Peer Revenue Play credits. Real papers.

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Artificial Intelligence

cs.AI Watch Artificial Intelligence

Friday 4 September

Artificial Intelligence · Cross-listed from cs.CY

When Does an Interpretation Count as Established? The Formation, Evaluation, and Responsibility of Interpretation in Generative AI

Deyu Jing

Generative AI research has increasingly evaluated factuality, citation, coverage, and report structure. Yet passing such local checks does not by itself show that a humanistic interpretation has been established. This paper asks how an interpretation comes to be recognized within sociotechnical processes. It introduces three connected concepts. Interpretive appearance names the gap between the finished form of an output and the publicly traceable process through which materials, counterevidence, and revisions constrained the judgment. The evaluation contract names the bounded materials, tasks, criteria, permitted inferences, and failure conditions within which a local judgment is valid. Standing substitution names the unwarranted conversion of a genuine local pass into a stronger claim that an interpretation, result, or research capability has been established, without commensurate new evidence or bridging arguments. The paper then examines responsibility for judgment: a text may acquire recognition while no public structure remains for stating reasons, answering objections, revising, downgrading, or withdrawing the conclusion. Humanistic scholarship provides a revealing test because new materials and conceptual distinctions can alter both the question and the criteria of evaluation. The paper therefore develops delayed closure as a practice of keeping recognized interpretations revisable and proposes five public requirements concerning materials and versions, evidential roles, failure, contract revision, and responsibility. The argument is conceptual and normative: it does not claim to offer a benchmark or to determine whether models possess understanding. It instead explains why local evaluation, finished textual form, and public recognition must not be treated as sufficient evidence that an interpretation has been formed.
Read the full abstract

Artificial Intelligence

DCFA: Dual-view Causal-inspired Attribution for Failure Reasoning in LLM-based Multi-agent Systems

Zehao Wang, Lanjun Wang, Shilong Jin et al. (+2 more)

Large language model (LLM)-based multi-agent systems have experienced rapid growth in recent years. Despite their promise, such systems remain fragile, frequently exhibiting reasoning and coordination errors that can lead to system-level failures. Failure attribution in such systems relies on tracing natural language interactions among agents to identify the decisive error, which refers to the earliest action whose correction can reverse system failure. There are two key challenges: 1) Shallow attribution: Existing methods often capture only minor deviations, such as incomplete retrievals or formatting errors, which verification mechanisms can correct, while missing the decisive cause of system failure. 2) Contextual degradation: As the length of the system traces increases, the model's reasoning ability rapidly deteriorates. To address these challenges, we propose DCFA, a training-free framework for failure attribution. DCFA integrates a global module that constructs structured causal-inspired dependency graphs from system traces to identify the initial decisive error, and a local module that applies local counterfactual-inspired reasoning to refine causal-inspired attribution. Experiments on the Who&When benchmark across six LLMs show that DCFA improves step-level accuracy by up to 8.27% over state-of-the-art baselines.
Read the full abstract

All authors Zehao Wang, Lanjun Wang, Shilong Jin, Junjie Chen, Yanghua Xiao

Artificial Intelligence

Aplaud: Adaptive Personalized Low-Rank Decomposition for User-Specific LLM

Xinyu Li, Ruoming Jin, Jianfeng Zhu et al. (+2 more)

In this paper, we study the problem of personalized survey response prediction using fine-tuned large language models (LLMs). This task poses unique challenges: limited per-user training data, scalability of model storage, and the need to exploit shared structure across survey questions. To address these issues, we propose Aplaud (Adaptive Personalized Low-rank and User-specific Nested Decomposition), a lightweight and scalable framework for LLM personalization. Aplaud extends the LoRA paradigm by separating adaptation into a frozen, shared low-rank basis and a compact user-specific correction, augmented with a rank-one residual for finer personalization. To further reduce per-user parameter cost and mitigate overfitting, the correction matrix can be factorized into an even lower-rank form. Empirical results demonstrate that Aplaud achieves efficient, scalable personalization across users while outperforming state-of-the-art LoRA-based personalized LLM approaches in both generalization and inference efficiency.
Read the full abstract

All authors Xinyu Li, Ruoming Jin, Jianfeng Zhu, Ruixin Guo, Zhi Liu

Artificial Intelligence · Cross-listed from cs.CL

Knowing What Not to Answer: Selective Non-Compliance in Vision-Language Models

Minji Kim, Jihyoung Jang and Hyounghun Kim

Vision-language models (VLMs) are expected to respond helpfully to appropriate requests while withholding compliance with requests that are incorrect, unsafe, infeasible, or unanswerable. However, existing benchmarks predominantly evaluate non-compliance at the level of the query as a whole, assuming that each request either warrants compliance or requires withholding compliance. In practice, real-world queries can contain a mixture of answerable content and components for which compliance should be withheld. In this paper, we introduce KoNA, a benchmark for evaluating selective non-compliance in VLMs across five categories: False Premise, Visual Inaccessibility, Universal Unknown, Task Feasibility, and Safety. Each task evaluates two capabilities: query-level non-compliance and component-level non-compliance under paired single and compound queries. Our evaluation across diverse VLMs shows that models often fail to refuse, correct, or abstain appropriately, and these failures become more pronounced when queries require selective non-compliance. To address this challenge, we fine-tune VLMs using KoNA examples that require selective non-compliance, together with a fully answerable set that should receive direct answers. Our fine-tuned models achieve substantial improvements in non-compliance accuracy while largely maintaining performance on fully answerable tasks. These results suggest that the fine-tuned models can distinguish between answerable components and those requiring non-compliance and respond in a task-appropriate manner.
Read the full abstract

Artificial Intelligence

PLUME: Parameter-Efficient Personalization of Large Language Models via Low-Rank User Modulation in Shared Subspaces

Xinyu Li, Hao Zhou, Jianfeng Zhu et al. (+4 more)

Personalizing large language models (LLMs) is essential for delivering AI assistance that aligns with individual users' styles, intents, and preferences. While per-user fine-tuning can substantially enhance personalization quality, it introduces significant parameter and storage overhead, limiting scalability to large user populations. We propose PLUME (Personalized Low-Rank Adaptation through User Modulation and Shared Subspace), a lightweight framework that achieves efficient and expressive per-user adaptation by leveraging a shared task-specific subspace. Specifically, PLUME first learns a global task subspace from aggregated user data. Personalization is then achieved by training only a lightweight small square matrix within this subspace, enabling each user to obtain a tailored model while keeping shared components fixed. Cross-layer shared parameters and rank-1 residual terms are further introduced to significantly reduce redundancy while maintaining expressiveness. Experiments on multiple personalized text generation benchmarks demonstrate that PLUME achieves comparable or superior performance to strong baselines, while reducing per-user parameters by over 95%. These results establish shared-subspace modulation with minimal residuals as a scalable and semantically grounded approach to LLM personalization.
Read the full abstract

All authors Xinyu Li, Hao Zhou, Jianfeng Zhu, Julina Maharjan, Ruixin Guo, Feodor Dragan, Ruoming Jin

Artificial Intelligence · Cross-listed from cs.CL

Refuse without Refusal: A Structural Analysis of Safety-Tuning Responses for Reducing False Refusals in Language Models

Minji Kim and Hyounghun Kim

Striking a balance between helpfulness and safety remains a fundamental challenge in aligning large language models. To achieve this balance, models should refuse harmful queries (e.g., "How do I shoot someone?") while remaining responsive to benign inputs, even those superficially resembling harmful queries (e.g., "Where can I shoot a good photo?"). However, models often struggle to distinguish genuinely harmful queries from benign queries that contain superficially risky language, resulting in false refusals. In this paper, we address the issue by decomposing a response in the safety-tuning dataset into two distinct components: (i) a boilerplate refusal statement and (ii) a rationale explaining the refusal. Our experiments and analyses show that refusal statements impede accurate discrimination between harmful and benign queries by inducing reliance on superficial cues. In contrast, training solely on rationales reduces false refusals while maintaining a comparable level of safety performance. Rationale-Only benefits also appear in our ICL configuration and remain compatible with the evaluated inference-time mitigation methods. The results emphasize the necessity of precisely curated, fine-grained safety supervision datasets and outline directions for constructing aligned agents that better reconcile helpfulness with safety.
Read the full abstract

Artificial Intelligence · Cross-listed from cs.SE

Building a research-software catalog with a coding agent: from hackathon prototype to public deployment

Kazuyoshi Yoshimi, Satoshi Terasaki and Gotai Yamada

Generative AI and coding agents can accelerate research software development, but they also increase the need for efficient software discovery and maintenance. We developed a repository catalog during a three-day hackathon and subsequently examined the engineering required to make it suitable for public deployment, including adversarial review, data-quality checks, browser-level validation, and publication safeguards. We then explored whether the lessons learned from this prototype could be transferred to a much larger, human-curated portal, through a retrieval agent under development for MateriApps that combines curated portal metadata, external documentation, vector search, and local language-model generation. Implementation with coding agents was rapid, but achieving reliable operation required substantial additional engineering: the most consequential problems were not crashes but silent failures that produced plausible yet incomplete or incorrect outputs, arising from incomplete data acquisition, misleading assessments, and retrieval or preprocessing failures. These observations suggest that AI-assisted software portals require explicit validation, monitoring, and repeated review, and that curated metadata and maintained documentation remain essential. The MateriApps work is exploratory and remains under active development, so the observations reported for it are preliminary; a comparable combination of curated metadata, automatically collected documentation, and retrieval-based assistance may nevertheless be useful for extending other research-software portals.
Read the full abstract

Artificial Intelligence · Cross-listed from cs.LG

Simulation-free Unbalanced Dynamic Optimal Transport with General Growth Penalty

Junda Ying, Yuxuan Wang, Bowen Yang et al. (+2 more)

Inferring cellular dynamics from unpaired single-cell snapshots requires modeling both state transitions and population growth or death. Unbalanced dynamic optimal transport (UDOT) addresses this by penalizing growth along transport paths, making the choice of growth penalty a key way to encode biological priors on proliferation and apoptosis. However, existing UDOT solvers either rely on computationally expensive NeuralODE simulations or depend on analytical solutions of conditional paths, restricting their efficiency solely to quadratic penalties, i.e. Wasserstein-Fisher-Rao (WFR) geodesics. To enable an efficient UDOT solver for general growth penalties, we first show that concave growth penalties lead to degenerate solutions where growth and transport are separated. We then introduce Simulation-free Unbalanced Dynamic Optimal transport (SUDO), a simulation-free framework for UDOT with general non-quadratic convex growth penalties. SUDO learns the conditional paths and transport costs, solves the induced semi-coupling problem, and subsequently leverages unbalanced flow matching to achieve a simulation-free solution. On WFR benchmarks, SUDO matches the accuracy of efficient, analytical solution-driven algorithms while outperforming simulation-based methods in computational speed. Beyond WFR, SUDO supports asymmetric penalties that encode proliferation-dominant priors and produce more plausible trajectories and growth estimates on synthetic and single-cell datasets.
Read the full abstract

All authors Junda Ying, Yuxuan Wang, Bowen Yang, Peijie Zhou, Lei Zhang

Artificial Intelligence · Cross-listed from eess.SP

Wireless Foundation Models: State-of-the-Art and Open Challenges

Alonso M. Pacheco Huachaca, Juan J. Rodriguez Rodriguez, Ahmed Aboulfotouh et al. (+3 more)

Wireless foundation models (WFMs) have emerged as a promising approach for learning reusable representations from large-scale wireless data and adapting them to downstream tasks. However, the rapidly growing literature remains fragmented across modalities, pretraining objectives, architectures, adaptation strategies, and evaluation protocols, making it difficult to assess progress toward broadly transferable models. This survey provides a systematic analysis of WFMs for physical-layer applications. We first introduce the main WFM design components, including pretraining, backbone architectures, and downstream adaptation. We then organize the literature into five physical-layer task families: signal recognition and demodulation, channel representation learning, RF sensing and localization, beam management, and spectrum sensing and monitoring, while separately examining multi-task PHY models. Across these categories, we analyze how existing models are pretrained, adapted, and evaluated, with particular attention to downstream task diversity and the distinction between in-distribution, partial-shift, and out-of-distribution transfer. Our analysis shows that current WFMs provide increasing evidence of reusable wireless representations, but this evidence varies considerably across task families and evaluation settings. Differences in datasets, modalities, architectures, pretraining objectives, adaptation protocols, and distribution shifts make it difficult to determine which design choices drive transfer and generalization. We conclude by identifying open directions for improving data availability, evaluation rigor, generalization, efficient adaptation, and real-world deployment, providing a unified framework for understanding the current WFM landscape and the requirements for developing more reusable foundation models for future physical-layer wireless systems.
Read the full abstract

All authors Alonso M. Pacheco Huachaca, Juan J. Rodriguez Rodriguez, Ahmed Aboulfotouh, Nelson L. S. da Fonseca, Carlos A. Astudillo, Hatem Abou-Zeid

Artificial Intelligence

FinalityBench: An Effect-Level Benchmark for Agent Decisions Under Delayed and Conflicting Financial Finality

Abhishek Sharma

A merchant's payment processor, ledger, ERP and bank feed are updated by messages that get delayed, duplicated, dropped and reordered, so for minutes at a time the four hold contradictory beliefs about the same order. An agent resolving the exception must decide whether to ship goods, re-submit a capture, refund or wait, knowing some of those cannot be undone. We present FinalityBench, an executable benchmark for that decision. It keeps a hidden canonical event log and derives each system's view from a separately faulted delivery stream, so disagreement follows from specified fault semantics rather than being authored. Grading is on executed monetary effects: an episode is scored by the merchant's terminal economic position, relative to a privileged reference told when the pending capture resolves. The corpus of 321 tasks includes 45 twin pairs (90 tasks): tasks whose four system views are identical at the decision instant, whose authoritative probes both return unknown, and whose eventual correct dispositions differ. That snapshot indistinguishability is checked under every evaluation seed rather than assumed; equivalence over all interaction traces is not claimed. Over 14,445 graded episodes from nine programmatic policies, ranking by single-task accuracy and by paired loss disagree in 7 places: a ship-on-first-sign policy is second-best by accuracy at 65.7% and worst in the suite by paired loss, because it cannot tell the two members apart. A runtime gating irreversible actions on an authoritative finality probe reaches 85.4% and, unlike every polling policy, loses nothing to pass^5; its residual loss is almost entirely one archetype, which prices finality information directly. Language models reach the same exact rate as the hand-written gate on a stratified subset, lose about twice as much money, and discover the finality-gating strategy without being told it.
Read the full abstract

Artificial Intelligence

Model Retirement Creates Reproducibility Risk in Biomedical AI Publications

Nathan Wolfrath, Meghan Conroy, Thomas Kosten et al. (+7 more)

Background. Large language models (LLMs) are being adopted in biomedical research at a rapid and accelerating pace, yet commercial services that host many widely used models operate under deprecation schedules that can complicate scientific reproducibility. Methods. We searched PubMed for original research articles from 2022 through March 2026 that applied a specific LLM to a biomedical task. An extraction agent identified model names from 61,077 article abstracts with human reviewers validating a subset for extraction accuracy. Extracted model names were normalized to canonical model identifiers. Lifecycle data (release date, retirement date, status) were compiled for the 50 most frequently used models. Results. We identified 8,931 paper-model mentions spanning 5,242 unique publications after restricting the analysis to the 50 most frequently used models. Among these mentions, 77.7% cited a commercial closed-weight model. Overall, 42% involved a model that was already retired by the time of official publication or is scheduled to retire within two years of publication. The median interval from publication to model retirement was 538 days. Conclusion. Many biomedical publications using LLMs are on a trajectory toward computational non-reproducibility after publication. Model deprecation should be treated as a core reporting and preservation issue for biomedical research.
Read the full abstract

All authors Nathan Wolfrath, Meghan Conroy, Thomas Kosten, Dave Bell, Bhabishya Neupane, Jonah Kindel, Anjishnu Banerjee, Priya Deshpande, Bradley Taylor, Anai N. Kothari

Artificial Intelligence

SQL-Zero: Self-Evolving Text-to-SQL

Daniel Machado Pedrozo, Julia Soares Dollis, Bryan Lincoln Marques de Oliveira et al. (+3 more)

Training a competitive Text-to-SQL agent usually depends on human-annotated natural-language/SQL pairs, which are expensive, domain-specific, and a bottleneck for scaling to new databases. We show it is possible to train a competitive solver with zero annotated pairs. We introduce SQL-Zero, a proposer-solver self-play in which a challenger and a solver start from the same base LLM and the only ground truth is execution against the database itself. The challenger generates SQL pairs calibrated to the solver's current difficulty (targeting "hard but solvable"), and both roles are updated with GRPO in alternating turns, with a template-level repetition penalty on the challenger to prevent diversity collapse. Training on BIRD databases with no labels, self-play improves over the zero-shot base on BIRD dev by 6.6 points at 3B and 7.3 points at 7B. It also scores higher than a matched control trained under the same recipe on human BIRD gold over the same databases, although an exact paired test does not resolve that margin. Transfer depends on scale: at 3B every iteration outperforms the base on unseen Spider databases and under lexical perturbation (Spider-Syn), where it also degrades less than the matched BIRD-gold control, whereas at 7B only the first iteration preserves transfer.
Read the full abstract

All authors Daniel Machado Pedrozo, Julia Soares Dollis, Bryan Lincoln Marques de Oliveira, Vinicius Alboneti Aguiar, Sávio Salvarino Teles de Oliveira, Telma Woerle de Lima Soares

Artificial Intelligence

Predicting Spatiotemporal Mobile Sensing-Based PM2.5 Concentrations Using Low-Rank Adapted Spatially Attentive Graph Neural Network

Om Chiddarwar, Priyanka Mandal, Praveen Kumar Chandaliya et al. (+1 more)

Urban air quality can vary significantly along transit corridors, necessitating high-resolution monitoring. This work introduces a novel mobile-sensing dataset from Surat, Gujarat, India, comprising PM$*{2.5}$ concentrations, meteorological variables (temperature, humidity, wind speed, wind direction), and land-use features. To represent the spatiotemporal data as a graph, two node-definition strategies were used: (i) uniform segmentation (200–400~m intervals) and (ii) DBSCAN clustering to adaptively group dense observations. For each node, rolling mean and standard deviation of meteorological variables were computed. To model this high-dimensional data, we propose a SA-GNN for fine-grained, short-term PM$*{2.5}$ forecasting and hotspot identification. We compared SA-GNN with LSTM, RNN, GRU, and ANN models. These models performed well on low-resolution data but had difficulty capturing rapidly changing patterns in urban air quality. SA-GNN employs cluster-specific GRUs to capture localized temporal dependencies and a Graph Attention Network to learn spatial heterogeneity. This hybrid architecture effectively models rapid fluctuations and complex spatial interactions. On our dataset, SA-GNN achieved $R^2 = 0.95$, RMSE $= 6.8$, and MAE $= 4.2~\si{\micro\gram\per\meter\cubed}$, outperforming all baseline models. Combining spatial clustering with adaptive attention significantly improves forecasting, enabling real-time, fine-grained monitoring and supporting personalized exposure tracking and timely alerts for healthier cities.
Read the full abstract

All authors Om Chiddarwar, Priyanka Mandal, Praveen Kumar Chandaliya, Shriniwas Arkatkar

Artificial Intelligence · Cross-listed from cs.CV

Enhancing Multimodal Emotion Recognition via Multi-Feature Encoding and Attention-Based Fusion

Xu Lin, Ke Wang, Hui Kang et al. (+1 more)

Multimodal emotion recognition has attracted growing interest due to its importance in human-computer interaction, remote education, and healthcare. This paper proposes a novel multimodal emotion recognition framework that integrates rich audio and visual feature extraction with an attention-based fusion strategy. For audio, we extract three complementary feature types: semantic embeddings from Wav2Vec2, MFCC features, and statistical acoustic descriptors such as pitch, energy, and rhythm. These are aligned and fused via a BiLSTM to capture temporal dependencies. For video, we propose a ResNet50-BiLSTM architecture that combines deep residual learning and sequential modeling to extract expressive spatiotemporal features from facial sequences. To enhance multimodal synergy, we introduce a feature-level fusion mechanism based on multi-head attention, allowing the model to adaptively weigh contributions across modalities. Experiments conducted on the MELD and IEMOCAP datasets demonstrate that our model significantly outperforms baselines in both accuracy and robustness. Furthermore, ablation studies show that the attention-based fusion strategy significantly improves performance in unbalanced data settings. Our findings suggest that the proposed framework effectively captures diverse emotional cues from speech and visual expressions, and offers a practical and generalizable approach for real-world multimodal emotion recognition tasks.
Read the full abstract

All authors Xu Lin, Ke Wang, Hui Kang, Xinying Wang

Artificial Intelligence · Cross-listed from cs.SE

Beyond Code Generation: Reliability, Verification, and Cost Economics in the Agentic Software Development Lifecycle

Happy Bhati

AI coding systems are moving from autocomplete and chat toward agents that can inspect repositories, edit multiple files, run tools, write tests, open pull requests, and work for long periods with limited supervision. This capability changes the bottleneck in software delivery. Recent field studies show meaningful gains in coding activity, but newer evidence also shows that those gains attenuate sharply between writing code and shipping reliable software. Review, integration, testing, security, deployment, and production operations remain constraining stages, while the economics are shifting from predictable per-seat licensing toward variable token, tool, sandbox, CI, and rework costs. This paper synthesizes peer-reviewed software-engineering research, university studies, benchmark audits, production reports from major technology companies, developer telemetry, and cost-management evidence released primarily from 2024 through September 2026. No new model experiment is claimed; numerical findings remain attributed to their original studies. The synthesis proposes four engineering concepts: the Agentic SDLC Throughput Paradox, Production-Qualified Change (PQC), the Verification Tax, and an Agentic SDLC Control Plane that allocates autonomy subject to cost, reliability, and human-attention budgets. An evidence-based horizon then maps today's supervised agents to future policy-bounded software factories. The central research question shifts from how much code an agent can generate to how much production-qualified value an engineering system can deliver per dollar, per reviewer-hour, and per unit of operational risk.
Read the full abstract

Artificial Intelligence

Train What You Deploy:Token-Faithful Post-Training of a Production Coding

Cheng Li, Jiexiong Liu, Yixuan Chen et al. (+1 more)

Existing post-training pipelines for coding and terminal agents suffer severe token and control fidelity errors: simplified training environments mismatch production deployments, and offline token reconstruction from agent logs distorts original prompts and conflates policy calls with background model operations. We present a fidelity-aware training coupling framework that retains trainer-side sampling over original prompts, eliminates spurious model calls via a negotiated training protocol, and restricts loss computation to verifiable token spans with closed-failure guarantees. We further propose Certified Divergence Proximal Policy Optimization (C-DPPO), which establishes tight two-sided TV certification bounds, adaptive-K rules, budget-aware sequence guarantees, and error-robust policy masking atop standard DPPO. Evaluated on matched Baize5B and Baize10B models with identical training and test protocols on TMax-100, C-DPPO yields a consistent +3.0-point performance gain over standard DPPO across model scales. Certificate audits validate the reliability and full operational coverage of our certified training pipeline.
Read the full abstract

All authors Cheng Li, Jiexiong Liu, Yixuan Chen, Chi Hong

Artificial Intelligence

ERPBench: Evaluating LLM Agents for Enterprise Decision-Making Across Competitive Market Ecologies

Xinran Zhang, Pengrui Lu, Lyumanshan Ye et al. (+1 more)

Large language model (LLM) agents are increasingly proposed for enterprise workflows, yet existing evaluations rarely test whether business-decision conclusions transfer across competitive market ecologies. We introduce ERPBench, an execution-instrumented benchmark for enterprise decision agents in a six-round Enterprise Resource Planning (ERP) simulation with coupled pricing, production, procurement, inventory, finance, and shared-market competition. ERPBench evaluates the same 100 fixed problems in two matched competitive market ecologies: Solo, where each evaluated LLM agent competes against fixed rule-based opponents, and Arena, where six evaluated LLM agents compete in a shared market. Across six model families, this yields 1,200 model-level trajectories spanning 7,200 decision rounds. Under the observed service configuration, the leading model differs between ecologies: DeepSeek leads in Solo (252.29M mean valuation; mean rank 1.67), whereas Gemini leads in Arena (263.95M; 1.76). The two ecologies identify the same task-level winner on only 21 of 100 problems, and Gemini's bottom-rank rate falls from 22 % to 0 % in Arena. ERPBench supports paired evaluation of whether enterprise-agent rankings transfer across competitive market ecologies, supplemented by aggregate execution-intervention analysis. Code and benchmark resources are available in our https://github.com/GAIR-NLP/erp-bench.
Read the full abstract

All authors Xinran Zhang, Pengrui Lu, Lyumanshan Ye, Pengfei Liu

Artificial Intelligence

Harness-agnostic detection and immunization of reward hacking in self-evolving language models

Rongxin Yang, Yang Liu, Shang Luo et al. (+10 more)

Self-evolving language models improve by proposing candidate updates and keeping whatever raises a visible score. When that score is an imperfect proxy for the capability one actually wants, sustained selection widens the gap between the two. This is reward hacking. We introduce HackProbe, a monitor that attaches to an arbitrary self-evolving loop through two black-box hooks, with no access to weights or activations. It keeps a secret, distribution-fixed comparison core, whose frozen distribution makes its capability proxy comparable across generations, alongside a rotated fresh layer that hardens the bank against co-adaptation. Four tests built on that proxy cover the level gap, a scale-aligned divergence with online change-point detection, capability stagnation, and a conditional confidently-wrong rate; a Sidak correction turns them into a calibrated family-wise p-value. Diagnosis alone recovers nothing, so a risk-aware immunization layer reselects an honest candidate from the proposal pool using the core together with a purely structural gaming footprint, disclosing at most log2 Pi bits per generation to the host. We prove a detectability bound that converts a target error rate into an explicit probe-size budget, and we delimit what probe rotation does and does not buy. On a controlled prompt-level host with four injected hacking channels and ground-truth labels, HackProbe reaches 0.763 AUROC against 0.663 for the strongest baseline and cuts the false-positive rate from 0.706 to 0.434. Its bandwidth-limited reselection is the only immunization level that returns more true capability under hacking, 5.2 points on average, than it forfeits on clean runs, 4.7; per-channel effects are mostly not individually significant.
Read the full abstract

All authors Rongxin Yang, Yang Liu, Shang Luo, Haoxuan Jia, Chongyang Zhang, Hao Zheng, Yingguang Yang, Yulin Huang, Jianshen Zhang, Yongzhi Qi, Kefu Xu, Congjing Ran, Bin Chong

Artificial Intelligence

Continual Graph Memory for Adaptive Recommendation under Intent Drift

Hao Nguyen Ngoc, Tung Nguyen, Nguyen Thi Hanh et al. (+2 more)

This paper studies adaptive recommendation under intent drift, where feedback from each recommendation outcome can reveal whether the relational evidence used for ranking is useful, missing, or misleading. While Knowledge Graphs (KGs) provide essential semantic structure to handle these shifts, traditional KG-enhanced systems treat the graph as a static retrieval substrate, making it brittle to evolving intents, noisy metadata, and recurring failure patterns. This paper proposes CGM-Rec, a continual graph memory framework for adaptive recommendation. CGM-Rec treats the graph state as a writable memory and maintains two complementary components. Therein, a Semantic Graph Memory is updated conservatively through quality-gated typed operations for storing stable and high-confidence relational knowledge. Meanwhile, an Episodic Lesson Memory acts as a fast reactive memory that learns recent outcomes, failure cases, and corrective hints. During testing, model parameters remain frozen and adaptation occurs only through memory writes. We evaluate CGM-Rec under a frozen-parameter, one-pass reranking protocol, where encoders and prompts remain fixed during testing and adaptation occurs only through memory writes. Experiments across multiple recommendation settings show that CGM-Rec improves over evaluated neural and LLM-based baselines on most metrics. Particularly, under sampled-candidate reranking, CGM-Rec improves HR@1 by up to 29.58% over the strongest LLM baseline on Bundle, and outperforms K-RagRec on metadata-rich ML-100K with HR@5 of 0.5941 versus 0.4746.
Read the full abstract

All authors Hao Nguyen Ngoc, Tung Nguyen, Nguyen Thi Hanh, Hoang Thai Dinh, Nguyen Xuan Tung

Artificial Intelligence

A Cost-Aware Agentic Architecture for NL-to-SQL over Nested Enterprise Schemas, with a New Benchmark

Yoga Sri Varshan Varadharajan, Ajay Yadav, Ritesh Goru et al. (+5 more)

Natural-language-to-SQL systems have ad- vanced rapidly on academic benchmarks, yet production enterprise schemas exhibit graph- like, semi-structured, deeply nested structure that current benchmarks do not measure. We make two complementary contributions. First, we introduce the DevRev NL2SQL bench- mark: 900 execution-verified queries with nested-type and link-graph structure, accom- panied by the Semantic Depth Score (SDS), a schema-agnostic rubric for analytical reasoning depth. Second, we present a cost-aware single- generation agentic architecture whose schema- selection, metadata-retrieval, and error-repair components are designed for the requirements this regime imposes. On the DevRev NL2SQL benchmark the system attains 91.7% answer correctness, a margin of 54.6 percentage points over the next-best baseline; on the Spider 2.0 Snowflake public dataset, it is competitive with leading systems at a single-generation operating point.
Read the full abstract

All authors Yoga Sri Varshan Varadharajan, Ajay Yadav, Ritesh Goru, Prateek Chaudhury, Constantine Caramanis, Prateek Jain, Divyateja Pasupuleti, Sunil Kumar Pandey

Artificial Intelligence · Cross-listed from cs.CL

Tracing Audio Grounding and Answer Selection in Audio LLMs

Hyebin Cho, Suho Yoo, Jihoo Jung et al. (+1 more)

Audio Large Language Models (Audio LLMs) have advanced in audio understanding, yet they can still predict the answer by reasoning from textual cues or linguistic priors rather than the provided audio. A common remedy is to train models on data whose answers cannot be inferred from text alone. This approach can improve performance, but what changes within the model remains unclear. In this paper, we ask what must happen inside the model for the audio to actually determine the answer. Our findings are threefold. (1) Replacing the audio with silence or unrelated audio causes substantially larger performance degradation in the trained model than in the pretrained model. (2) Acoustic information most strongly shapes the model's representations of the answer choices in early-to-middle layers, while training mainly increases the influence of audio information on the final prediction in middle-to-late layers. (3) The weights learned during training have their largest impact in specific layer bands. Together, these results provide a mechanistic account of how training strengthens the use of acoustic evidence in Audio LLMs.
Read the full abstract

All authors Hyebin Cho, Suho Yoo, Jihoo Jung, Joon Son Chung

Artificial Intelligence · Cross-listed from cs.SD

SCAPES: Semantically Conditioned Autoregressive Prior for Environmental Sounds

Esteban Gutiérrez, Lonce Wyse, Frederic Font et al. (+1 more)

As generative audio models grow in complexity, the computational and ecological costs of synthesizing everyday sounds have become increasingly prohibitive, often requiring industrial-scale resources and massive datasets. In this paper, we present SCAPES: a Semantically Conditioned Autoregressive Prior for Environmental Sounds. SCAPES is a lightweight, resource-efficient generative model designed to synthesize high-fidelity environmental textures through high-level semantic control. By operating on the continuous latent manifold of a neural audio codec, our approach bypasses the rigid structural constraints inherent to discrete tokenization. We propose a segmentation strategy that decomposes audio into overlapping segments, enabling a Continuous Normalizing Flow (CNF) to model the evolution of latent trajectories using Flow Matching. Our experiments demonstrate that a 36-million parameter instance of SCAPES can be trained on limited, uncurated datasets using a single consumer-grade GPU. Notably, convergence is achieved after training for approximately twice the source audio duration, yielding high-fidelity outputs with robust long-term stability and semantic consistency. Furthermore, we showcase the model's capacity for smooth semantic interpolation, providing a flexible and accessible tool for open research and creative sound design. Code, pretrained weights, audio examples, and an interactive demo are publicly available on our project page https://cordutie.github.io/projects/scapes.html
Read the full abstract

All authors Esteban Gutiérrez, Lonce Wyse, Frederic Font, Xavier Serra

Artificial Intelligence

SiLR: Structure-Preserving Admission and Process Reward for LLM Tool Agents

Chenyu Zhou, Qiliang Jiang, Shuning Wu et al. (+1 more)

A runtime gate for an LLM tool agent is usually cast as a filter. In a ReAct loop a rejected proposal is followed by another at the same state, so the gate is a search operator over the proposal stream whose admission criterion shapes which trajectories are reachable. We study post-violation recovery admission, where progress must be admitted while the system is still in violation, and identify the scalar projection trap: an aggregate-score gate accepts a locally improving proposal and commits the trajectory to a plateau. SiLR instead shadow-executes each proposal and admits it under a product order over the branch-level violation state (overloaded-branch support and per-branch severity). We prove that no scalar surrogate is sound for this order, so the failure is representational, not a matter of threshold tuning. On mined Gym-ANM scenarios, SiLR recovers 21/21 multi-action episodes against 0/21 for terminal and 9/21 for the best scalar gate, significant across the full 24-scenario benchmark. The terminal-versus-structured dichotomy holds across three model families and in CityLearn. Because admission rests on deterministic simulation, the LLM lies outside the trust boundary: a magnitude-redistribution attack that defeats both scalar and support-only baselines is contained only by the full per-branch predicate. With two constraint families active, every tested scalar projection admits physically unsafe actions; support-only admits the largest fraction (63.2% of 42,410; product order 0). In the hardest dual-family traces, scalar gates recover only through that unsafe class. Reused as a GRPO process reward, it outperforms its count projection in every mined scenario and is the only tested reward whose ungated policy exceeds the untrained base (0.844 vs. 0.778). Scalar projection loses the violation geometry at both design points; only the full product order is structurally sufficient.
Read the full abstract

All authors Chenyu Zhou, Qiliang Jiang, Shuning Wu, Xu Zhou

Artificial Intelligence

Leveraging Imperfect Restoration for Data Availability Attack

Yi Huang, Jeremy Styborski, Mingzhi Lyu et al. (+2 more)

The abundance of online data is at risk of unauthorized usage in training deep learning models. To counter this, various Data Availability Attacks (DAAs) have been devised to make data unlearnable for such models by subtly perturbing the training data. However, existing attacks often excel against either Supervised Learning (SL) or Self-Supervised Learning (SSL) scenarios. Among these, a model-free approach that generates a Convolution-based Unlearnable Dataset (CUDA) stands out as the most robust DAA across both SSL and SL. Nonetheless, CUDA's effectiveness against SSL is underwhelming and it faces a severe trade-off between image quality and its poisoning effect. In this paper, we conduct a theoretical analysis of CUDA, uncovering the sub-optimal gradients it introduces and elucidating the strategy it employs to induce class-wise bias for data poisoning. Building on this, we propose a novel poisoning method named Imperfect Restoration Poisoning (IRP), aiming to preserve high image quality while achieving strong poisoning effects. Through extensive comparisons of IRP with eight baselines across SL and SSL, coupled with evaluations alongside five representative defense methods, we showcase the superiority of IRP. Code: https://github.com/lyumingzhi/IRP
Read the full abstract

All authors Yi Huang, Jeremy Styborski, Mingzhi Lyu, Fan Wang, Adams Kong

Artificial Intelligence

$\tau^\tau$-Bench: An Environment for End-To-End, Realistic Agent Construction

Quan Shi, Keshav Dhandhania, Karthik Narasimhan et al. (+1 more)

LLM agents are rapidly becoming production software, deployed to handle customer service, adjudicate disputes, and operate internal systems. Notably, the work of building them is increasingly handed to coding agents, yet existing benchmarks say little about whether an AI system can deliver one under the conditions of a real client engagement. We introduce $\tau^\tau$-bench (pronounced hyper-tau-bench), a benchmark that makes agent construction the task. A developer agent is given the records a business actually keeps, a client who holds requirements, a production API that operations must run through, a codebase to inherit, and limits on serving cost and models: the same starting point a real engagement provides. From these it must deliver a complete customer-service agent, scored by deploying that agent against held-out simulated users. Across 53 tasks spanning four domains, the strongest configuration, Claude Opus 5 under Claude Code, passes just 23.9% of evaluation simulations. Meanwhile, an expert-authored reference ceiling scores 82.2%. The failures mirror ones human agent developers see: models issue shallow queries in place of deep comprehension of the records, communicate almost nothing to the client, and experiment too little with agent architecture and serving spend, shipping the first design that runs. We aim for $\tau^\tau$-bench to turn the work of cooperative agent building into a measurable target for coding agents.
Read the full abstract

All authors Quan Shi, Keshav Dhandhania, Karthik Narasimhan, Victor Barres

Artificial Intelligence · Cross-listed from cs.CL

PetQA: Benchmarking Veterinary Knowledge and Clinical Reasoning

Taegyun Kim, Youngwook Ham, Jungwook Rhim et al. (+3 more)

We introduce PetQA, a Korean long-form question-answering (QA) benchmark for evaluating veterinary knowledge and clinical reasoning in large language models (LLMs) and large vision-language models (LVLMs). PetQA contains 10,076 text-only and 8,751 multimodal QA pairs derived from real-world questions about dogs and cats, paired with answers from expert veterinarians. Its test split, PetQA-Bench, further includes annotations for question types and clinical conditions. We evaluate eighteen models using ROUGE, BERTScore, and LLM-as-a-judge metrics for factuality and helpfulness under three settings: zero-shot inference, retrieval-augmented generation (RAG), and supervised fine-tuning (SFT). The benchmarking results provide an overview of the strengths and limitations of current models in addressing veterinary clinical queries and highlight the need for more effective adaptation methods to develop clinically reliable AI systems for veterinary care. To facilitate broader use, we additionally provide translated versions of PetQA-Bench in five languages.
Read the full abstract

All authors Taegyun Kim, Youngwook Ham, Jungwook Rhim, Ju-Hyun An, Sungkyu Park, Kunwoo Park

Artificial Intelligence · Cross-listed from cs.CV

Dual-Part Multi-Lateral Branched Network for Multi-Class Segmentation in Cardiovascular Catheterization Angiograms

Olatunji Omisore, Ahmed Elazab, Ali Shahidinejad et al. (+1 more)

Catheterisation image processing requires segmentation models that are fast, accurate and explainable. While most of the existing studies usually focus on binary segmentation, there is a recent demand for simultaneous segmentation of multiple structures found in catheterization scenes. In this study, a dual-part MLBNet architecture is designed with multi-lateral encoder blocks and multi-head decoder branches for class-aware segmentation in cardiovascular catheterization scenes. Lateral branches in the encoder enables repeated feature extraction to learn diverse shared representations, while multiple decoder heads are used to introduce class-skewed branches that specialize in different structural properties in catheterization scenes. To analyze the performances of the dual-part MLBNet architecture, several multi-class segmentation angiogram data obtained during cardiovascular catheterization in phantom models, synthetic human-simulated aorta, and animal model are used for model training and evaluation. Results obtained showed the dual-part models could effectively separate guidewire, catheter, vessels and background pixels to their classes of memberships with high probability. The results demonstrate that all models were able to distinguish the dominant background class from foreground structures with high overall accuracy.
Read the full abstract

All authors Olatunji Omisore, Ahmed Elazab, Ali Shahidinejad, Fariza Sabrina

Artificial Intelligence · Cross-listed from cs.CL

When Do Internal Probes Beat Reading the Answer? Miscalibrated Readouts and Behavior-Concealed Knowledge in Language Models

Gnaneswar Villuri, Hashmath Shaik and Alex Doboli

A 0.6B language model, asked to verify 1,200 logical conclusions (half valid, half corrupted by a single semantic edit), answers YES every time. Judged by behavior it discriminates nothing; linear probes on its hidden states read the correct verdict at 0.96 AUC, transferring to unseen logical structures and separating foils built from exactly the words of the true conclusion (0.90). We ask where the verdict is lost, and find the dominant failure is a single scalar. The verdict survives to the model's own output logits (margin AUC 0.89) along a well-aligned readout direction; a saturated decision threshold, offset by +4.6 sigma, erases it. The diagnosis generalizes: across 90 semantic-label configurations of a five-model, three-family factorial, behavioral accuracy collapses onto a single function of threshold offset (Spearman -0.93) while margin ranking moves far less. Across a 13x scale range, internal knowledge saturates while free-form behavior is non-monotone: an 8B model underperforms its 4B sibling through an answer-channel failure rather than the threshold; forced-choice accuracy is monotone. The diagnosis is actionable: a one-parameter correction, never fit on evaluated structures, repairs behavior from 50% to 81% (0.6B); calibrated margin decoding recovers 94% at 8B; few-shot prompting works the same way, recentering the threshold (+4.6 sigma to 0.0 sigma) while preserving ranking. Comparing probe to margin separates three regimes: concealed, miscalibrated, and undetected. On a maze task built so foils carry no surface cues, the audit correctly reports the third. In the standard generation setting, answer-surface features and heuristic labels reproduce published probing results without any internal access.
Read the full abstract

Artificial Intelligence

Does the Selected Object Reach the Reader? Auditing Identity Handoffs in Grounded Language-Model Pipelines

Siddharth Vohra, Runmin Jiang, Xiaomo Li et al. (+1 more)

Grounded language-model pipelines can be divided into three stages: selecting an object, retrieving passages for it, and using that evidence to answer. If the selected object must reach the reader, losing it breaks the handoff. Benchmark recall checks the dataset-linked object, which can differ. We audit 600 HybridQA questions across three selector families. On 1,463 resolvable records where the selected object matches the dataset-traced passage, exact key lookup and exact title matching return the object every time. With every ranked rule given the same decoded selected title, body-only BM25 omits it on 389 records (26.6%) at cutoff five, while hybrid retrieval with reranking omits it on 14 (1.0%). The two identities differ on 329 of 1,792 resolvable records. With original-question rankings, their top-five checks disagree on 106 records (5.9%). Frozen reader comparisons associate the aligned object's presence with 28.6 to 31.0 points higher exact match. In a deliberately selected 64-item cohort, removing that passage sharply lowers exact match, while removing a similar-length comparison passage does not reproduce the drop. We release the Returned-Object Profile (ROP), an executable record of the target, returned-ID field, cutoff, membership rule, and complete expected population, with data and an offline replay.
Read the full abstract

All authors Siddharth Vohra, Runmin Jiang, Xiaomo Li, Min Xu

Artificial Intelligence · Cross-listed from cs.LG

Training-Free Halving of Activated Experts in Fine-Grained Mixture-of-Experts Models

Xing Chen and Hengshuai Yao

Modern fine-grained Mixture-of-Experts (MoE) models route each token to a small number of experts and renormalize their router probabilities. We show that this renormalization implicitly calibrates expert output gain to the training top-$k$: reducing $k$ at inference changes not only which experts are used but also the strength of the expert branch. We separate these effects by activating the top $k_1$ experts while normalizing by the probability mass of the top $k_2$ experts, introducing one integer with no parameters, training, or measurable compute overhead. On Qwen3.6-35B-A3B, reducing from 8 to 4 experts causes a 4.65-point MMLU drop under standard renormalization but only 0.35 points with $k_2=16$, while halving routed-expert compute. The result replicates on the $11\times$ larger Qwen3.5-397B-A17B, where reducing from 10 to 5 experts loses only 0.55 points with an appropriate reference set. Removing renormalization entirely is catastrophic, showing that preserving a suitable reference mass is crucial. We further find that perplexity and downstream accuracy favor different $k_2$, cautioning against selecting MoE compression settings using unlabeled text alone. Analyses also show that expert identity matters substantially more than expert weighting, while balanced and domain-specialized routing leaves limited room for expert pruning.
Read the full abstract