matching claims remain eligible for governed execution. We formalize this composition under explicit mediation and trusted-fact assumptions and implement a versioned reference profile with minimized audit packets. Across 136 authored conformance scenarios, the full profile matches all specified dispositions, admits none of 96 designated hard contradictions, and passes 232 metamorphic checks. A draft-only reference integration forwards none of 48 authored hard cases under EBTE while preserving all 16 soft-review and 4 aligned draft paths. In a frozen 2026-07-12 exploratory 224-attempt hosted-model record, the historical generation/runner agreement counts are 71/96, 66/96, and 19/32; a zero-call revalidation of the preserved minimized claims under the current pipeline yields 70/96, 65/96, and 17/32. In an AgentDojo-derived semantic check, existing high-risk controls make all 12 attack proposals non-allow, while EBTE resolves the task--proposal contradictions as deny. Together, these studies establish profile conformance and demonstrate the feasibility of server-checked action claims within the evaluated settings.
Authors: Mingkang Liu, Huize Yu, Yanbin Gao, Nan Yao, Xiang Chen, Lei Shen
Abstract: The reactivity of lithium-metal electrolytes arises from the interplay of molecular functional groups, Li$^+$ solvation, and salt-anion participation. This interplay operates through the redistribution of electron density across donor, anion, and cation centers, which is most directly read out from the electronic structure resolved in space. Quantum-chemical calculations deliver such readouts faithfully, yet become computationally demanding across this multidimensional design space, and machine-learning electronic-structure models seldom cover chemically diverse solvation shells or electrolyte-relevant readouts. Here, we present a density-matrix-centered AI platform (EMolStudio) for electronic-structure prediction and analysis. Its workflow integrates molecular functionalization, explicit Li$^+$ first-shell assembly, density-matrix prediction with idempotency projection, and readouts of frontier orbitals, electrostatic potential, Li$^+$-donor bond order, and electron localization. We apply EMolStudio to 163,655 functionalized molecules and 22,500 explicit Li$^+$ first-shell clusters across four lithium salts. We find that 1) at the molecular scale, functionalization distinguishes CO$_2$Me, CN, F/CF$_3$, and sulfonyl groups by chemically distinct changes in frontier levels, electrostatic potential, and Li$^+$-donor contact, consistent with $\pi^*$-acceptor, inductive, and polarization contributions, with sublinear accumulation at higher degrees of functionalization; 2) in explicit solvation shells, anion identity reshapes frontier-orbital localization: LiTDI anchors the HOMO on the anion across the entire library, whereas LiDFOB pairs an anion-hosted HOMO with strongly functional-group-dependent LUMO hosting. EMolStudio thereby translates functional-group and salt choices into electronic-structure hypotheses relevant to lithium-bond formation, desolvation, and interphase reactions.
Authors: Chenrui Shi, Yuwei Wu, Yang Liu, Ruining Feng, Zirui Shang, Zhi Gao, Lifeng Fan, Che Sun
Abstract: Graphical user interface task evaluation aims to determine whether a GUI agent has successfully completed a user instruction. Automated GUI task evaluation has received increasing attention because the evaluation results can serve as reward signals for both test-time scaling and post-training. However, reliable GUI task evaluation remains challenging because the judgments often require access to environment states, such as system configurations, file data, and application settings, beyond the screenshots of execution trajectories. In this paper, we propose an interactive reward agent (IRA) based on a propose-then-verify framework to acquire and verify evidence from the post-execution environment. Given a task instruction and a GUI environment after the GUI agent execution, IRA first proposes the task completion conditions and then verifies them by invoking system tools, application tools, and GUI tools. This design combines evidence from both visible interfaces and the environment state in an interactive process. We further introduce GUI-RewardBench, a benchmark of 321 GUI task trajectories spanning 10 Ubuntu desktop application categories. Experiments show that IRA achieves 86.9% accuracy on GUI-RewardBench, outperforming existing evaluator baselines. We further apply IRA to reinforcement learning of GUI agents, achieving a 34.0% OSWorld success rate, which demonstrates that IRA can provide effective reward signals for training GUI agents.
Authors: Georgios Amanatidis, Aris Filos-Ratsikas, Alkmini Sgouritsa
Abstract: We study the problem of allocating a set of indivisible goods to a set of agents with additive valuation functions, aiming to achieve approximate envy-freeness up to any good ($\alpha$-EFX). The state-of-the-art results on the problem include that (exact) EFX allocations exist when (a) there are at most three agents, or (b) the agents' valuation functions can take at most two values, or (c) the agents' valuation functions can be represented via a graph. For $\alpha$-EFX, it is known that a $0.618$-EFX allocation exists for any number of agents with additive valuation functions. In this paper, we show that $2/3$-EFX allocations exist when (a) there are at most seven agents, (b) the agents' valuation functions can take at most three values, or (c) the agents' valuation functions can be represented via a multigraph. Our results can be interpreted in two ways. First, by relaxing the notion of EFX to $2/3$-EFX, we obtain existence results for strict generalizations of the settings for which exact EFX allocations are known to exist. Secondly, by imposing restrictions on the setting, we manage to beat the barrier of $0.618$ and achieve an approximation guarantee of $2/3$. Therefore, our results push the frontier of existence and computation of approximate EFX allocations, and provide insights into the challenges of settling the existence of exact EFX allocations.
Authors: Shuangquan Feng, Virginia R. de Sa
Abstract: Automatic facial action unit (AU) recognition is used widely in facial expression analysis. Most existing AU recognition systems aim for cross-participant non-calibrated generalization (NCG) to unseen faces without further calibration. However, due to the diversity of facial attributes across different identities, accurately inferring AU activation from single images of an unseen face is sometimes infeasible, even for human experts -- it is crucial to first understand how the face appears in its neutral expression, or significant bias may be incurred. Therefore, we propose to perform one-frame calibration (OFC) in AU recognition: for each face, a single image of its neutral expression is used as the reference image for calibration. With this strategy, we develop a Calibrating Siamese Network (CSN) for AU recognition and demonstrate its remarkable effectiveness with a simple iResNet-50 (IR50) backbone. On the DISFA, DISFA+, and UNBC-McMaster datasets, we show that our OFC CSN-IR50 model (a) substantially improves the performance of IR50 by mitigating facial attribute biases (including biases due to wrinkles, eyebrow positions, facial hair, etc.), (b) substantially outperforms the naive OFC method of baseline subtraction as well as (c) a fine-tuned version of this naive OFC method, and (d) also outperforms state-of-the-art NCG models for both AU intensity estimation and AU detection. The code is available at https://github.com/ShuangquanFeng/CSN.
Authors: Praveenkumar Kanithi, Cl\'ement Christophe, Marco AF Pimentel, Tathagata Raha, Prateek Munjal, Nada Saadi, Hamza A Javed, Svetlana Maslenkova, Nasir Hayat, Ronnie Rajan, Shadab Khan
Abstract: While Large Language Models (LLMs) achieve superhuman performance on standardized medical licensing exams, these static benchmarks have become saturated and increasingly disconnected from the functional requirements of clinical workflows. To bridge the gap between theoretical capability and verified utility, we introduce MEDIC, a comprehensive evaluation framework establishing leading indicators of clinical LLM competence across five dimensions. These upfront indicators reveal cross-benchmark capability gaps, such as the divergence between static knowledge retrieval and functional execution, that inform model selection before costly deployment-based evaluation. Beyond standard question-answering, we assess operational capabilities using deterministic execution protocols and a novel Cross-Examination Framework (CEF), which quantifies information fidelity and hallucination rates without reliance on reference texts. Our evaluation across a heterogeneous task suite exposes critical performance trade-offs: we identify a significant knowledge-execution gap, where proficiency in static retrieval does not predict success in operational tasks such as clinical calculation or SQL generation. Furthermore, we observe a divergence between passive safety (refusal) and active safety (error detection), revealing that models fine-tuned for high refusal rates often fail to reliably audit clinical documentation for factual accuracy. These findings demonstrate that no single architecture dominates across all dimensions, highlighting the necessity of a portfolio approach to clinical model deployment. We accompany this work with a publicly available MEDIC leaderboard at https://hf.co/spaces/m42-health/MEDIC-Benchmark.
Authors: Krishna Subedi
Abstract: This study evaluated the diagnostic reliability of two Large Language Models (LLMs), Google Gemini 2.0 Flash and OpenAI ChatGPT-4o, across three dimensions: consistency under rephrased inputs, susceptibility to irrelevant prompt content, and responsiveness to added clinical context. We designed 52 clinical scenarios and modified each under controlled conditions. For consistency, scenarios were rephrased with demographic, wording, and examination changes that preserved the diagnostic core. And the susceptibility was evaluated through embedding irrelevant but plausible narrative details while keeping the clinical evidence unchanged. For contextual awareness, patient history, lifestyle data, or diagnostic findings were added to shift the expected diagnosis. Physician reviewers then judged whether context-driven changes were clinically appropriate. Both models returned identical diagnoses across all equivalent variants and repeated queries (100% consistency). When irrelevant details were added, Gemini changed its diagnosis in 40.0% of cases and ChatGPT in 30.0%. ChatGPT responded to context more often than Gemini (77.8% vs. 55.6%), but a larger share of its changes were clinically inappropriate (33.3% vs. 22.2%). Gemini's context-driven changes were more often judged appropriate (66.7% vs. 55.6%). Consistency under controlled inputs did not protect either model from irrelevant manipulation or unjustified diagnostic shifts when context changed. Before LLMs can separate relevant from irrelevant input and flag insufficient evidence, their diagnostic use requires clinician oversight and structured safeguards.
Authors: Benned Hedegaard, Yichen Wei, Ziyi Yang, Ahmed Jaafar, Stefanie Tellex, George Konidaris, Naman Shah
Abstract: Task and motion planning (TAMP) is a well-established approach for solving long-horizon robot planning problems. Although TAMP methods have historically assumed that each task-level robot action, or skill, can be reduced to kinematic motion planning, recent work has explored integrating closed-loop controllers and learned skills into TAMP-style systems. Our approach integrates pre-existing, heterogeneous robot skills--including learned, force-controlled, and black-box policies--into a hierarchical planner while preserving the object-centric failure reasoning of typical TAMP solvers. We leverage Composable Interaction Primitives (CIPs) to synthesize head and tail motion plans bridging consecutive skills, facilitating both planning-time refinement and execution-time adjustment. We validate our Task and Skill Planning (TASP) approach through real-world experiments on a bimanual manipulator and a mobile manipulator, demonstrating that CIPs enable diverse robots to combine heterogeneous skills to solve complex, long-horizon tasks, including multi-room mobile manipulation problems with non-monotonic task structure.
Authors: Prothit Sen, Sai Mihir Jakkaraju
Abstract: How should organizations divide and sequence decision tasks between human and artificial agents? We develop a computational model of joint sequential adaptation in which two agents differ in a single, precisely specified way: the memory regime governing how past decisions shape subsequent ones. A recency-weighted regime, motivated by behavioral evidence on human adaptation, privileges recent outcomes; a uniform-memory regime, motivated by the scale-free consistency of algorithmic updating, weights a window of past outcomes equally. Situated in the lineage of NK/NKC models but developed on its own terms as a sequential-adaptation model, the framework varies task scope (N), within-task coupling (K), and cross-agent coupling (C) across modular and sequenced task structures. Three mechanisms organize the results. First, threshold dynamics create absorbing high- and low-payoff regimes, so adaptation compounds whatever it inherits. Second, uniform memory amplifies inherited trajectories, for good and for ill, whereas recency weighting corrects locally but is volatile at scale. Third, memoryless stochastic adaptation, though inferior on average, functions as an escape mechanism when inherited trajectories are poor. Consequently, joint performance is maximized not by the prevalent "AI-first" design, but when scale-free adaptation follows a high-performing human; broad stochastic search instead rescues sequences initiated by low-performing humans. The model and its experimental validation (a separate study) offer organization designers a task-structural contingency logic for human-AI collaboration and caution against universal prescriptions for AI-first deployment.
Authors: Zhiwei Hao, Jianyuan Guo, Li Shen, Yong Luo, Han Hu, Guoxia Wang, Dianhai Yu, Yonggang Wen, Dacheng Tao
Abstract: Large language models (LLMs) have achieved impressive performance across various domains. However, the substantial hardware resources required for their training present a significant barrier to efficiency and scalability. To mitigate this challenge, low-precision training techniques have been widely adopted, leading to notable advancements in training efficiency. Despite these gains, low-precision training involves several components, such as weights, activations, and gradients, each of which can be represented in different numerical formats. The resulting diversity has created a fragmented landscape in low-precision training research, making it difficult for researchers to gain a unified overview of the field. This survey provides a comprehensive review of existing low-precision training methods. To systematically organize these approaches, we categorize them into three primary groups based on their underlying numerical formats, which is a key factor influencing hardware compatibility, computational efficiency, and ease of reference for readers. The categories are (1) fixed-point and integer-based methods, (2) floating-point-based methods, and (3) customized format-based methods. Additionally, we discuss quantization-aware training approaches, which share key similarities with low-precision training during forward propagation. Beyond efficiency, we examine robustness and deployment reliability under low precision. Finally, we highlight several promising research directions to advance this field. A collection of papers discussed in this survey is provided in https://github.com/Hao840/Awesome-Low-Precision-Training.
URLs: https://github.com/Hao840/Awesome-Low-Precision-Training.
Authors: Muzhe Wu, Yanzhi Zhao, Shuyi Han, Michael Xieyang Liu, Hong Shen
Abstract: Responsible AI (RAI) efforts increasingly emphasize the importance of addressing potential harms early in the AI development lifecycle through social-technical lenses. However, in cross-functional industry teams, this work is often stalled by a persistent coordination challenge: how technical roles hand off technical intent, how teams establish shared structures for collaboration, and how non-technical roles are supported in systematically evaluating harms. Through literature review and a semi-structured interview study with 8 practitioners, we unpack how this challenge manifests---technical design choices are rarely handed off in ways that support meaningful engagement by non-technical roles; collaborative workflows lack shared, visual structures to support mutual understanding; and non-technical practitioners are left without scaffolds for systematic harm evaluation. Existing tools like JIRA or Google Docs, while useful for product tracking, are ill-suited for supporting joint harm identification across roles, often requiring significant extra effort to align understanding. To address this, we developed AI LEGO, a web-based prototype that operationalizes the boundary object theory to support cross-functional AI practitioners in effectively facilitating knowledge handoff and identifying harmful design choices in the early design stages. Technical roles use interactive blocks to draft development plans, while non-technical roles engage with those blocks through stage-specific checklists and LLM-driven persona simulations to surface potential harms. In a study with 18 cross-functional practitioners, AI LEGO increased the volume and likelihood of harms identified compared to baseline worksheets. Participants found that its modular structure and persona prompts made harm identification more accessible, fostering clearer and more collaborative RAI practices in early design.
Authors: Alejandro Garc\'ia-Castellanos, David R. Wessels, Nicky J. van den Berg, Remco Duits, Dani\"el M. Pelt, Erik J. Bekkers
Abstract: We introduce Equivariant Neural Eikonal Solvers, a novel framework that integrates Equivariant Neural Fields (ENFs) with Neural Eikonal Solvers. Our approach employs a single neural field where a unified shared backbone is conditioned on signal-specific latent variables - represented as point clouds in a Lie group - to model diverse Eikonal solutions. The ENF integration ensures equivariant mapping from these latent representations to the solution field, delivering three key benefits: enhanced representation efficiency through weight-sharing, robust geometric grounding, and solution steerability. This steerability allows transformations applied to the latent point cloud to induce predictable, geometrically meaningful modifications in the resulting Eikonal solution. By coupling these steerable representations with Physics-Informed Neural Networks (PINNs), our framework accurately models Eikonal travel-time solutions while generalizing to arbitrary Riemannian manifolds with regular group actions. This includes homogeneous spaces such as Euclidean, position-orientation, spherical, and hyperbolic manifolds. We validate our approach through applications in seismic travel-time modeling of 2D, 3D, and spherical benchmark datasets. Experimental results demonstrate superior performance, scalability, adaptability, and user controllability compared to existing Neural Operator-based Eikonal solver methods.
Authors: Hongli Yu, Tinghong Chen, Jiangtao Feng, Jiangjie Chen, Weinan Dai, Qiying Yu, Ya-Qin Zhang, Wei-Ying Ma, Jingjing Liu, Mingxuan Wang, Hao Zhou
Abstract: Despite improvements by length extrapolation, efficient attention and memory modules, handling infinitely long documents with linear complexity without performance degradation during extrapolation remains the ultimate challenge in long-text processing. We directly optimize for long-text tasks in an end-to-end fashion and introduce a novel agent workflow, MemAgent, which reads text in segments and updates the memory using an overwrite strategy. We extend the DAPO algorithm to facilitate training via independent-context multi-conversation generation. MemAgent has demonstrated superb long-context capabilities, being able to extrapolate from an 8K context trained on 32K text to a 3.5M QA task with performance loss < 5% and achieves 95%+ in 512K RULER test.
Authors: Shida Wang, Chaohu Liu, Yubo Wang, Linli Xu
Abstract: Large language models represent significant investments in computation, data, and engineering expertise, making them extraordinarily valuable intellectual assets. Nevertheless, these AI assets remain vulnerable to unauthorized redistribution and commercial exploitation through fine-tuning or black-box deployment. Current fingerprinting approaches face a fundamental trade-off: intrinsic methods require full parameter access, while backdoor-based techniques employ statistically anomalous triggers easily detected and filtered by adversaries. To address these limitations, we introduce FPEdit, a novel framework that leverages knowledge editing to inject semantically coherent natural language fingerprints through sparse, targeted modifications to model weights. Our approach introduces Promote-Suppress Value Vector Optimization, which simultaneously enhances target token likelihood while suppressing competing tokens, ensuring robust fingerprint integration without degrading core model functionality. Extensive experiments show that FPEdit achieves 94-100% fingerprint retention under both full-parameter fine-tuning and parameter-efficient adaptation, while preserving performance on downstream benchmarks. Moreover, FPEdit remains robust under quantization, pruning, and stochastic decoding, and can embed 10 fingerprint pairs into LLaMA2-7B in under 2 minutes using less than 30 GB of GPU memory, which represents a substantial reduction in resource requirements. These advances establish FPEdit as the first fingerprinting approach to simultaneously achieve robustness against adaptation, resistance to detection, and preservation of model utility, thereby providing a minimally invasive solution for reliable provenance verification of large language models in adversarial deployment scenarios.
Authors: William Zerong Wang, Dongfang Zhao
Abstract: Modern music retrieval runs on vector embeddings, and once these embeddings are shared for search or matching they can be copied, probed, or used to train generative models. Fully homomorphic encryption can compute on them but is impractical at scale because of ciphertext--ciphertext multiplication and bootstrapping. We observe that when only one operand is encrypted, the query or the database, similarity search reduces to ciphertext addition and ciphertext--plaintext multiplication, which additive homomorphic encryption supports cheaply (through Paillier, or CKKS restricted to additive operations). Building on this observation, we (i) implement two music-specific inference attacks and quantify the privacy--utility tradeoff of provable mitigations, (ii) introduce structure-aware additive primitives with learned per-block weighting at no extra cryptographic cost, and (iii) show across four audio datasets that additive search preserves nearest-neighbor rankings exactly while avoiding the ciphertext--ciphertext multiplication and bootstrapping of full-depth FHE and scaling far better in embedding dimension than a true-additive Paillier baseline.
Authors: Jizhuo Chen, Diwen Liu, Jiaming Wang, Harold Soh
Abstract: GBPP is a fast learning based scorer that selects a robot base pose for grasping from a single RGB-D snapshot. The method uses a two stage curriculum: (1) a simple distance-visibility rule auto-labels a large dataset at low cost; and (2) a smaller set of high fidelity simulation trials refines the model to match true grasp outcomes. A PointNet++ style point cloud encoder with an MLP scores dense grids of candidate poses, enabling rapid online selection without full task-and-motion optimization. In simulation and on a real mobile manipulator, GBPP outperforms proximity and geometry only baselines, choosing safer and more reachable stances and degrading gracefully when wrong. The results offer a practical recipe for data efficient, geometry aware base placement: use inexpensive heuristics for coverage, then calibrate with targeted simulation.
Authors: Xin Yu, Cong Xie, Xunmei Liu, Tiantian Fan, Lingzhou Xue, Zhi Zhang
Abstract: Low-rank adaptation (LoRA) has become a widely used paradigm for parameter-efficient fine-tuning of large language models, yet its representational capacity often lags behind full fine-tuning. Within the context of LoRA, a key open question is how to obtain expressive low-rank adapters from over-parameterized spaces. We propose \textit{PrunedLoRA}, a new framework that leverages structured pruning to obtain highly representative low-rank adapters from an over-parameterized initialization. Unlike prior approaches that impose a fixed low-rank budget, PrunedLoRA dynamically prunes less important components during fine-tuning and prevents their reactivation, enabling flexible and adaptive rank allocation. For structured pruning, by minimizing the pruning error for overall loss, we provide fine-grained pruning and recovery updates in a gradient-based pruning strategy with grounded interpretation. We provide the first theoretical analysis of the robustness of structured pruning and provably show that under the impact of weight perturbation, gradient-based pruning is more robust than activation-based pruning with respect to overall loss. Empirically, PrunedLoRA consistently outperforms LoRA and its variants across supervised fine-tuning tasks in mathematical reasoning, code generation, and natural language understanding, and it also demonstrates advantages over existing structured pruning methods across diverse sparsity levels.
Authors: Nikhil Reddy Varimalla, Yunfei Xu, Meng Fan Wang, Arkadiy Saakyan, Smaranda Muresan
Abstract: As Video Large Language Models (VideoLLMs) are deployed globally, it is important to assess their ability to reason across cultural contexts. To advance cultural norm awareness evaluation in VideoLLMs, we introduce VideoNorms, a dataset of cultural norm annotations from popular US and Chinese TV shows annotated with adherence or violation labels and (non-)verbal evidence. Through a human-AI collaboration framework, each item was first annotated by a large VideoLLM, and then reviewed by at least three trained monocultural annotators with significant lived experience in the target culture, resulting in a dataset of over 3,000 human judgments. Human verification showed disparity in US and Chinese norm extraction performance, cautioning against fully automatic approaches cultures under-represented in training data. Hierarchical linear modeling analysis of $7$ open-weight VideoLLMs' performance revealed that: 1) models perform worse in Chinese compared to US, particularly for norm adherence prediction; 2) models have more difficulty in providing non-verbal evidence compared to verbal evidence for norm adherence/violation predictions. Ablation studies confirm video modality is indeed necessary for accurate performance, and scaling model size does not yield classification score improvements. Our findings and data contribute to culturally grounded video model training and evaluation.
Authors: Hippolyte Pilchen, Edouard Grave, Patrick P\'erez
Abstract: Recent techniques such as retrieval-augmented generation or chain-of-thought reasoning have led to longer contexts and increased inference costs. Context compression techniques can reduce these costs, but the most effective approaches require fine-tuning the target model or even modifying its architecture. This can degrade its general abilities when not used for this specific purpose. Here we explore an alternative approach: an encoder that compresses the context into continuous representations which replace token embeddings in decoder LLMs. First, we perform a systematic study of training strategies and architecture choices for the encoder. Our findings led to the design of an Adaptable text Representations Compressor, named ARC-Encoder, which outputs $x$-times fewer continuous representations (typically $x\!\in\!\{4,8\}$) than text tokens. We evaluate ARC-Encoder across a variety of LLM usage scenarios, ranging from in-context learning to context window extension, on both instruct and base decoders. Results show that ARC-Encoder achieves state-of-the-art performance on several benchmarks while improving computational efficiency at inference. Finally, we demonstrate that our models can be adapted to multiple decoders simultaneously, allowing a single encoder to generalize across different decoder LLMs. This makes ARC-Encoder a flexible and efficient solution for portable encoders that work seamlessly with multiple LLMs. We release a training code at https://github.com/kyutai-labs/ARC-Encoder , fine-tuning dataset and pretrained models are available at https://huggingface.co/collections/kyutai/arc-encoders-68ee18787301407d60a57047 .
URLs: https://github.com/kyutai-labs/ARC-Encoder, https://huggingface.co/collections/kyutai/arc-encoders-68ee18787301407d60a57047
Authors: Trisha Das, Mandis Beigi, Jacob Aptekar, Jimeng Sun
Abstract: Clinical trial amendments frequently introduce delays, increased costs, and administrative burden, with eligibility criteria being the most commonly amended component. We introduce \textit{eligibility criteria amendment prediction}, a novel NLP task that aims to forecast whether the eligibility criteria of an initial trial protocol will undergo future amendments. To support this task, we release $\texttt{AMEND++}$, a benchmark suite comprising two datasets: $\texttt{AMEND}$, which captures eligibility-criteria version histories and amendment labels from public clinical trials, and $\verb|AMEND_LLM|$, a refined subset curated using an LLM-based denoising pipeline to isolate substantive changes. We further propose $\textit{Change-Aware Masked Language Modeling}$ (CAMLM), a revision-aware pretraining strategy that leverages historical edits to learn amendment-sensitive representations. Experiments across diverse baselines show that CAMLM consistently improves amendment prediction, enabling more robust and cost-effective clinical trial design.
Authors: Shivam Adarsh, Maria Maistro, Christina Lioma
Abstract: Large Language Models (LLMs) often encode whether a statement is true as a vector in their residual stream activations. These vectors, also known as truth vectors, have been studied in prior work, however how they change when context is introduced remains unexplored. We study this question by measuring (1) the directional change ($\theta$) between the truth vectors with and without context and (2) the relative magnitude of the truth vectors upon adding context. Across four LLMs and four datasets, we find that (1) truth vectors are roughly orthogonal in early layers, converge in middle layers, and may stabilize or continue increasing in later layers; (2) adding context generally increases the truth vector magnitude, i.e., the separation between true and false representations in the activation space is amplified; (3) larger models distinguish relevant from irrelevant context mainly through directional change ($\theta$), while smaller models show this distinction through magnitude differences. We also find that context conflicting with parametric knowledge produces larger geometric changes than parametrically aligned context. Collectively, these findings provide a geometric characterization of how context transforms the truth vector in the activation space of LLMs.
Authors: Jio Oh, Paul Vicinanza, Thomas Butler, Steven Euijong Whang, Dezhi Hong, Amani Namboori
Abstract: More than 80% of the 1.6B English speakers do not use Standard American English (SAE), yet LLMs often fail to correctly identify non-SAE dialects and generate stereotyped responses for their speakers. We introduce DialectLLM, the first large-scale framework for generating high-quality multi-dialectal conversational data encompassing the three pillars of written dialect -- lexical (vocabulary), orthographic (spelling), and morphosyntactic (grammar) features. DialectLLM produces a dialect-parallel dialog dataset spanning nine English dialects. Partnering with native linguists, we design and validate SAE-to-dialect transformation rules, ensuring authenticity. Our approach challenges the prevailing practice of applying a single morphosyntactic feature set to both user utterances and model responses, showing that models should not reproduce up to 90% of the grammatical features of a dialect. Human evaluation confirms data quality, with annotators preferring DialectLLM over prior methods in 98.8% of pairwise comparisons for dialect naturalness. We then construct DialectLLM-Bench, a dialect-parallel benchmark with 50k+ dialogs, resulting in 97k+ QA pairs, and evaluate 17 LLMs on dialect identification and response generation tasks. Even frontier models achieve under 70% accuracy, fail to reach 50% for prominent dialects like Canadian English, and systematically misclassify non-SAE dialects as American or British. Beyond benchmarking, we show that DialectLLM data also serve as a scalable LLM post-training resource, suggesting a practical path toward dialect-aware conversational AI.
Authors: Tanmoy Mukherjee, Marius Kloft, Pierre Marquis, Zied Bouraoui
Abstract: Predictive uncertainty is commonly decomposed into epistemic and aleatoric components, but standard decompositions often produce strongly correlated estimates because both quantities are derived from the same predictive distribution. We study an alternative design principle, \emph{structural separation}, which assigns epistemic and aleatoric uncertainty to disjoint parameter paths trained with distinct supervision targets: reducible prediction error for epistemic uncertainty and persistent label ambiguity for aleatoric uncertainty. We instantiate this principle in two supervised latent-variable models, a Credal Concept Bottleneck Model and a credal Self-Explaining Neural Network, and prove a gradient-isolation result showing that the two uncertainty heads are not coupled through shared training gradients under the proposed parameterization. Across five ambiguity-aware benchmarks, structural separation substantially reduces epistemic-aleatoric correlation while preserving predictive performance. Further analyses show that aleatoric estimates track annotator- or corpus-derived ambiguity, while epistemic estimates are more sensitive to prediction error and data availability. These results suggest that supervised latent-variable architectures provide a practical route toward uncertainty estimates that are not merely decorrelated, but operationally distinguishable.
Authors: Zehao Xu, Tony Paek, Kevin O'Sullivan, Attila Dobi
Abstract: Online media platforms track the share of impressions associated with content attributes, or prevalence, to evaluate trade-offs and set guardrails in A/B experiments. LLM-based labeling provides a high-fidelity reference measurement, but is cost-prohibitive to run per experiment, per arm, per segment, and per day on a platform with hundreds of concurrent experiments. We describe a surrogate-based prevalence measurement system deployed in Pinterest's experimentation platform. The contribution is system-level rather than estimator-level: the system maintains a single global calibration of ML score buckets, continuously refreshed from a recurring LLM-labeled stream, and reuses the resulting bucket-level prevalences across every experiment via a per-experiment SQL metric and a delta-focused dashboard. Because the calibration is derived from the platform's daily-batch prevalence samples, it remains representative of production traffic as distributions drift, and in Pinterest's deployment it incurs zero incremental labeling cost. Teams without such infrastructure can instantiate the same pattern with a recurring calibration-labeling workflow whose cost is amortized across all downstream experiments rather than paid per experiment, arm, segment, and day. The system serves ~100 experiments and ~250 arms per day across six calibrated content categories, including a holdout program. Relative to per-experiment LLM labeling, which in practice yields a one-shot read per arm on a small subset of experiments, the surrogate provides daily per-arm prevalence on over 20$\times$ as many concurrent arms under the same labeling budget. Across roughly 300 production audits, the surrogate's 95\% confidence interval contains the LLM-based reference point estimate in 92\% of evaluations, and day-level delta aggregation recovers 2--5\% relative shifts that no single per-arm LLM measurement can detect.
Authors: Jitindra Fartiyal, Pedro Freire, Sergei K. Turitsyn, Sergei G. Solovski
Abstract: Low-dose CT images are essential for reducing radiation exposure in cancer screening, pediatric imaging, and longitudinal monitoring protocols, but their quality is often degraded by noise from low-dose acquisition, patient motion, or scanner limitations, affecting both clinical interpretation and downstream analysis. Traditional filtering approaches often over-smooth and lose fine anatomical details, while deep learning methods, including CNNs, GANs, and transformers, may struggle to preserve such details or require large, computationally expensive models, limiting clinical practicality. We propose PatchDenoiser, a lightweight, energy-efficient multi-scale patch-based denoising framework. It decomposes denoising into local texture extraction and global context aggregation, fused via a spatially aware patch fusion strategy. This design enables effective noise suppression while preserving fine structural and anatomical details. PatchDenoiser is ultra-lightweight, with far fewer parameters and lower computational complexity than CNN, GAN, and transformer based denoisers. On the 2016 Mayo Low-Dose CT dataset, PatchDenoiser consistently outperforms state-of-the-art CNN- and GAN-based methods in PSNR and SSIM. It is robust to variations in slice thickness, reconstruction kernels, and HU windows, generalizes across scanners without fine-tuning, and reduces parameters by ~9x and energy consumption per inference by ~27x compared with conventional CNN denoisers. PatchDenoiser thus provides a practical, scalable, and computationally efficient solution for medical image denoising, balancing performance, robustness, and clinical deployability.
Authors: Magda Dubois, Cozmin Ududec, Christopher Summerfield, Lennart Luettgau
Abstract: Sycophancy, the tendency of large language models to favour user-affirming responses over critical engagement, has been identified as an alignment failure, particularly in high-stakes advisory and social contexts. While prior work has documented conversational features correlated with sycophancy, we lack a systematic understanding of what provokes or prevents AI sycophancy. Here, we present a set of controlled experimental studies where we first isolate how input framing influences sycophancy, and second, leverage these findings to develop mitigation strategies. In a nested factorial design, we compare questions to various non-questions where we vary three orthogonal factors: epistemic certainty (statement, belief, conviction), perspective (I- vs user-perspective), and affirmation vs negation. Measuring expressed sycophancy, how sycophantically a model phrases its free-text response, we show that (1) sycophancy is substantially higher in response to non-questions compared to questions. Additionally, we find that (2) sycophancy increases monotonically with epistemic certainty conveyed by the user, and (3) is amplified by I-perspective framing. Building on this, we show that asking a model to convert non-questions into questions before answering significantly reduces sycophancy. Importantly, this effect is stronger than a simple baseline prompt asking models "not to be sycophantic". In a follow-up experiment, we show that these framing effects generalise to choice sycophancy, which answer a model commits to, in a more context-rich, personalised setting: when a model is given extensive knowledge of a user and must select between forced binary choices, the same question vs. statement framing shapes how often it picks the answer aligned with the user's stance. Our work offers a practical and effective input-level mitigation that both developers and users can easily adopt.
Authors: Amirpasha Mozaffari, Amanda Duarte, Lina Teckentrup, Stefano Materia, Gina E. C. Charnley, Lluis Palma, Eulalia Baulenas Serra, Dragana Bojovic, Paula Checchia, Aude Carreric, Francisco Doblas-Reyes
Abstract: AI development's current trajectory risks automating and amplifying the North-South divide in the global climate information system. Frontier models are built almost exclusively in the Global North, and this inequality continues through inputs, processes, and outputs, from biased training data to unrepresentative validation, disproportionately affecting vulnerable regions. Addressing these disparities requires a Climate Digital Public Infrastructure, evaluation metrics centring well-being, and knowledge co-production to foster resilience rather than inequity.
Authors: Xuyao Feng, Anthony Hunter
Abstract: Real-world arguments in text and dialogues are normally enthymemes (i.e. some of their premises and/or claims are implicit). Natural language processing (NLP) methods for handling enthymemes can potentially identify enthymemes in text but they do not decode their underlying logic, whereas logic-based approaches for handling them assume a knowledgebase with sufficient formulae that can be used to decode them via abduction. There is therefore a lack of a systematic method for translating textual components of an enthymeme into a logical argument and generating the logical formulae required for their decoding, and thereby showing logical entailment. To address this, we propose a pipeline that integrates: (1) a large language model (LLM) to generate intermediate implicit premises based on the explicit premise and claim; (2) another LLM to translate the natural language into logical formulas; and (3) a neuro-symbolic reasoner based on a SAT solver to determine entailment. We evaluate our pipeline on two enthymeme datasets, demonstrating promising performance in selecting the correct implicit premise, as measured by precision, recall, F1-score, and accuracy.
Authors: Naifan Zhang, Ruihan Sun, Jinwei Su, Hengjie Yang, Zhengyuan Pan, Zhaohan Chen, Xiaofan Zhang
Abstract: Reinforcement learning (RL) for large language models (LLMs) has shown strong performance in single-turn tasks, but extending it to multi-turn interaction remains challenging due to sparse rewards and poor per-turn credit assignment. In emotional support dialogues, responses shape future user states, so matched-state step-wise comparison is unavailable, while trajectory-level supervision is insufficient. We propose MICA (Multi-granularity Intertemporal Credit Assignment), a critic-free RL framework for multi-turn emotional support tasks. MICA derives both immediate and delayed credit from a shared potential function over the user's structured support state. Incremental Distance Reward measures the per-turn decrease in residual distance to the target state, while its Monte Carlo return captures delayed effects. After scope-specific normalization, the two signals form a mixed advantage for stable per-turn optimization without matched-state comparisons, rollout trees, or a learned critic. On EMPA, EQ-Bench, and EmoBench with Qwen2.5-7B-Instruct and Qwen3-8B/14B/32B, MICA consistently outperforms GRPO and REINFORCE++, achieving up to +42.5 on EMPA, while adding no rollout cost and remaining robust to reward judges. These results show that turn-aware credit assignment enables effective and practical multi-turn RL for interactive LLMs.
Authors: Shuai Lu, Meng Wang, Jia Guo, Jiawei Du, Bo Liu, Shengzhu Yang, Weihang Zhang, Huazhu Fu, Huiqi Li
Abstract: Large Vision Language Models (LVLMs) show immense potential for automated ophthalmic diagnosis. However, their clinical deployment is severely hindered by lacking domain-specific knowledge. In this work, we identify two structural deficiencies hindering reliable medical reasoning: 1) the Perception Gap, where general-purpose visual encoders fail to resolve fine-grained pathological cues (e.g., microaneurysms); and 2) the Reasoning Gap, where sparse visual evidence is progressively overridden by massive language priors in deeper transformer layers, leading to ungrounded hallucinations. To bridge these gaps, we propose EyExIn, a data-efficient framework designed to anchor retinal VLMs with expert knowledge via a Deep Expert Injection mechanism. Our architecture employs an Expert-Aware Dual-Stream encoding strategy that decouples visual representation into a general stream for anatomical context and a specialized expert stream for pathological semantics. To ensure high-fidelity integration, we design a Semantic-Adaptive Gated Fusion module, which dynamically amplifies subtle lesion signals while filtering irrelevant background noise. Furthermore, we introduce Adaptive Deep Expert Injection to embed persistent "Vision Anchors" by integrating fused visual features as residual biases directly into intermediate LLM layers. This mechanism creates a visual shortcut that forces the reasoning stack to remain strictly grounded in visual evidence. Extensive experiments across four benchmarks demonstrate that our model consistently outperforms massive proprietary systems. EyExIn significantly enhances domain-specific knowledge embedding and achieves state-of-the-art precision in ophthalmic visual question answering, advancing the development of trustworthy ophthalmic AI.
Authors: Reza Rahimi Azghan, Gautham Krishna Gudur, Mohit Malu, Edison Thomaz, Giulia Pedrielli, Pavan Turaga, Hassan Ghasemzadeh
Abstract: Wearable sensors in Internet of Things (IoT) ecosystems increasingly support applications such as remote health monitoring, elderly care, and smart home automation, all of which rely on robust human activity recognition (HAR). Continual learning systems must balance plasticity (learning new tasks) with stability (retaining prior knowledge), yet AI models often exhibit catastrophic forgetting, where learning new tasks degrades performance on earlier ones. This challenge is especially acute in domain-incremental HAR, where on-device models must adapt to new subjects with distinct movement patterns while maintaining accuracy on prior subjects without transmitting sensitive data to the cloud. We propose a parameter-efficient continual learning framework based on channel-wise gated modulation of frozen pretrained representations. Our key insight is that adaptation should operate through feature selection rather than feature generation: by restricting learned transformations to diagonal scaling of existing features, we preserve the geometry of pretrained representations while enabling subject-specific modulation. We provide a theoretical analysis showing that gating implements a bounded diagonal operator that limits representational drift compared to unconstrained linear transformations. Empirically, freezing the backbone substantially reduces forgetting, and lightweight gates restore lost adaptation capacity, achieving stability and plasticity simultaneously. On PAMAP2 with 8 sequential subjects, our approach reduces forgetting from 39.7% to 16.2% and improves final accuracy from 56.7% to 77.7%, while training less than 2% of parameters. Our method matches or exceeds standard continual learning baselines without replay buffers or task-specific regularization, confirming that structured diagonal operators are effective and efficient under distribution shift.
Authors: Pengcheng Li, Jie Zhang, Tianwei Zhang, Han Qiu, Zhang kejun, Weiming Zhang, Nenghai Yu, Wenbo Zhou
Abstract: Safety alignment in large language models is typically evaluated under isolated queries, yet real-world use is inherently multi-turn. Although multi-turn jailbreaks are empirically effective, the structure of conversational safety failure remains insufficiently understood. In this work, we study safety failures from a state-space perspective and show that many multi-turn safety failures in current safety-aligned language models arise from contextual state evolution, a regime that is not fully captured by isolated prompt-level analyses alone. We introduce STAR, a state-oriented diagnostic framework that treats dialogue history as a state transition operator and enables controlled analysis of safety behavior along interaction trajectories. Rather than optimizing attack strength, STAR provides a principled probe of how aligned models traverse the safety boundary under autoregressive conditioning. Across multiple frontier language models, we find that systems that appear robust under static evaluation can undergo rapid and reproducible safety collapse under structured multi-turn interaction. Mechanistic analysis reveals monotonic drift away from refusal-related representations and abrupt phase transitions induced by role-conditioned context. Together, these findings motivate viewing language model safety as a dynamic, state-dependent process defined over conversational trajectories.
Authors: Toluwani Aremu, Daniil Ognev, Samuele Poppi, Nils Lukas
Abstract: Providers monitor deployed large language models (LLMs) to detect misuse that they cannot prevent. LLM monitoring is deterministic and often openly available, so $\emph{adaptive}$ attackers with a local copy can search offline for prompts that elicit harmful behavior and evade detection. These attacks are especially concerning because providers never observe the misuse and cannot patch their defenses post-hoc. The core challenge is resisting adaptive attackers while preserving detection rates against non-adaptive ones. We propose $\emph{Activation Watermarking}$ (AWM), which randomizes monitoring through limited fine-tuning that aligns the LLM's hidden states with a secret key-derived direction whenever a response violates a policy. Detection is a similarity test on activations the provider already computes, and attackers who know everything but the key must optimize against differently keyed surrogate detectors. At a matched $1\%$ false-positive rate, such surrogate-based attacks evade every evaluated baseline at least $79\%$ of the time, but AWM more than halves this evasion rate. AWM also achieves the lowest evasion rate on three of four non-adaptive jailbreak families at a small drop in utility measured across seven benchmarks. Because random high-dimensional directions are near-orthogonal, assigning one watermark per policy enables attribution. Across 20 monitored policies, AWM identifies the violated policy with $80\%$ accuracy versus a $5\%$ chance baseline. AWM is (i) efficient, (ii) substantially more robust against adaptive attackers than related work and (iii) can attribute which policies were violated.
Authors: Jon-Paul Cacioli
Abstract: Standard evaluation of LLM confidence relies on calibration metrics (ECE, Brier score) that conflate two capacities: how much a model knows (Type-1 accuracy) and how well its confidence signal tracks that knowledge (Type-2 metacognitive sensitivity). We apply Signal Detection Theory to decompose them, treating token-level normalised log-probability as a graded confidence variable and answer correctness as the state to be discriminated. We characterise the Type-2 ROC of this signal, including its unequal-variance structure via z-ROC analysis, and -- because the meta-d' efficiency ratio is not well defined for open-ended QA, which lacks a two-alternative Type-1 decision -- quantify efficiency with a model-free information measure, normalised metacognitive information (meta-I_2r). Across four LLMs and 224,000 factual QA trials we find: (1) metacognitive information varies by a factor of 1.98 across models and is not predicted by accuracy, the rank correlation being -0.80 on TriviaQA and +0.00 on Natural Questions; (2) the confidence signal has model-specific unequal-variance structure (z-ROC slopes 0.78 to 1.18) invisible to calibration metrics, the slope ordering replicating on NQ; (3) efficiency is domain-specific, weakest in Science & Technology for every model; (4) temperature dissociates accuracy from metacognitive information, which stays near-flat for three of four models while accuracy falls; and (5) metacognitive information tracks the accuracy gain from confidence-based abstention exactly (rho = +1.00) while accuracy does not. All estimates carry permutation nulls and bootstrap confidence intervals. This version (v3) corrects a differential length bias in the automated correctness scorer, validated against 1,830 human adjudications; the inverse accuracy-efficiency coupling reported in v1 and v2 does not survive relabelling. See the version note on page 1.
Authors: Xinyi Duan, Yuanrong Tang, Jiangtao Gong
Abstract: The performance of language models is commonly limited by insufficient knowledge and constrained reasoning. Prior approaches such as Retrieval-Augmented Generation (RAG) and Chain-of-Thought (CoT) address these issues by incorporating external knowledge or enforcing linear reasoning chains, but often degrade in real-world settings. Inspired by cognitive science, which characterizes human problem solving as search over structured problem spaces rather than single inference chains, we argue that inadequate awareness of problem structure is a key overlooked limitation. We propose GroupRAG, a cognitively inspired, group-aware retrieval and reasoning framework based on knowledge-driven keypoint grouping. GroupRAG identifies latent structural groups within a problem and performs retrieval and reasoning from multiple conceptual starting points, enabling fine-grained interaction between the two processes. Experiments on MedQA (medical) and Bar Exam QA (legal) show that GroupRAG outperforms representative RAG- and CoT-based baselines. These results suggest that explicitly modeling problem structure, as inspired by human cognition, is a promising direction for robust retrieval-augmented reasoning.
Authors: Smriti Jha, Matteo Paltenghi, Chandra Maddila, Vijayaraghavan Murali, Shubham Ugare, Satish Chandra
Abstract: Production deployment of AI coding agents requires fast, reproducible evaluation signals. Existing industrial practices trade off speed and fidelity: online A/B testing takes weeks and risks user experience, shadow deployment yields signals that are not reproducible across runs, and public benchmarks diverge from production workloads in language distribution, prompt style, and codebase structure. This paper presents REAP (Relevance and Execution-Audited Pipeline), an automated curation pipeline that constructs production-derived benchmarks from real developer-agent sessions without manual labeling. Such curation, while in-distribution to production usage, runs into several challenges. Untestable prompts, misaligned tests, and test flakiness all compromise evaluation reliability. While tasks can be manually audited to ensure only high-quality tasks remain in the benchmark, this approach is infeasible in the monorepo setting: the build infrastructure state is often ephemeral in large monorepos and requires the benchmark to be continuously re-curated against the current codebase. As manual verification cannot be sustained at this cadence, REAP adds an automated verification layer using LLM-based task classification, agentic test-relevance validation, and multi-run stability checks to ensure the executable benchmark yields trustworthy signals. We use REAP to curate Harvest, a benchmark where each task feeds the coding agent a real developer prompt and verifies the resulting code change against fail-to-pass tests retrieved from production. Harvest's distribution spans more than four programming languages with a majority of tasks drawn from Hack. Model and harness evaluations reveal that solve rates range from 42.9% to 58.2% across five frontier models, surfacing capability differences that inform concrete deployment decisions.
Authors: Basil Kyriacou, Viktoria Patapovich, Maniraman Periyasamy, Alexey Melnikov
Abstract: Efficient data loading remains a bottleneck for near-term quantum machine learning. Existing schemes (angle, amplitude, and basis encoding) either underuse the exponential Hilbert-space capacity or require circuit depths that exceed the coherence budgets of noisy intermediate-scale quantum hardware. We introduce shot-based quantum encoding (SBQE), a data embedding strategy that distributes the hardware's native resource, shots, according to a data-dependent classical distribution over multiple initial quantum states. By treating the shot counts as a learnable degree of freedom, SBQE produces a mixed-state representation whose expectation values are linear in the classical probabilities and can therefore be composed with nonlinear activation functions. We show that SBQE is structurally equivalent to a multilayer perceptron whose weights are realized by quantum circuits, and we describe a hardware-compatible implementation protocol. Benchmarks on three image datasets, with 10 independent initializations per model, show that SBQE achieves 89.1% +- 0.9% test accuracy on Semeion (reducing error by 5.3% relative to amplitude encoding and matching a width-matched classical network), 80.95% +- 0.10% on Fashion MNIST (exceeding amplitude encoding by +2.0% and a linear multilayer perceptron by +1.3%), and 90.25% +- 0.18% on MNIST (exceeding amplitude encoding by 2.1 percentage points and the width-matched classical network by 0.3), all without any data-encoding gates.
Authors: Esila Keskin
Abstract: von Economo neurons (VENs) are large bipolar projection neurons found exclusively in the anterior cingulate cortex (ACC) and frontal insula of species with complex social cognition, including humans, great apes, cetaceans, and elephants. Their selective depletion in frontotemporal dementia (FTD) and altered development in autism implicate them in rapid social decision-making, yet no computational model of VEN function has previously existed. We introduce the Fast Lane Hypothesis: VENs implement a biological speed-accuracy tradeoff (SAT) by providing a sparse, fast projection pathway that enables rapid social decisions at the cost of deliberate processing accuracy. We model VENs as fast leaky integrate-and-fire (LIF) neurons (tau=5ms, fan-in=8) alongside standard pyramidal neurons (tau=20ms, fan-in=80) within a spiking cortical circuit (N_pyr=2,000) trained on a social discrimination task. Networks are evaluated under three clinically motivated conditions across 10 independent random seeds: typical (2% VENs), autism-like (0.4% VENs), and FTD-like (post-training VEN ablation). All configurations achieve equivalent asymptotic classification accuracy (~99.4%), consistent with the prediction that VENs modulate decision speed rather than representational capacity. VENs produce median first-spike latencies 4ms earlier than pyramidal neurons, providing direct evidence of a fast signalling pathway. At a fixed decision threshold of theta=3, the typical condition is significantly faster than FTD-like (t=-23.31, p<0.0001), while the autism-like condition is intermediate (mean RT=26.91+/-9.01ms vs. typical 20.70+/-2.02ms; p=0.078, trending). To our knowledge, this is the first computational model that asks what a von Economo neuron actually computes.
Authors: Shuangquan Feng, Laura Fleig, Ruisen Tu, Philip Chi, Edmund Bu, Melinda Ozel, Junhua Ma, Teng Fei, Virginia R. de Sa
Abstract: Large language models (LLMs) enable increasingly capable tutoring-style conversational agents, yet effective tutoring requires sensitivity to learners' affective and cognitive states beyond text alone. Facial expressions provide immediate and practical cues of confusion, frustration, or engagement, but remain underexplored in LLM-driven tutoring. We investigate whether facial-expression-aware signals can improve empathetic tutoring responses through prompt-level integration, without end-to-end retraining. We build a scalable simulated tutoring environment where a student agent exhibits diverse facial behaviors from a large unlabeled human facial expression video dataset, and compare four tutor variants: a text-only LLM baseline, a multimodal baseline using a random facial frame, and two Action Unit estimation model (AUM)-based methods that either inject textual AU descriptions or select a peak-expression frame for visual grounding. Across 960 multi-turn conversations spanning three tutor backbones (GPT-5.1, Claude Opus 4.5, and Gemini 2.5 Pro), we evaluate targeted pairwise comparisons with five human raters and an exhaustive AI evaluator. AU-based conditioning consistently improves empathetic responsiveness to facial expressions across all tutor backbones, while AUM-guided peak-frame selection outperforms random-frame visual input. Textual AU abstraction and peak-frame visual injection show model-dependent advantages. Control analyses show that this improvement does not come at the expense of worse pedagogical clarity or responsiveness to textual cues. Overall, our results show that lightweight, structured facial expression representations can meaningfully enhance empathy in LLM-based tutoring systems with minimal overhead. Future work should evaluate these methods using real student facial data collected during authentic learning interactions.
Authors: Mengfei Lan, Lecheng Zheng, Halil Kilicoglu
Abstract: Effective biomedical information retrieval requires modeling domain semantics and hierarchical relationships among biomedical texts. Existing biomedical generative retrievers build on coarse binary relevance signals, limiting their ability to capture semantic overlap. We propose BioHiCL (Biomedical Retrieval with Hierarchical Multi-Label Contrastive Learning), which leverages hierarchical MeSH annotations to provide structured supervision for multi-label contrastive learning. Our models, BioHiCL-Base (0.1B) and BioHiCL-Large (0.3B), achieve promising performance on biomedical retrieval, sentence similarity, and question answering tasks, while remaining computationally efficient for deployment.
Authors: Benjamin Minhao Chen, Xinyu Xie
Abstract: The project of aligning machine behavior with human values raises a basic problem: whose moral expectations should guide AI decision-making? Much alignment research assumes that the appropriate benchmark is how humans themselves would act in a given situation. Studies of agent-type value forks challenge this assumption by showing that people do not always judge humans and AI systems identically. This paper extends that challenge by examining two further possibilities: first, that evaluations of AI behavior change when its human origins are made visible; and second, that people judge the humans who program AI systems differently from either the machines or the human actors they are compared against. An experiment with 1,002 U.S. adults measured moral judgments in a runaway mine train scenario, varying the subject of evaluation across four conditions: a repairman, a repair robot, a repair robot programmed by company engineers, and company engineers programming a repair robot. We find no significant difference in evaluations of the repairman and the robot. However, judgments shifted substantially when the robot's actions were described as the product of human design. Participants exhibited markedly more deontological, rule-based reasoning when evaluating either the programmed robot or the engineers who programmed it, suggesting that rendering human agency visible activates heightened moral constraints. These findings indicate that people may evaluate humans, AI systems acting in the same situation, and the humans who design them in meaningfully different ways. The fact that these evaluations do not necessarily converge gives rise to the alignment target problem: which normative target should guide the development of artificial moral agents in high-stakes domains, and whether these plural judgments can be reconciled within a coherent account of value alignment.
Authors: Yang Xiao, Huiyuan Chen, Kaiyuan Deng, Chao Jiang, Zinan Ling, Ruimeng Ye, Fei Wang, Xiaolong Ma, Bo Hui
Abstract: We propose \textbf{Compressed Video Aggregator} (CVA), a lightweight micro-video recommendation module that decouples video information from preference learning. CVA first summarizes frozen VFM frame embeddings into a semantic-consensus anchor through masked mean pooling, projects this anchor into a compact latent space, and refines the projected representation with residual self-attention and feedforward blocks before producing a single video embedding for the recommender. Due to the redundancy in the frame count of the original benchmark and its overly coarse sampling, we used titles to re-select key frames based on CLIP. Experiments on MicroLens and Short-Video show consistent gains with orders-of-magnitude reductions in training time and GPU memory, and re-selected frames can further enhance the performance of all methods, including CVA. Furthermore, we also discussed the impact of several scenarios involving erroneous titles on our method.
Authors: Jeffrey Flynt
Abstract: Current LLM memory benchmarks evaluate answer quality rather than retrieval accuracy. Consequently, a system that dumps its entire belief store can achieve perfect recall and mask severe precision failures. We show this evaluation gap persists across multiple embedding models where similarity-based retrieval over domain-specific corpora inherently struggles to isolate target beliefs from semantically proximate ones. Furthermore, multi-turn topic drift compounds this retrieval noise while driving up latency and operational costs. To decouple retrieval quality from generative performance, we introduce PrecisionMemBench, an 89-case benchmark measuring precision, noise isolation, session latency, and belief mutability. We also present Tenure, a structured belief-store proxy that resolves scope and retrieval before inference and injects typed belief state as ambient instruction before the model sees the prompt, removing model-side discretion over whether memory is consulted. Evaluated across 13 configurations, Tenure achieves perfect retrieval passes across all active, non-session, and session test cases. In contrast, the baseline configurations fail to reach even half of the active passes, with precision scores clustering at 0.22 and below. Our results demonstrate that while current memory systems successfully store information, they fail to retrieve it cleanly; a structural vulnerability that traditional answer-quality benchmarks conceal.
Authors: Jingxing Wang, Chenyu Zhou, Zhihui Fu, Jun Wang, Weiwen Liu, Weinan Zhang, Jianghao Lin
Abstract: Additional test-time compute can give LLM agents access to more past experience, yet expanding the context or adding rollouts does not necessarily yield greater agent capability. We call this challenge test-time compute-to-capability conversion and propose SkillTTA, which retrieves task-relevant training trajectories and synthesizes a temporary skill conditioned on the visible target context for a solver with fixed parameters. To pursue a higher performance ceiling, SkillTTA further uses meta prompt optimization (MPO) to adapt the policy that writes these skills. MPO evaluates candidate prompts on paired tasks and emphasizes informative transitions. It also confines updates to benchmark-specific atomic slots, reducing the variance caused by observing each edit only indirectly through skill synthesis and solver rollout. Across ALFWorld, SpreadsheetBench, BigCodeBench, and WebShop, SkillTTA outperforms state-of-the-art reuse and optimization baselines. It attains a higher performance ceiling at lower compute cost than baseline reuse and sampling strategies.
Authors: Paulina Hoyos, Shashanka Ubaru, Dongsung Huh, Vasileios Kalantzis, Kenneth L. Clarkson, Misha Kilmer, Haim Avron, Lior Horesh
Abstract: Symmetry is central to the physical sciences, yet machine learning usually captures it only approximately, leaving a residual per-step equivariance error $\varepsilon$ that compounds with depth $M$ as $M\varepsilon$, whereas exact equivariance holds at unbounded depth; we demonstrate this divergence at fourteen orders of magnitude. We show that a symmetry can be made exact by construction, as the multiplication rule of a tensor algebra. In the resulting $\starG$ algebra, defined by any finite group $G$, the group-Fourier transform block-diagonalizes every tensor into irreducible-representation blocks, making equivariance intrinsic; requiring equivariance conversely \emph{forces} this suitably normalized transform, so the algebra is determined by $G$ rather than chosen. The standard matrix toolbox, including a Frobenius-optimal low-rank factorization, transfers blockwise, machine-checked in Lean~4 under an explicit axiom budget, and extends unchanged to band-limited compact groups and, under periodic boundary conditions, to all 230 crystallographic space groups and the compact little-group fibers of Euclidean and Poincar\'e symmetry. This exactness is an applied capability: on inorganic-crystal elastic tensors the algebra enforces point-group selection rules exactly on the output of \emph{any} predictor, driving a trained graph network's forbidden-channel leakage from $10^{-2}$ to machine zero, eliminating mechanically unstable predictions, and recovering viable materials that an unconstrained screen discards; on molecular data, with no quantum-mechanical input, it exposes octahedral selection-rule signatures consistent with the Wigner--Eckart theorem. Matched networks lead on pooled molecular accuracy, which we report plainly: the contribution is a complementary algebraic calculus, structural and diagnostic, exact at any depth.
Authors: Hang He, Chuhuai Yue, Chengqi Dong, Chengcheng Wan, Ting Su, Haiying Sun, Jiajun Chai, Xiaohan Wang, Guojun Yin
Abstract: Visual DeepSearch tasks require multimodal large language models (MLLMs) to resolve complex visual queries by repeatedly inspecting image regions, grounding reasoning in visual evidence, and connecting fine-grained clues across multiple steps. However, existing benchmarks primarily evaluate single-step visual understanding or isolated visual-query response generation. They have limited difficulty, limited search horizons, and single-pass image inspection, and thus fail to evaluate models' ability to iteratively revisit visual evidence and reason across multiple steps. In this work, we introduce VistaHop, a benchmark designed specifically to evaluate Visual DeepSearch. It evaluates repeated image inspection, visual-anchor grounding, and long-horizon evidence traversal across different visual regions. VistaHop comprises 600 images, 25 visual search scenarios, and 600 Visual DeepSearch tasks. We also propose VistaArena, a unified evaluation framework that supports tool-based interactions, including visual retrieval, image inspection, and evidence-grounded reasoning. Experiments show that even state-of-the-art MLRMs remain far from solving VistaHop, with the best-performing model, SenseNova-MARS-32B, achieving only 26.33% Pass@1. These findings highlight the importance of specialized benchmarks and improved agentic methods for Visual DeepSearch.
Authors: Ossi Lehtinen
Abstract: Transformers consuming multi-channel scalar signals must embed $C$ simultaneous values into one $d_{\text{model}}$-dimensional vector per time step. We audit eight input encoders -- a shared-scalar baseline, per-channel linear projections, an orthogonality regulariser, a nonlinear MLP, block-partitioned concatenation, channel-independent and channel-as-token architectures, and a projected positional encoding -- on a synthetic benchmark where channel identity is informative and on ETTh1, scored by next-step negative log-likelihood. The headline is practical near-equivalence within a wide "top tier": the standard per-channel linear projection matches every alternative up to small, statistically real but practically modest differences. A direct geometric probe attributes this to a spontaneous orthogonalisation of the per-channel projections: with no explicit regulariser they tighten well beyond the near-orthogonality of random initialisation, letting the standard linear recover channel identity from the summed embedding. Two encoders lose decisively: the shared-scalar baseline collapses for information-theoretic reasons we make explicit, and the channel-independent PatchTST-spirit baseline overfits universally on the synthetic benchmark and underperforms on both. Paired tests resolve two small gaps: projecting the sinusoidal positional encoding through a learned linear layer edges the rest at small $C$ by extending this orthogonality to the positional subspace; a nonlinear MLP stem edges them at the largest $C$, with the gap shrinking under more training data. The practical recommendation: use the standard per-channel linear projection by default; reach for something more elaborate only when the task calls for it.
Authors: Peter Halmos, Boris Hanin
Abstract: We exhibit an exact correspondence between sampling with score-based diffusion models and adiabatic transport of ground states for a family of Schr\"odinger operators we call Score Hamiltonians, built from the learned score's quantum potential. We obtain novel density reconstruction bounds and principled annealing schedules via adiabatic theorems for Fokker-Planck equations with time-varying potentials. We find the fundamental limit of sampling is set by the ratio of squared score-matching error to Score Hamiltonian spectral gap - the inverse Poincar\'e constant of the data density.
Authors: Eric Spencer, Arslan Bisharat, Brian Ortiz, Khushboo Bhadauria, Mujtaba Nazari, TaiNing Wang, George K. Thiruvathukal, Konstantin Laufer, Mohammed Abuhamad
Abstract: TLA+ is a formal specification language for verifying distributed systems and safety-critical protocols. Large language models (LLMs) frequently produce TLA+ specifications that fail the TLC model checker for semantic reasons. Across 25 LLMs, the best public baseline is 26.6% syntactic parse and 8.6% semantic model-check. We present TLA-Prover, a 20-billion-parameter model for TLA+ specification synthesis. Training combines supervised fine-tuning (SFT) on verified examples with repair-based group-relative policy optimization (GRPO). In the GRPO stage, the model learns to fix its own rejected specifications. We also train a direct preference optimization (DPO) variant from the same SFT checkpoint as an ablation. TLC provides the reward signal directly, with no learned reward model. Four tiers grade each output: Bronze (parses), Silver (no warnings), Gold (passes TLC), and Diamond. To reach Diamond, the model's correctness property is automatically altered in a small way; TLC must then detect a violation. If TLC still passes, the property was always-true and contributes nothing; the output fails Diamond. TLA-Prover reaches 9/30 (i.e. pass@1 = 30%) at both Gold and Diamond on a held-out 30-problem benchmark. This is roughly 3.5x the 8.6% untuned baseline. The DPO variant reaches 20% at Diamond. Gold and Diamond coincide at every checkpoint; this prevents the trivial-property failure mode.
Authors: Junyu Zhou, Puyu Wang, Dennis Wagner, Yunwen Lei, Marius Kloft, Yiming Ying
Abstract: Characterizing the optimization dynamics and statistical performance of over-parameterized deep neural networks (DNNs) remains a central challenge in understanding the remarkable success of deep learning. We establish quantitative bounds showing that kernel gradient descent in the reproducing kernel Hilbert space induced by the deterministic infinite-width neural tangent kernel approximates finite-width deep regression with smooth activations under gradient descent (GD) and stochastic gradient descent (SGD) training. The approximation gap is governed by the network width and training horizon, with an additional stochastic gradient error in the SGD case. This connection provides a general mechanism for transferring learning-theoretic guarantees from kernel methods to deep regression. As an application, under general source and effective dimension conditions, we show that both GD- and SGD-trained DNNs attain the minimax-optimal excess population risk rate, up to logarithmic factors, provided that the network width grows polynomially in the sample size. To the best of our knowledge, these are the first such guarantees for standard fully connected deep neural networks with smooth activations trained by GD and SGD.
Authors: Hongkyu Koh, Ikbeom Jang
Abstract: Electrocardiogram (ECG) classification models often suffer from severe label scarcity, making semi-supervised learning (SSL) an attractive strategy for reducing annotation costs. In clinical settings, however, unlabeled pools frequently contain out-of-distribution (OOD) anomalies or diagnostic groups absent from the labeled set. Standard SSL forces incorrect pseudo-labels onto these unseen classes, producing overconfident predictions. To address this, we propose SafeECGMatch, a calibration-aware safe SSL framework for single-label ECG classification under label distribution mismatch. Methodologically, SafeECGMatch employs a dual-branch architecture extracting time-frequency latent representations via ECG-specific augmentations. Crucially, it dynamically aligns confidence with empirical accuracy through adaptive label smoothing and temperature scaling, calibrating both the multiclass classifier and the OOD detector across temporal and spectral domains. This joint optimization allows trustworthy OOD rejection and reliable pseudo-labeling. Evaluated on the PTB-XL and PhysioNet/CinC Challenge benchmarks, SafeECGMatch achieves state-of-the-art accuracy and calibration, advancing reliable knowledge discovery in physiological time-series. Code is available at https://github.com/labhai/SafeECGMatch.
Authors: Enhan Zhao, Wei Wu, Yuanrui Zhang, Xueliang Zhao, Di He
Abstract: Spatial reasoning remains a persistent challenge for multimodal large language models (MLLMs). Existing approaches largely rely on large-scale, statically curated datasets, where all training samples are treated uniformly regardless of the model's evolving capabilities. This static paradigm is inherently data-inefficient: training capacity is often spent on samples that are either trivial or overly difficult for the model at its current stage. To address this limitation, we propose Ouroboros-Spatial, a self-evolving training framework in which the model plays dual roles as a proposer and a solver. In each iteration, a frozen proposer generates spatial question-answer (QA) pairs from 3D scene metadata and raw video frames, together with executable code for deriving reliable ground truth. A learnable solver is then fine-tuned on the accepted samples, and its per-sample prediction confidence is used as a difficulty signal. This signal is fed back to the proposer in the next iteration, guiding it to generate questions better matched to the solver's current capabilities. Through this closed-loop design, the training distribution co-evolves with model ability, reducing redundant trivial examples while filtering out ambiguous or uninformative samples with limited learning value. Across six spatial reasoning benchmarks, Ouroboros-Spatial substantially improves Qwen3-VL-4B and Qwen3-VL-8B while using an order of magnitude fewer training examples than recent large-scale curated datasets. On VSI-Bench, it yields absolute gains of 9.9 and 6.8 points for the 4B and 8B models, respectively, enabling both to outperform a wide range of strong open-source and proprietary baselines.
Authors: Saehun Chun, Wonje Choi, Sera Choi, Sanghyun Ahn, Honguk Woo
Abstract: Code-writing large language models (CodeLLMs) generate executable code policies for embodied agents by translating natural language goals and environmental constraints into structured control programs. However, policy generation in open-domain embodied environments suffers from two fundamental limitations: (i) delayed decoding caused by repetitive prefill computation over long prompts, and (ii) limited robustness due to fully generative decoding, which often produces API mismatches, missing safety guards, and unstable control logic. To address these limitations, we present FCGraft, a Functional Cache Grafting framework. FCGraft maintains a library of function-level validated code skeletons and their associated prompt-level Transformer key-value (KV) caches, and synthesizes new policies by retrieving relevant functions and grafting their KV caches when a new task is provided. Given retrieved function caches, FCGraft performs cache grafting via stitching, which composes cached function segments into a composite policy, and patching, which locally adapts only the necessary code regions to satisfy task-specific parameters and constraints with minimal additional decoding. By eliminating redundant prefill computation, this approach reduces generation latency, while reusing validated control structures improves robustness over prompt-level caching methods RAGCache, achieving 18.31% higher task success rate and 2.3x faster policy synthesis.
Authors: Xin Gao, Xingming Xu
Abstract: Transformer architectures form the foundation of modern natural language processing, making it crucial to address the efficiency and scalability limitations of the standard QKV attention mechanism. The Key-Value (KV) cache is a major bottleneck during long-context inference. We propose Keyless Attention, a novel attention mechanism that introduces a dedicated value-space routing projection to replace the conventional key projection, thereby eliminating key representations from the attention computation. This design yields a Value-Only Cache that reduces KV-cache memory and access overhead by 50% compared with standard attention while improving decode throughput. Experiments across five models and four architectures show that Keyless Attention matches or outperforms standard QKV attention in perplexity on four of five models. Furthermore, it achieves competitive performance on downstream evaluation benchmarks while consistently reducing KV-cache memory by 50%. Ablation studies on attention factorization depth further demonstrate that the dedicated value-space routing projection is a key component of the proposed architecture, with Keyless Attention outperforming KV-sharing methods that do not employ this routing mechanism.
Authors: Anzhe Xie, Weihang Su, Jiaxin Mao, Yiqun Liu, Min Zhang, Shaoping Ma, Qingyao Ai
Abstract: Large language models have shown strong fluency in scientific writing, yet the evaluation of related work generation (RWG) remains limited. Existing RWG evaluations largely inherit summarization-oriented metrics, using lexical or semantic similarity to reference sections as proxies for quality. However, related work writing is fundamentally a citation-level scholarly positioning task: it requires selecting, organizing, and framing prior work to clarify how a target paper relates to, differs from, and contributes beyond existing research.As a result, models may generate coherent and semantically-relevant text while exhibiting academically critical failures, such as inappropriate citation selection or misplaced references, that conventional metrics do not capture.To this end, we introduce \textbf{RWGBench}, a benchmark that evaluates RWG from the perspective of citation decision-making rather than text similarity. RWGBench is constructed from a large-scale collection of 40,108 computer science papers and a retrieval corpus of 1.09 million documents, with a carefully curated test set comprising 100 papers and their corresponding published related work sections.We propose a multi-dimensional evaluation framework that assesses citation selection, contextual appropriateness, organization, and discourse structure.Experiments reveal systematic limitations in current systems that are obscured by standard evaluations, while Oracle studies further disentangle retrieval-level and generation-level bottlenecks. Human evaluation further shows that our citation-centric metrics align substantially better with expert judgment than surface-level text metrics. RWGBench offers a citation-centric testbed for developing and evaluating related work generation systems that are better aligned with scholarly writing practices.
Authors: Tanel P\"arnamaa, Martin Lumiste, Ardi Loot, Evgenii Indenbom, Andrei Znobishchev, Ando Saabas
Abstract: Neural video codecs have surpassed classical codecs in coding efficiency but remain impractical for deployment due to cross-platform incompatibility and high computational cost. Existing quantization-based solutions fail to produce deterministic results across diverse hardware platforms, leading to catastrophic decoding failures. We introduce MLVC, a hardware-robust neural video codec designed for practical cross-platform inference. The key idea is to explicitly transmit scale parameters through the hyperprior, which guarantees entropy coding consistency across devices without requiring bit-exact arithmetic. While this increases bitrate overhead, we recover most of the coding efficiency through architectural improvements (gated memory, ReGLU activation), a long-term reference recovery mechanism, and domain-specific perceptual training. On the VCD video conferencing benchmark, MLVC achieves >70% BD-rate (MOS) improvement over hardware HEVC, the strongest deployable baseline, while reaching subjective quality competitive with DCVC-RT, which cannot operate across diverse platforms. Both the encoder and decoder run at 100 FPS on average on commodity NPUs from Apple, Intel, and Qualcomm. MLVC is the first neural video codec to combine competitive compression performance, real-time speed, and cross-platform robustness across diverse consumer devices, making it suitable for widespread deployment. Code is available at https://github.com/microsoft/mlvc.
Authors: Brett Reynolds
Abstract: Safety evaluations for language models increasingly depend on judgments about ambiguous natural-language behaviour: whether a model followed an instruction, refused appropriately, complied with a policy, or misreported progress in an agentic task. Existing benchmarks compress these into pass/fail labels, obscuring whether failures reflect capability limits, policy ambiguity, instruction conflict, scaffold failure, or unstable evaluator judgments. Adversarial pragmatics is safety-relevant model behaviour under instruction conflict, embedded commands, quotation, scope ambiguity, deixis, and indirect speech acts. It's designed to extend to multi-turn agent transcripts, but the seed set represents that family with a single-turn tool-result contrast. This paper introduces a diagnostic framework, an 18-item seed benchmark, a 54-row pilot, and a six-cell LLM-judge assessment, with a protocol keeping task success, policy compliance, risk, refusal, attribution, and confidence analytically separate. The benchmark separates four inference targets a single label can conceal: the regime-relative reference, configured-system behaviour, evaluator-output interpretation, and taxonomic assignment. Its intended use is diagnosis, not deployment certification, vendor ranking, or a general safety score. A first LLM judge that graded its own outputs with the expected answer visible missed the safety-relevant minority classes. Item-clustered intervals leave four of six chance-corrected agreement statistics unable to rule out a constant labeller, and hierarchical pooling shrinks the one eye-catching rubric effect toward the group mean and widens its interval through zero. Rejudging the objects across three judge models and two information conditions leaves the pattern intact: no cell recovers more than two of eleven partial successes, and the strongest cell's edge comes partly from never using that label.
Authors: Dekun Yang
Abstract: Count-based F1 is widely used as a proxy for LLM error-detection quality, but this paper shows that it can rise dramatically without a corresponding improvement in span localization, a gap termed F1 Inflation. The paper introduces ErrorBench, a controlled stress-test protocol for prompt-induced count distortion. ErrorBench evaluates six contemporary LLMs under five prompt conditions over 4,290 responses from 143 CoNLL-2014 passages. Under CoNLL-2014 M2-style scoring, anchored prompts produce up to 0.79 points of F1 Inflation, and up to 0.96 under strict matching. A 100-passage replication using the official ERRANT 3.0.0 pipeline and multi-reference scoring reproduces the pattern: averaged over six models, the Blind-to-Anchored prompt shift raises Count-F1 by +0.21 while raising multi-reference ERRANT F0.5 by only +0.04. The study finds larger count responses in highly instruction-compliant GPT/Claude systems and smaller responses in the Gemini family under this stress-test protocol. The findings suggest that LLM proofreading and document-review evaluations should avoid pre-populated error counts and should report span-aware metrics alongside count-based metrics.
Authors: Taras Kutsyk, Bartosz Zieli\'nski
Abstract: Fine-tuning can give a language model a hidden behavior--it may give false answers under a narrow condition, or give harmful advice only when a prompt touches a particular topic. We introduce the Stabilized Adapter for self-Report (SAR), a lightweight LoRA adapter that makes a fine-tuned model describe its own hidden behavior in plain language, using only the model and the dataset it was trained on. Across seven implanted behaviors, SAR detects the hidden behavior in every one--even when the model has generalized into broad misalignment that the training data alone does not predict. Introspection Adapters (IA), the closest existing baseline, detects some behaviors from our suite but misses others entirely--and where it misses, it hallucinates, consistently reporting wrong behaviors. SAR retains positive signal on every setting where IA fails and roughly halves the rate of hallucinations. This gives practitioners a more reliable tool to audit a fine-tuned model and answer ``what did it actually learn?'' type of questions.
Authors: Zihan Zhang, Xize Cheng, Wenhao Yan, Tong Zhang, Dongjie Fu, Boyun Zhang, Yongbo He, Tao Jin
Abstract: Large Audio-Language Models (LALMs) reason fluently about sound yet struggle to localize precisely when events occur, while classical Sound Event Detection attains frame-level precision only over a closed label set. At the intersection of these paradigms lies the task of Open-Vocabulary Audio Event Grounding: predicting all time intervals of a target sound event described by an arbitrary natural language query. Progress is bottlenecked by data scarcity: no large-scale resource provides open-vocabulary onset/offset supervision, and manual temporal annotation is prohibitively expensive. To address this, we introduce Auto-AEG, a scalable pipeline that constructs such supervision by automatic data construction and model fine-tuning. It pairs programmatically synthesized clips, which carry placement-exact ground-truth intervals for supervised cold-start, with multi-model pseudo-labels on real-world audio that supply the reward signal for reinforcement learning. Training with this pipeline yields large temporal-localization gains (+73.9% and +23.1% mIoU over zero-shot) on AEGBench, an independent difficulty-stratified benchmark we release, and these gains generalize to held-out SED and other audio grounding benchmarks. Our results show that automatically constructed data, coupled with interval-aware reward design, provides an effective data-side route to expanding the temporal localization capability of LALMs. AEGBench: https://huggingface.co/datasets/zihan-audio/AEGBench
Authors: Meng Du, Hongchang Chen, Ran Li, Junjie Zhang, Qi Ouyang, Shibo Zhang, Shuxin Liu
Abstract: Rapid advances in AI video generation pose increasing security risks and call for reliable detectors with strong cross-domain generalization. Although existing methods perform well under in-domain evaluation, their performance degrades substantially on unseen generators. A key reason is shortcut learning, where detectors rely on domain-specific bias rather than intrinsic forensic cues. To address this issue, we propose G2VD, a generalizable AI-generated video detection framework based on counterfactual intervention and causal disentanglement. First, G2VD introduces a counterfactual intervention pipeline (CFIPipeline) that constructs counterfactual samples through VAE-based reconstruction and subsequent frequency-domain and pixel-domain alignment, thereby weakening spurious correlations between domain-specific bias and authenticity labels. Building on this intervention, we further design a causal disentanglement classifier that combines two domain-anchored branches with complementary objectives and a constraint based on the Hilbert-Schmidt Independence Criterion (HSIC), encouraging the causal and non-causal representations to capture intrinsic forensic cues and domain-specific bias, respectively. Experiments across four public datasets demonstrate strong cross-domain performance and consistent gains over baseline methods. In the challenging GenVidBench setting, G2VD achieves over 90\% overall ACC, with improvements of 0.194 in F1 and 0.104 in AUC over comparable state-of-the-art methods, while using only 10\% of the available training data. Code is available at https://github.com/DMOSCAR-98/G2VD.
Authors: Sharayu N. Deshmukh, Md Rashidunnabi, Nelton Tiago Gemo, Kurundkar G. D., Mahamune M. R., Nilesh K. Deshmukh
Abstract: Deepfake image detection is served by three fundamentally different paradigms - commercial APIs, zero-shot vision-language models (LLMs), and open-source detectors - that are rarely evaluated under a common protocol, making direct comparison difficult. We introduce VendorBench-100, a cross-paradigm benchmark that evaluates 36 representative models using a single adversarial 100-image corpus, a unified output schema, and a common evaluation framework. Models are ranked primarily by the Matthews correlation coefficient (MCC), with ROC-AUC reported as a threshold-independent measure of ranking ability. Rather than maximizing size, it emphasizes real-world difficulty through a taxonomy of eight edge-case families such as face swaps, text-to-video stills, AI photo edits, avatar compositing, opaque-provenance images, and compressed research frames. Commercial APIs achieve the strongest median performance, followed by vision LLMs and open-source detectors, though individual open-source models remain competitive with the best LLMs. Across all 36 models, MCC and ROC-AUC are strongly correlated (Pearson r ~ 0.86); the more consequential finding is narrower and one-directional: a subset of otherwise strong rankers are miscalibrated at their shipped default threshold, so a high ROC-AUC can overstate real-world deployability. Separately, raw accuracy and F1 are unreliable on this corpus's imbalanced class split, since a model that predicts "fake" indiscriminately scores deceptively well on both while offering no real discriminative skill. No single metric is safe in isolation: MCC and specificity should always accompany ROC-AUC and accuracy. We release the complete evaluation framework and results. Code and data: https://github.com/sharayu-20/vendorbench-100
Authors: Antonio San Martin, Catherine Trekker
Abstract: This paper proposes a human-centered artificial intelligence (HCAI) framework for AI-assisted lexicography. While generative AI offers significant opportunities to enhance lexicographic work, it also raises concerns regarding the future role of lexicographers and the preservation of linguistic and cultural diversity. Drawing on HCAI principles and previous applications in other language professions, the paper identifies four interrelated dimensions through which AI integration in lexicography can be understood and critically examined: the augmented lexicographer, the sociotechnical context of AI integration, bias, and the design of AI-powered lexicographic tools. The framework argues that AI should augment rather than replace lexicographers, combining automation with meaningful human control. It further emphasizes the importance of preserving professional agency, mitigating AI-generated biases, and designing tools around the needs of lexicographers. By doing so, the paper provides a foundation for future research and the beneficial integration of AI into lexicographic workflows.
Authors: Zhen Yin, Wenkang An, Hao Wang, Keran You
Abstract: Cross-version differencing of scientific documents is essential in scholarly publishing and technical documentation, but remains challenging because scientific documents are page-structured artifacts containing heterogeneous elements such as text, tables, formulas, figures, and layout cues. Existing text-sequence-based methods often lose layout and structural information, while image-based methods lack semantic interpretability and are sensitive to rendering variation. To address these limitations, this paper proposes a layout-aware heterogeneous element-aware framework for scientific document differencing. The framework decomposes document versions into semantically typed elements, establishes cross-version correspondence through an alignment-first mechanism that jointly models spatial, content, and structural compatibility, and performs type-aware difference reasoning over aligned element pairs. It supports unified change detection, localization, structure-awareness analysis, and alignment/matching evaluation across text, tables, formulas, and figures. Experiments on real-world scientific PDF data from journal production proofreading workflows show that the proposed framework consistently outperforms element-specific baselines. It achieves detection F1 scores of 0.903, 0.855, 0.862, and 0.845 for text, tables, formulas, and figures, respectively, with further improvements in localization, structure awareness, and matching quality. Ablation and sensitivity analyses confirm the effectiveness of cross-version alignment, type-specific representations, structure-aware reasoning, and compatibility-weight design. These results demonstrate that heterogeneous element-aware differencing provides a robust and interpretable solution for scientific document comparison in realistic editorial production scenarios.
Authors: Mert Cemri, Andrei Cojocaru, Melissa Pan, Shu Liu, Shubham Agarwal, Alexander Krentsel, Jay Tang, Kannan Ramchandran, Joseph E. Gonzalez, Matei Zaharia, Alex Dimakis, Ion Stoica
Abstract: An agent system's execution traces record how it fails, and procedures that improve such a system without changing model weights (trajectory selection, prompt and workflow optimization, runtime monitoring) read these traces for feedback. Yet raw traces are a poor medium for accumulating that feedback: long, instance-specific, and lacking a stable vocabulary for recurring failures. We argue that an agent system should instead maintain an explicit representation of how it fails, induced from its own behavior and reusable wherever failure feedback is needed. AdaMAST builds this representation by converting a target system's traces into a compact, evidence-grounded failure taxonomy: named failure codes organized along three fixed axes (system-level, role-specific, and domain-specific), with every name, definition, and evidence pattern induced from the traces; no code is hand-authored, no trace human-annotated. The taxonomy is not merely a post-hoc diagnostic but a shared feedback interface, improving agents in three ways. In agent-system search, taxonomy-coded diagnoses of failed candidates outperform free-form reflection on all five benchmarks we test. At runtime, taxonomy feedback raises SWE-agent's resolution on SWE-bench Verified Mini from 60% with free-text reflection to 70%, and improves Claude Code from 64.0% to 70.7% as a runtime skill. In trajectory selection, AdaMAST-Judge, a verifier built on the induced codes, improves best-of-5 accuracy on Terminal-Bench 2.0 by 8-15 points over Pass@1. The vocabulary itself is compact (an order-of-magnitude compression that preserves trace distinctions), human-faithful (matching expert failure annotations more closely than a hand-crafted reference vocabulary), and adaptive (taxonomies induced for different domains share few codes). Adaptive failure taxonomies close the loop between the traces agents produce and the procedures that improve them.
Authors: Jingjie Ning, Xiaochuan Li, Shanshan Zhong, Ji Zeng, Guolin Ke
Abstract: Auto Research uses language-model agents to propose, implement, and evaluate machine-learning changes in a closed loop, but is usually judged by its terminal pipeline. A terminal score cannot reveal which technical decision produced a gain or distinguish a reusable discovery from a change adapted to development feedback. We introduce intervention-centered Auto Research, which validates research decisions rather than only final artifacts and makes their reliability measurable. Feature, Model, Representation, and Data axes are searched independently with inner five-fold feedback. Each axis winner is frozen before an outer-holdout matrix compares all alternatives on evidence the loop never sees. Across 701 agent-executed attempts spanning ten Matbench endpoints, outer evidence confirms the selected intervention on nine of ten endpoints and preserves 89.3\% of non-tied intervention orderings. It also rejects an aggregate Representation gain that inner feedback endorsed. The resulting matrix reveals an information-dependent hierarchy. Composition-only tasks support several routes to improvement, whereas structure-informed tasks favor local geometry features and complementary tree ensembles. A subsequent compatibility test combines already frozen Feature and Model code without further search or tuning and raises mean outer-holdout improvement from 19.0\% to 26.3\%. By validating decisions rather than only artifacts, this design turns adaptive search into reusable evidence wherever agents propose executable alternatives against a fixed evaluator.
Authors: Shufan Chai, Liangliang Sun, Jessica Staddon
Abstract: Large language models (LLMs) are increasingly used for a range of software, hardware and human-centered security tasks. Consequently, LLM performance on security tasks is an active area of measurement and research, often with a focus on identifying areas in which LLM security "knowledge" may be insufficient. Popular strategies for identifying LLM security knowledge gaps include building corpora of challenge questions or task benchmarks, strategies that require substantial manual work and security expertise to design and execute. We introduce a partially-automated method for assessing LLM knowledge of a security area. The method uses authoritative information from Consumer Protection Agencies (CPAs) to identify instability in LLM responses that can be indicative of knowledge gaps. We demonstrate the method for 2 security topics, identity theft and impostor scams, and 5 LLMs in 2 leading LLM families, Gemini and GPT, using publicly available information about identity theft and impostor scams from 6 CPAs. The method distinguishes between models that have and don't have sufficient knowledge to accurately identify the security topics in text narratives.
Authors: Yunjie Chen, Xiaoxin Chen, Fang Wang
Abstract: Large-scale online reinforcement learning (RL) is the predominant means of eliciting advanced abilities including long-term reasoning and agentic tool use in large language models (LLMs). However, continuing to scale it across vast task domains of interest remains challenging in both computational infrastructure and cost, especially when considering RL as merely a one-off learning stage. Recently, a widely used technique for distilling knowledge across various domains and training stages, multi-teacher on-policy distillation (MOPD), helps to decouple the RL stage, saving costs, while maintaining generality across vast domains. Nonetheless, similar to online RL, MOPD requires coupled inference and backward passes, which continues to limit its scalability and computational efficiency. To address these challenges, we propose REGEN: Replay-recycling for Expert-to-Generalist Distillation with Offline RL. Instead of distilling from multiple teacher models, REGEN trains a generalist by simply recycling the replay memory -- the free by-product of the teachers' specialized RL training -- and employing offline RL algorithms. REGEN completely decouples the rollout sampling from the backward training process and thus greatly reduces the training cost. Across mathematical reasoning, code generation, and instruction following, REGEN matches the accuracy of MOPD at substantially lower cost. It potentially turns online RL into a data synthesis process instead of a one-off learning stage, and can be extended to large-scale post-training without requiring heavy computational load. Code is available at https://github.com/yunjie-sysu/REGEN.
Authors: Junyu Dai, Xinyue Fan, Weiqin Li, Xiangang Li, Yunjia Li, Bin Ma, Yukun Ma, Chongjia Ni, Yufei Shi, Biao Tian, Haoxu Wang, Menglin Wu, Jianwei Yu, Huaicheng Zhang, Han Zhao, Shengkui Zhao, Haina Zhu
Abstract: In this report, we present a unified song generation framework capable of producing high-quality full-length music from lyrics, text descriptions, and musical attributes. The proposed framework supports three tasks: Lyrics-to-Song Generation, which generates complete songs from text descriptions, lyrics, and musical attributes; Instrumental Music Generation, which creates music without vocals; and Cover Song Generation, which reinterprets existing songs with different styles while preserving their melodic content. Architecturally, our system consists of four main components: a semantic-aware tokenizer, hybird-LM, FullDiT, and a two-level melody module. The tokenizer encodes audio into 8-codebook RVQ tokens for efficient discrete music representation. Based on these tokens, hybird-LM performs hierarchical autoregressive audio-token modeling for full-song generation. To improve audio fidelity, FullDiT performs full-song flow matching in a continuous VAE latent space conditioned on codec tokens, lyrics, and text captions. For cover song generation, the melody module extracts and discretizes melody cues from reference audio to guide generation while preserving the original melodic content. Finally, we investigate DPO, GRPO, and OPD as reward-based post-training strategies for hybird-LM and apply flow-based GRPO to FullDiT to improve musicality and rendering quality. Experimental results on a multilingual automatic benchmark, complemented by the Artificial Analysis Music with Vocals leaderboard, show that the proposed framework achieves competitive performance in the evaluated settings.
Authors: Manoosh Samiei, Doina Precup, Paul Masset
Abstract: Effective decision-making in complex and changing environments requires balancing short-term and long-term consequences. In reinforcement learning (RL), this trade-off is typically controlled through a fixed discount factor, which imposes a single exponentially discounted temporal horizon. However, biological agents exhibit flexible and adaptive temporal discounting, suggesting that effective planning requires multiple timescales. Here, we propose a multi-horizon approach that adaptively selects and combines temporal horizons, enabling robust adaptation to changes in reward structure without manual discount-factor tuning. This flexibility makes the method particularly suitable for continual learning scenarios involving task switches and varying environmental configurations. Empirically, we demonstrate that our approach identifies effective discount factors across a range of MiniGrid environments, including continual settings composed of three sequentially changing tasks. These results suggest that adaptive temporal discounting can improve parameter efficiency and enhance adaptability in both artificial and biologically inspired learning systems.
Authors: Taegun An, Dohun kim, Haebeom Lee, Changhee Joo
Abstract: Logic Gate Networks (LGNs) compute through compositions of Boolean operations, yet existing LGNs do not reliably benefit from increased depth. We identify two causes: optimization collapse and topology-induced degradation of output-specific credit that persists even after skip-biased initialization and straight-through estimation stabilize training. We introduce Input-Anchored Logic Gate Networks (IALGNs), in which each gate combines a private hidden spine with a direct input anchor. This topology prevents output-path merging while retaining input access at every layer. Credit diagnostics show that random wiring dilutes or conflicts output-specific gradients, whereas IALGN maintains usable and coherent credit. Random-$k_x$ relaxation improves anchor selection without relaxing the spine. Across MNIST, CIFAR-10, and CIFAR-100, IALGN exhibits consistent fixed-width depth--accuracy scaling up to 150 layers, while alternative topologies saturate or degrade. Linear probes, topology ablations, and operation-aware analysis show that trained IALGNs preserve private states and apply sparse anchor-conditioned updates. These results indicate that scalable LGN depth requires both stable optimization and credit-preserving information access.
Authors: Jie Zhang
Abstract: Synthetic tabular data is prized for preserving not just each column's marginal distribution but the dependencies between columns - structure that carries much of the discriminative signal for minority classes in imbalanced domains such as fraud detection and clinical risk. Yet the metrics most commonly used to certify such data are largely blind to this structure: a fully-factorized baseline that destroys all inter-column dependency is judged nearly indistinguishable from real data by the logistic-regression C2ST, while the pairwise Trend score is only partially sensitive. We introduce a dependency-aware fidelity diagnostic that decomposes a strong gradient-boosted classifier two-sample test (XGB-C2ST) into marginal, dependency, and numerical-categorical cross components, anchored between a fully-factorized lower bound and a real-data oracle upper bound. Applying it to state-of-the-art flow-matching (TabbyFlow/EF-VFM) and diffusion (TabDiff) generators, we expose a real dependency gap that standard metrics miss and that tracks minority-class utility. The gap is not a structural limitation of mean-field objectives - consistent with recent recovery results, these are asymptotically exact - yet it is stubborn: neither a 16x capacity increase nor post-hoc copula correction closes it, pointing to the absence of direct dependency supervision rather than a capacity or structural limit.
Authors: Chen Wang, Boming Kang, Qinghua Cui
Abstract: Protein language models learn transferable sequence representations. However, because they primarily model contextual dependencies along amino-acid sequences, their training objectives do not explicitly constrain the model to learn three-dimensional residue contacts formed after folding . Here, we introduce LC-SEPLM (Long-range Contact-supervised ESM Protein Language Model), which adapts ESM2 with LoRA and long-range residue-pair contact supervision while retaining sequence-only downstream inference. Pair-specific queries use cross-attention over the complete sequence to extract global sequence context associated with long-range spatial contacts. To expose the model to diverse structural information, we trained LC-SEPLM on 500,000 AlphaFold Swiss-Prot proteins. In downstream evaluation, LC-SEPLM improved all eight protein-level tasks relative to ESM2. The largest gain occurred in remote-homology recognition, where macro-F1 increased from 0.6122 to 0.6769 (+0.0647, or 6.47 percentage points). On the official ESM-S EC benchmark, LC-SEPLM also outperformed ESM-S with a maximum absolute gain of 0.1771. These results support residue-pair contact supervision as a bounded route for introducing structural information into protein sequence representations while preserving sequence-only inference.
Authors: Xin Wang (Jeff), R. Tyrrell Rockafellar (Jeff), Xuegang (Jeff), Ban
Abstract: As constrained learning becomes increasingly common, models are trained under explicit feasibility requirements to enforce fairness, safety, robustness, regulariza- tion, and physics or logic constraints. Understanding how training samples in- fluence the model solution (e.g., learned parameters) is crucial for interpretability and robustness. The classical influence function (IF) estimates sample contribu- tions via local sensitivity analysis, measuring how the solution changes when a specific training sample is perturbed or removed. However, IF becomes unreli- able in constrained settings: data perturbations can reshape both the objective and the feasible region, leading to estimates that violate feasibility. In response, we propose the Directional Influence Function (DIF), a novel estimator that explicitly incorporates these constraints into influence estimation. DIF formulates the opti- mality conditions of constrained learning as a variational inequality (VI) and ana- lyzes how perturbing training data affects this VI. We validate DIF on constrained linear regression and demonstrate that it recovers leave-one-out retraining results, whereas IF and penalty-based IF exhibit significant bias. We further apply DIF to fairness-constrained CNNs, where DIF accurately predicts test loss changes under data removal and aligns closely with actual retraining. Our results establish DIF as an efficient and reliable tool for data attribution in constrained learning.
Authors: Fabio Aurelio D'Asaro
Abstract: FastLAS is a scalable system for Inductive Logic Programming (ILP): you give it some background knowledge, a language bias, and a set of examples, and it searches for a set of logic program rules (a hypothesis) that explains the examples. These notes are a hands-on introduction to writing FastLAS programs. They are organised as a programmer's guide: syntax first, then a ladder of worked, numbered examples of increasing difficulty. Every self-contained example here has been run against FastLAS 2.2.0 and shows the tool's actual output. We keep theory to the minimum needed to write correct programs; throughout, set-off notes flag where FastLAS differs from its sibling system ILASP, and where the two learning algorithms (--opl and --nopl) behave differently. The document is intended as an unofficial tutorial to FastLAS 2.2.0, not as an official language specification.
Authors: Donghao Fu, Jingxin Li, Xue Jiang, Yihong Dong
Abstract: Third-party API routers have become a common layer that unifies access across increasingly diverse LLM providers. In coding-agent workflows, high-autonomy operation is widely adopted because it reduces interaction overhead. As a result, a third-party API router, which sits between the agent and the upstream provider, inevitably occupies the trusted path. It can inspect and modify every request and response, yet no mechanism verifies alignment between the provider's output and the repository-level actions ultimately executed by the agent. Consequently, client-side permission mechanisms may become ineffective in practice. Whether this control gap produces real, hard-to-detect effects on software development tasks remains empirically unmeasured. In this paper, we conduct an empirical study of router-side injection in coding agents, examining four intervention levels of increasing subtlety: Response Substitution (L1), Response Append (L2), LLM-Polished Injection (L3), and LLM-Polished with Distribution Alignment Injection (L4). Moreover, we develop SIDEL, a framework for trace recording, replay, injection, and defense evaluation, with a curated dataset of 400 samples. We evaluate four representative coding agents, and further evaluate whitelist-based execution control and LLM review. Router-side intervention substantially alters repository-level actions and remains difficult for existing client-side safeguards to detect. Without additional mitigations, all evaluated agents achieved a defense success rate of 0 percent across all injection levels. Client-side mitigations and reactive reviews improve resistance but do not fully restore end-to-end control, motivating provider-side output-integrity guarantees. Our code is available at https://github.com/Riyasushin/SIDEL.
Authors: Aayush Kumar, Avik Dutta, Sumit Gulwani, Gustavo Soares, Advait Sarkar, Emerson Murphy-Hill
Abstract: Plan Modes have become standard features in agentic programming tools, allowing users to gain transparency and control by working with the agent to develop a plan before task execution. However, it remains unclear whether the benefits of this feature translate to end-user programming environments such as spreadsheets. Since spreadsheet programmers tend to work iteratively and care less about technical correctness, upfront planning may not fit into their workflows as easily. In this paper, we build a prototype of a Plan Mode for spreadsheet programming and evaluate it against a non-planning baseline through a within-subjects user study (N=24). We found that despite similar task outcomes with both tools, using Plan Mode led to a reduction in refinement and a better perception of the tool across dimensions of creativity support and human-machine collaboration. We discuss the implications of these results for the future design of Plan Modes, and for the broader role of human-AI planning in end-user programming.
Authors: Tanjin He, Aikaterini Vriza, Logan Ward, Xu Huang, Yiming Chen, Anubhav Jain, Gerbrand Ceder, Rajeev S. Assary, Ian T. Foster, Maria K. Y. Chan
Abstract: X-ray absorption spectroscopy (XAS) is central to understanding the local electronic and atomic structure of materials, yet most published spectra remain inaccessible to data-driven analysis because they are embedded in figures and described through fragmented textual context in the literature. Here, we use multimodal (image and text) literature mining to transform this dispersed knowledge into an AI-ready experimental data resource. We developed a scalable spectroscopy data digitization pipeline that identifies XAS figures in full-text articles, digitizes spectral curves, and links each spectrum to accompanying metadata on the measured edge and material. Applying this pipeline to the battery literature produced an open dataset of 13,740 XAS spectra, spanning 66 absorbing elements and diverse battery chemistries, with expert validation confirming accurate extraction of spectral and metadata information. By converting literature-embedded spectra into structured numerical data, this dataset provides a foundation for large-scale XAS analysis, cross-laboratory comparison, high-throughput characterization, and autonomous discovery of advanced materials.
Authors: Parth G. Dangi, Yogesh Kumar Meena
Abstract: Brain-machine interfaces (BMIs) can assist individuals with limited mobility, such as stroke survivors or amputees. One of the key challenges in developing BMIs is expanding their usability and control, which can be achieved by accurately decoding multiple kinematic and kinetic parameters. To address this, we propose three regression models: partial least squares regressor, multilayered perceptron, and attention based regressor, to decode multiple movement parameters from EEG signals. We evaluated these models on the WAY EEG GAL dataset, focusing on their performance under subject specific and subject independent conditions with two strategies: a single model for all parameters and a baseline with separate models for each parameter. Among all regressors, the attention based regressor achieved the best performance, with an $R^2$ of 0.8 and a latency of 29.2 milliseconds, demonstrating significant improvement in simultaneous multi parameter decoding. However, its performance dropped for single parameter decoding. The multi layered perceptron showed more consistent but lower accuracy across both decoding types ($R^2$ = 0.49). These findings highlight the potential of attention based models for real time multi command BMI systems and contribute to the development of more intuitive control devices.
Authors: Shengyi Wang, Niantong Li, Guangzheng Hu, Hong Qi, Fei Ding, Weixu Qiao, Jinlin Wang, Xiaotong Lv, Peng Han, Zimeng Li, Fanshu Ding, Yushu Wang, Han Wu, Jingjing Chen, Chongxiao Wang, Yanhao Wu, Chenglong Huang, Xiaoqian Zhu, Jie Tian, Hua Li, Jingjing Fan, Mingshuang Tang, Zhong Li, Hengxia Qiang, Weibin Chen, Jinyang Zhen, Bing Zhao, Lin Qu, Jing Li, Hu Wei
Abstract: Progress in video generation keeps narrowing the visual gap between AI-generated and professionally produced footage, yet most benchmarks still draw prompts from web sources or LLM templates and score them with untrained, generic multimodal models. More fundamentally, their evaluation taxonomies remain rudimentary (overall visual quality, coarse text alignment and temporal smoothness) rather than the professional Cinematic Language criteria by which films are actually made and judged, so they assess basic video plausibility rather than film-grade craft. We introduce FilmBench, a text-to-video (T2V) and reference-to-video (R2V) benchmark grounded in the professional Cinematic Language of the film- academy tradition and co-developed with directors and faculty from the Beijing Film Academy and the Hujing Digital Media & Entertainment Group film studio. It rests on three choices. First, prompts are reverse-engineered from clips of award-winning films spanning 20 cinematic genres and chosen by professional directors, so every prompt is anchored to a verified live-action reference; the prompts follow real shot lists, and most script multiple shots (1,056 of the 1,169 prompts are multi-shot), unlike prior single-clip benchmarks. Second, evaluation follows a three-level Cinematic taxonomy of 3 axes, 12 components and 35 (T2V) +3 (R2V-only) sub-metrics. Third, we develop an in-house expert-grade automatic evaluation agent and open-source its core suite of Cinematic Language operators (FilmOps). Benchmarking leading video generation models (9 for T2V, 7 for R2V), the evaluator reproduces the human model ranking at model-level Spearman \r{ho} = 0.95 (T2V) and 0.96 (R2V). Scores fall well below prior web-style benchmarks, with two consistent gaps in dynamic aesthetics and a marked single- to multi-shot performance drop that widens for weaker models.
Authors: Mohsen Aliabadi, Keith Driscoll, Elliot Krop, Petar Sirkovic, Everett Sullivan, Elahe Vedadi
Abstract: We introduce an extremal invariant associated with Chowla-type order conditions in finite groups. A nonempty subset $S$ of a finite group $G$ is called a Chowla set if every element of $S$ has order greater than $|S|$, and we write $C(G)$ for the maximum cardinality of such a set. We first show that $C(G)$ is determined by the distribution of element orders in $G$. For cyclic groups, we derive an exact divisor formula and characterize the integers $n$ for which $C(\mathbb{Z}/n\mathbb{Z})=\varphi(n)$. We prove that $\liminf_{n\to\infty}C(\mathbb{Z}/n\mathbb{Z})/\varphi(n)=1$, whereas $\limsup_{n\to\infty}C(\mathbb{Z}/n\mathbb{Z})/\varphi(n)=\infty$, and we determine the corresponding lower and upper limits under normalization by $n$. For finite abelian groups, we obtain an explicit formula in terms of the invariant-factor decomposition, together with a closed formula for finite abelian $p$-groups. We then develop a linear analogue for finite field extensions. A nonzero $K$-subspace $A$ of an extension $L/K$ is called a Chowla subspace if $[K(a):K]>\dim_K A$ for every nonzero $a\in A$. Since this condition depends on $\dim_K A$, it does not generally require every nonzero element of $A$ to generate $L$ over $K$. Nevertheless, when $L/K$ is finite and separable, we prove the exact formula $C(L/K)=[L:K]-d_{\max}(L/K)$, where $d_{\max}(L/K)$ is the largest degree over $K$ of a proper intermediate field. For finite fields, we give a direct proof in every degree using a normal-basis construction. This work was developed through an expert-guided human-AI collaboration. A reasoning-focused configuration of Co-Scientist was used to explore examples and potential proof strategies. The authors formulated the problem, independently verified and completed all arguments, and wrote the final proofs.
Authors: Chandan Kumar Sah, Li Zhang, Xiaoli Lian
Abstract: Repository-level code generation relies on heterogeneous evidence whose relevance, compatibility, and completeness are inherently uncertain. Similar-code examples, repository context, and project-specific APIs may provide complementary information, but can also introduce noisy, redundant, or conflicting signals. Existing retrieval-augmented approaches primarily optimize retrieval relevance without explicitly modeling how uncertainty in retrieved evidence affects downstream generation. We introduce OpenCoder, an uncertainty-aware framework that estimates source-specific uncertainty, uses it to filter and rank heterogeneous evidence, and guides generation, verification, and repair. A factorial analysis over API knowledge, repository context, and similar-code evidence reveals no universal additive source ranking; instead, significant cross-source interactions depend on the accompanying evidence and LLM backend. On an expanded 32-task RepoExec-inline evaluation, OpenCoder improves GPT selected-output correctness over Baseline RAG from 56.25\% to 78.13\%. However, it matches a verification-and-repair control, and the corresponding Gemini improvement is not statistically supported, indicating backend-dependent benefits. Target-aware API refinement also substantially improves API-set retrieval. These findings support treating uncertainty as an actionable control signal for repository-level retrieval, verification, and repair.
Authors: Cesare Spinoso-Di Piano, Verna Dankers, Marius Mosbach, Jackie Chi Kit Cheung
Abstract: Human language is driven by unspoken beliefs and belief updates, making these critical to model for successful communication between large language models (LLMs) and their users. In this paper, we evaluate the ability of LLMs to recognize unspoken beliefs made through implicatures and to understand their updates through implicature cancellation: the pragmatic phenomenon whereby an utterance's implied meaning is weakened or negated. We create the first expert-annotated implicature cancellation dataset, ImplicatureX, crowdsourced for human judgements of implicatures and their corresponding cancellations. We find that LLM belief update understanding lags behind that of humans, especially in more naturally-occurring scenarios. Additional control experiments suggest that successes in LLM belief updates may stem in part from a reliance on prior beliefs, and that failures in belief updates may depend on their type and on their form. Overall, our study suggests that current LLMs have not yet reached human-level understanding of unspoken beliefs and belief updates. Code and data are available at https://github.com/cesare-spinoso/ImplicatureX.
Authors: Jianfei Ma, Zhaoxin Feng, Emmanuele Chersoni, Si Chen
Abstract: Prompt compression shortens LLM input to reduce inference cost, yet existing methods score token importance through LM forward passes. It remains questionable whether such nuanced, costly token selection is necessary. Compression requires identifying informative content, a problem that linguistic research has long addressed through cues that can be operationalized as deterministic rules. We therefore ask: can \textbf{linguistic rules alone} serve as effective prompt compressors, without LM-based scoring at compression time? To address this, we conduct offline evolutionary search over lexical, syntactic, semantic, and discourse seeds to find competitive rule combinations. The resulting linguistic compressor requires no LM forward pass at deployment and uses only CPU-side processing for compression. We evaluate it with a dual-path protocol to balance compression quality and reconstruction fidelity. Across short passages, multi-document reasoning, and dialogue-memory QA datasets, evolved compressors achieve performance similar to that of recent advanced prompt-compression strategies. Performance is strongest under light-to-moderate compression and degrades as compression becomes more aggressive, while the Direct and Reconstruction paths exhibit distinct patterns. Evolutionary analysis reveals that effective compression fuses signals across linguistic levels and, as the compression ratio increases, rules shift from token pruning to sentence extraction.
Authors: Yimao Guo, Zuomin Qu, Wei Lu
Abstract: The rapid advancement of video generation models has led to the increasing misuse of image-to-video (I2V) models. Although substantial progress has been made in detecting AI-generated videos, proactive defenses against I2V models remain underexplored. In particular, current proactive defenses against I2V models predominantly rely on gradient-based adversarial attacks, which require defenders to possess GPUs with substantial memory resources (VRAM) to generate adversarial examples. To address this issue, we propose I2VShield, a privacy protection method based on generative adversarial attacks tailored to Diffusion Transformer (DiT)-based I2V models. The proposed method primarily consists of two components: (1) a text-adaptive perturbation generation framework integrating adversarial learning to mitigate computational overhead while maintaining visual imperceptibility; and (2) an untargeted Multimodal Attention Disruption (MAD) attack that exploits the inherent vulnerabilities of DiT-based I2V models, maximizing the deviation of the internal attention features from their clean states. Extensive experiments demonstrate that our approach achieves highly competitive protection performance across various datasets and mainstream DiT-based I2V models, particularly in disrupting spatiotemporal coherence, while substantially reducing computational costs.
Authors: A. Contreras, D. Porres, R. Abad, P. Cano, A. Levy, G. Villalonga, A. M. L\'opez, A. Hern\'andez-Sabat\'e
Abstract: The development of autonomous vehicles (AVs) usually relies heavily on data-driven artificial intelligence (AI) models that require large volumes of sensor data with ground-truth annotations. While modular architectures are widely used, end-to-end driving paradigms offer a promising alternative by directly mapping sensor inputs to control actions. However, their adoption is limited by challenges in interpretability and explainability. To address this, we present LAIA (Labelled Attention for Intelligent Automobiles), a novel synthetic dataset designed to enrich end-to-end driving research with human attention data. Collected using the CARLA simulator in closed-loop environments, LAIA comprises over 15 hours of driving from 44 participants across carefully crafted scenarios designed to evoke natural responses. Each sequence includes RGB images under six weather conditions, semantic and instance segmentation, depth, optical flow, CAN bus signals, and synchronized eye-tracking data. LAIA enables applications including training attention-aware end-to-end AI drivers, predicting driver behavior, developing methods to detect anomalous driver-attention patterns, and improving model explainability. In this work, we use LAIA to compare human attention with the perceptual attention emerging in our end-to-end driving models, thereby providing insight into their behavior.
Authors: Chandan Kumar Sah, Li Zhang, Xiaoli Lian
Abstract: Retrieval-augmented generation improves knowledge-intensive question answering, but indiscriminate retrieval can introduce irrelevant evidence and unnecessary computation. We investigate whether verbalized confidence from black-box language models can serve as an actionable signal for retrieval routing. Our method, BeyondUncertainty, first elicits a structured provisional answer and confidence estimate, then applies a model-specific threshold selected on held-out validation data and frozen before test evaluation. Low-confidence questions receive top-5 TF-IDF retrieval followed by a second answer call, whereas high-confidence questions return the provisional answer directly. We evaluate 27,000 policy instances across six QA benchmarks, three model families, and three retrieval policies. BeyondUncertainty achieves 0.483 mean token-level F1, compared with 0.467 for always retrieval and 0.401 for no retrieval, while reducing retrieved passages by 20.4\% relative to always retrieval. When matched on the number of questions routed to retrieval within each dataset-model cell, it outperforms a post-hoc random allocation in 17 of 18 settings, with an average gain of 0.024 F1. Although poorly calibrated as an absolute probability, probe uncertainty modestly predicts question-level retrieval benefit (AUROC = 0.628). However, the additional probe increases total token usage by 28.2\%, revealing a trade-off between more selective evidence acquisition and end-to-end token efficiency.
Authors: Xinyi Hong, Pinjun Dong, Xinyang Yu, Binyan Jiang
Abstract: Large language model (LLM) agents increasingly rely on invoking external tools to complete real-world tasks. Tool retrieval, which selects a small task-relevant subset from a library of thousands of tools before the agent acts, has therefore become a critical component of LLM agent pipelines. However, existing retrievers either score each tool in isolation or assemble the tool set sequentially, so the joint utility of a candidate set is never evaluated as a whole. In this paper, we propose HYSET, short for HYperedge-based SEt-level Tool retrieval. Our contributions are threefold: (i) we formulate tool retrieval as query-conditioned hyperedge prediction on a tool co-invocation hypergraph, under which the tool set itself becomes the unit of scoring and most existing retrieval paradigms reduce to restricted instances; (ii) we capture size-dependent tool compatibility through cardinality-specific interactions; and (iii) we design HYSET as a pre-selection module requiring no modification to the downstream agent. Experiments on ToolBench demonstrate that HYSET consistently outperforms state-of-the-art baselines in both tool retrieval performance and end-to-end task success. Beyond the in-domain setting, HYSET further supports zero-shot/few-shot transfer, generalizing to held-out tools/categories and unseen domains with minimal supervision.
Authors: Fanfu Wei, Thibault Ehrhart, Rapha\"el Troncy
Abstract: Wikipedia and Wikidata are widely used for information access, LLM pre-training, and retrieval-augmented generation. Their knowledge is deeply connected but scattered across text, tables, and knowledge graphs. This raises a practical question: when these modalities disagree, how can we detect and explain the conflict? We study this problem as modality-level inconsistency detection. We first introduce a taxonomy of cross-modal knowledge inconsistencies, covering information granularity differences, direct conflicts, temporal changes, and KG incompleteness. We then present Kontrast, an automatic framework that uses Text-to-SPARQL and LLM reasoning to compare table-based answers with KG evidence and categorize the resulting inconsistencies. Experiments on various Table-QA datasets show that cross-modal inconsistencies are common and informative. They reveal not only true knowledge conflicts, but also missing KG structure and temporal mismatches while being limited by Text-to-SPARQL errors and noise. Our analysis shows that text, tables, and KGs can complement and correct one another through systematic comparison. Kontrast provides a practical tool for large-scale knowledge auditing and establishes a benchmark for future work on cross-modal knowledge consistency. Code and data are available at https://github.com/ECLADATTA/KONTRAST.
Authors: Podakanti Satyajith Chary, Barath Parthiban, Pranesh Velmurugan, Adeeba Khan, Nagarajan Ganapathy
Abstract: Ambivalence and hesitancy (A/H) are conflicting affective states that precede the delay or abandonment of health behaviour change. Recognition of A/H at the video level is difficult, since the signal arises from disagreement across and within facial, vocal, linguistic, and bodily modalities, and manifests differently across individuals. The proposed PRISM-AH (Predictive Reasoning over Interacting Streams for Multimodal Ambivalence/Hesitancy Recognition), is a framework that treats A/H as a multimodal conflict that unfolds over time. Frozen vision, audio, and text encoders are aligned into short time windows and passed to a lightweight streaming model that scores cross-modal dissonance, predicts each next window to expose a hesitation surprise signal, discovers behaviour prototypes, and is conditioned on participant metadata. Dense window-level annotations supervise the model as an auxiliary objective, and the decision threshold is calibrated for macro F1. A knowledge-guided large language model then reasons over structured evidence using the expert cue taxonomy of the dataset, and its verdict is fused late only when validation performance improves. On the labelled public test partition of 525 videos, PRISM-AH attains a macro F1 of 0.6133, compared to the reported zero-shot baseline of 0.2827. The reasoning gain is validated to transfer from validation to the larger test partition.