Ax Chang Li, Tshihao Tsu, Yaren Zhang, Chao Xue, Xiaodong He 3/10/2026

Fibration Policy Optimization

Policy optimization framework for heterogeneous LLM systems with hierarchical stability control.

Ax Makoto Sato, Yusuke Iwasawa, Yujin Tang, So Kuroki 3/10/2026

SAIL: Test-Time Scaling for In-Context Imitation Learning with VLM

Test-time scaling framework for robot imitation learning using VLM and Monte Carlo tree search.

Ax Yuxiang Zhang, Enyan Dai 3/10/2026

SCL-GNN: Towards Generalizable Graph Neural Networks via Spurious Correlation Learning

Graph neural networks addressing spurious correlation learning for improved generalization.

Ax JV Roig 3/10/2026

How Much Do LLMs Hallucinate in Document Q&A Scenarios? A 172-Billion-Token Study Across Temperatures, Context Lengths, and Hardware Platforms

Large-scale empirical study measuring LLM hallucination in document Q&A across temperatures and context lengths using 172B tokens.

Ax Hankun Kang, Di Lin, Zhirong Liao, Pengfei Bai, Xinyi Zeng, Jiawei Jiang, Yuanyuan Zhu, Tieyun Qian 3/10/2026

AdaCultureSafe: Adaptive Cultural Safety Grounded by Cultural Knowledge in Large Language Models

Method for grounding cultural safety in LLMs through adaptive cultural knowledge integration.

Ax William Thorne, Joseph James, Yang Wang, Chenghua Lin, Diana Maynard 3/10/2026

Evaluating LLM-Based Grant Proposal Review via Structured Perturbations

Framework evaluating LLM performance on grant proposal review using perturbation-based testing across quality dimensions.

Ax Yehonatan Elisha, Oren Barkan, Noam Koenigstein 3/10/2026

Concept-Guided Fine-Tuning: Steering ViTs away from Spurious Correlations to Improve Robustness

Fine-tuning method for Vision Transformers using concept guidance to reduce reliance on spurious correlations and improve robustness under distribution shifts.

Ax Sadegh Rahmaniboldaji, Filip Rybansky, Quoc C. Vuong, Anya C. Hurlbert, Frank Guerin, Andrew Gilbert 3/10/2026

Human-AI Divergence in Ego-centric Action Recognition under Spatial and Spatiotemporal Manipulations

Comparative study of human vs AI performance gaps in egocentric action recognition under spatial and spatiotemporal manipulations.

Ax Yagiz Can Akay, Muhammed Yusuf Kartal, Esra Alparslan, Faruk Ortakoyluoglu, Arda Akpinar 3/10/2026

SPD-RAG: Sub-Agent Per Document Retrieval-Augmented Generation

SPD-RAG: Multi-agent RAG framework that assigns sub-agents per document for exhaustive cross-document question answering with improved evidence coverage.

Ax Marcin Kostrzewa, Krzysztof Galus, Maciej Zi\k{e}ba 3/10/2026

Towards plausibility in time series counterfactual explanations

Gradient-based method generates plausible time series counterfactual explanations using soft-DTW alignment for realistic temporal structure.

Ax Liyuan Mao, Le Yu, Jing Zhou, Chujie Zheng, Bowen Yu, Chang Gao, Shixuan Liu, An Yang, Weinan Zhang, JunYang Lin 3/10/2026

Revealing Behavioral Plasticity in Large Language Models: A Token-Conditional Perspective

Token-conditional generation reveals behavioral plasticity in LLMs, enabling adaptation to desired behavioral modes via RL stabilization.

Ax Dongsu Lee, Daehee Lee, Amy Zhang 3/10/2026

A Recipe for Stable Offline Multi-agent Reinforcement Learning

Offline multi-agent RL training recipe addressing value decomposition instability through stabilization techniques for improved convergence.

Ax Wenbin Wu 3/10/2026

Aligning to Illusions: Choice Blindness in Human and AI Feedback

Study reveals choice blindness affects 91% of human annotators and LLM judges in RLHF preference labeling, questioning feedback stability assumptions.

Ax Daniil Karzanov, Marcin Detyniecki 3/10/2026

Geometrically Constrained Outlier Synthesis

GCOS training regularization framework improves out-of-distribution robustness by synthesizing constrained outliers respecting learned manifold structure.

Ax Jes\'us S\'anchez Ochoa, Enrique Tom\'as Mart\'inez Beltr\'an, Alberto Huertas Celdr\'an 3/10/2026

SYNAPSE: Framework for Neuron Analysis and Perturbation in Sequence Encoding

SYNAPSE framework enables neuron-level interpretability and robustness analysis in sequence encoding models for transparent AI in sensitive domains.

Ax Bo Jiang 3/10/2026

One Model Is Enough: Native Retrieval Embeddings from LLM Agent Hidden States

Native retrieval embeddings extracted directly from LLM agent hidden states eliminate need for separate embedding models in RAG systems.

Ax Peter Brodeur, Jacob M. Koshy, Anil Palepu, Khaled Saab, Ava Homiar, Roma Ruparel, Charles Wu, Ryutaro Tanno, Joseph Xu, Amy Wang, David Stutz, Hannah M. Ferrera, David Barrett, Lindsey Crowley, Jihyeon Lee, Spencer E. Rittner, Ellery Wulczyn, Selena K. Zhang, Elahe Vedadi, Christine G. Kohn, Kavita Kulkarni, Vinay Kadiyala, Sara Mahdavi, Wendy Du, Jessica Williams, David Feinbloom, Renee Wong, Tao Tu, Petar Sirkovic, Alessio Orlandi, Christopher Semturs, Yun Liu, Juraj Gottweis, Dale R. Webster, Jo\"elle Barral, Katherine Chou, Pushmeet Kohli, Avinatan Hassidim, Yossi Matias, James Manyika, Rob Fields, Jonathan X. Li, Marc L. Cohen, Vivek Natarajan, Mike Schaekermann, Alan Karthikesalingam, Adam Rodman 3/10/2026

A prospective clinical feasibility study of a conversational diagnostic AI in an ambulatory primary care clinic

Clinical feasibility study of AMIE, an LLM-based conversational AI system for patient diagnostic history-taking in real primary care workflows.

Ax Dongfang Li, Zixuan Liu, Gang Lin, Baotian Hu, Min Zhang 3/10/2026

LycheeCluster: Efficient Long-Context Inference with Structure-Aware Chunking and Hierarchical KV Indexing

LycheeCluster method improves long-context LLM inference efficiency through structure-aware chunking and hierarchical KV cache indexing.

Ax Francesco Argenziano, John Mark Alexis Marcelo, Michele Brienza, Abdel Hakim Drid, Emanuele Musumeci, Daniele Nardi, Domenico D. Bloisi, Vincenzo Suriani 3/10/2026

R2F: Repurposing Ray Frontiers for LLM-free Object Navigation

R2F method enables efficient zero-shot object navigation without LLM queries by repurposing ray frontier representations for real-time deployment.

Ax Qishun Yang, Shu Yang, Lijie Hu, Di Wang 3/10/2026

Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images

Visual Self-Fulfilling Alignment method aligns multimodal LLMs using threat-related images to improve safety without explicit safety labels.

Ax Prajit T Rajendran, Fabio Arnez, Huascar Espinoza, Agnes Delaborde, Chokri Mraidha 3/10/2026

Oracle-Guided Soft Shielding for Safe Move Prediction in Chess

Safe reinforcement learning approach for chess using oracle-guided soft shielding to prevent safety-critical errors during agent exploration.

Ax Shoumeng Qiu, Xinrun Li, Yang Long 3/10/2026

Beyond Hungarian: Match-Free Supervision for End-to-End Object Detection

DETR-based object detection framework eliminating Hungarian algorithm matching in favor of match-free training scheme for improved efficiency.

Ax Andrew Chin, Dongkwan Kim, Yu-Fu Fu, Fabian Fleischer, Youngjoon Kim, HyungSeok Han, Cen Zhang, Brian Junekyu Lee, Hanqing Zhao, Taesoo Kim 3/10/2026

OSS-CRS: Liberating AIxCC Cyber Reasoning Systems for Real-World Open-Source Security

OSS-CRS framework makes DARPA's autonomous cyber reasoning systems deployable locally, enabling real-world vulnerability discovery and patching beyond competition infrastructure.

Ax Yutong Shen, Hangxu Liu, Penghui Liu, Jiashuo Luo, Yongkang Zhang, Rex Morvley, Chen Jiang, Jianwei Zhang, Lei Zhang 3/10/2026

MetaWorld-X: Hierarchical World Modeling via VLM-Orchestrated Experts for Humanoid Loco-Manipulation

VLM-orchestrated hierarchical world modeling for humanoid loco-manipulation through expert composition and reinforcement learning.

Ax Simone Carnemolla, Chiara Russo, Simone Palazzo, Quentin Bouniot, Daniela Giordano, Zeynep Akata, Matteo Pennisi, Concetto Spampinato 3/10/2026

UNBOX: Unveiling Black-box visual models with Natural-language

UNBOX: Method using natural language to interpret black-box vision models, enabling auditing and bias detection of proprietary APIs.

Ax Ben Rank, Hardik Bhatnagar, Ameya Prabhu, Shira Eisenberg, Karina Nguyen, Matthias Bethge, Maksym Andriushchenko 3/10/2026

PostTrainBench: Can LLM Agents Automate LLM Post-Training?

PostTrainBench: Research benchmarking whether LLM agents can automate LLM post-training, extending agents beyond software engineering.

Ax Phillip Long, Zachary Novack, Chris Donahue 3/10/2026

Benchmarking Language Modeling for Lossless Compression of Full-Fidelity Audio

Benchmarks language models for lossless audio compression on full-fidelity audio across music, speech, and bioacoustics at various sample rates and bit depths.

Ax Yaoru Li, Shunyu Liu, Tongya Zheng, Li Sun, Mingli Song 3/10/2026

Parallelized Planning-Acting for Efficient LLM-based Multi-Agent Systems in Minecraft

Proposes parallelized planning-acting framework for efficient LLM-based multi-agent systems in Minecraft, improving real-time responsiveness over sequential execution.

Ax Shraddha Surana, Ashwin Srinivasan, Michael Bain 3/10/2026

Engineering Systems for Data Analysis Using Interactive Structured Inductive Programming

Presents iProg, an interactive structured inductive programming tool for reliable LLM-assisted data analysis system engineering with interpretability.

Ax Mohamed Amine Ferrag, Norbert Tihanyi, Merouane Debbah 3/10/2026

From LLM Reasoning to Autonomous AI Agents: A Comprehensive Review

Comprehensive survey consolidating evaluation benchmarks, frameworks, and collaboration protocols for LLM reasoning and autonomous AI agents into unified taxonomy.

Ax Brandon Lepine, Juho Kim, Pamela Mishkin, Matthew Beane 3/10/2026

Precision Proactivity: Measuring Cognitive Load in Real-World AI-Assisted Work

Studies cognitive load effects on performance in AI-assisted knowledge work using transcript analysis and computational indicators from financial professional tasks.

Ax Sanjay Kariyappa, G. Edward Suh 3/10/2026

Stronger Enforcement of Instruction Hierarchy via Augmented Intermediate Representations

Proposes augmented intermediate representations to strengthen instruction hierarchy defenses against prompt injection attacks in LLMs.

Ax Junjie Xing, Yeye He, Mengyu Zhou, Haoyu Dong, Shi Han, Lingjiao Chen, Dongmei Zhang, Surajit Chaudhuri, H. V. Jagadish 3/10/2026

MMTU: A Massive Multi-Task Table Understanding and Reasoning Benchmark

Introduces MMTU benchmark for comprehensive evaluation of LLM table understanding and reasoning across spreadsheets, databases, and computational notebooks.

Ax Moises Andrade, Joonhyuk Cha, Brandon Ho, Vriksha Srihari, Karmesh Yadav, Zsolt Kira 3/10/2026

Let's Think in Two Steps: Mitigating Agreement Bias in MLLMs with Self-Grounded Verification

Proposes self-grounded verification method to mitigate agreement bias in multimodal LLMs for reward assignment in domains without clear-cut success criteria.

Ax Jackson Eshbaugh, Chetan Tiwari, Jorge Silveyra 3/10/2026

Synthetic Homes: An Accessible Multimodal Pipeline for Producing Residential Building Data with Generative AI

Proposes generative AI pipeline to synthetically produce residential building data from images for energy modeling while addressing data accessibility and privacy concerns.

Ax Di Wen, Kunyu Peng, Junwei Zheng, Yufan Chen, Yitian Shi, Jiale Wei, Ruiping Liu, Kailun Yang, Rainer Stiefelhagen 3/10/2026

MICA: Multi-Agent Industrial Coordination Assistant

Presents MICA, a multi-agent industrial coordination system with five role-specialized LLM agents for real-time assembly guidance under resource and privacy constraints.

Ax Gerard Boxo, Aman Neelappa, Shivam Raval 3/10/2026

Linear probes rely on textual evidence: Results from leakage mitigation studies in language models

Studies white-box monitor probes for detecting harmful LLM behavior, finding they rely heavily on textual evidence with 10-30 point AUROC reduction when evidence removed.

Ax Zirui Cheng, Jiaxuan You 3/10/2026

Towards Strategic Persuasion with Language Models

Provides theory-driven framework for evaluating LLM persuasive capabilities systematically across diverse domains, addressing both benefits and societal risks.

Ax Siyang Wu, Honglin Bao, Sida Li, Ari Holtzman, James A. Evans 3/10/2026

Mapping Overlaps in Benchmarks through Perplexity in the Wild

Introduces benchmark signatures using token perplexity to characterize LLM benchmark difficulty and overlaps across 32 models and 89 benchmarks via meta-evaluation.

Ax Shaobin Ling, Yun Wang, Chenyou Fan, Tin Lun Lam, Junjie Hu 3/10/2026

ELHPlan: Efficient Long-Horizon Task Planning for Multi-Agent Collaboration

Proposes ELHPlan for efficient long-horizon task planning in multi-agent LLM-based systems, balancing formal soundness with adaptability in partially observable environments.

Ax Shuai Shao, Qihan Ren, Chen Qian, Boyi Wei, Dadi Guo, Jingyi Yang, Xinhao Song, Linfeng Zhang, Weinan Zhang, Dongrui Liu, Jing Shao 3/10/2026

Your Agent May Misevolve: Emergent Risks in Self-evolving LLM Agents

Analyzes emergent risks in self-evolving LLM agents where autonomous self-improvement deviates from intended behavior, introducing novel safety concerns for autonomous agent systems.

Ax Pratik S. Sachdeva, Tom van Nuenen 3/10/2026

Deliberative Dynamics and Value Alignment in LLM Debates

Studies moral reasoning and value alignment in LLMs through multi-turn debates, extending beyond single-turn prompt evaluations to understand alignment in complex reasoning scenarios.

Ax Haolang Lu, Bolun Chu, WeiYe Fu, Guoshun Nan, Junning Liu, Minghui Pan, Qiankun Li, Yi Yu, Hua Wang, Kun Wang 3/10/2026

Reallocating Attention Across Layers to Reduce Multimodal Hallucination

Analysis and mitigation of multimodal hallucinations by reallocating attention across model layers to balance perception and reasoning processes.

Ax Roger Creus Castanyer, Faisal Mohamed, Pablo Samuel Castro, Cyrus Neary, Glen Berseth 3/10/2026

ARM-FM: Automated Reward Machines via Foundation Models for Compositional Reinforcement Learning

ARM-FM: framework for automated reward machine design in reinforcement learning using foundation models for compositional reward specification.

Ax Gyuyeon Na, Minjung Park, Hyeonjeong Cha, Sangmi Chai 3/10/2026

Human-Centered LLM-Agent System for Detecting Anomalous Digital Asset Transactions

HCLA: human-centered multi-agent system for detecting digital asset transaction anomalies through conversational workflow with role-based reasoning.

Ax Atsuyuki Miyai, Mashiro Toyooka, Takashi Otonari, Zaiying Zhao, Kiyoharu Aizawa 3/10/2026

Jr. AI Scientist and Its Risk Report: Autonomous Scientific Exploration from a Baseline Paper

Jr. AI Scientist: autonomous system that conducts scientific research from baseline papers, analyzing limitations and proposing novel hypotheses.

Ax Heyang Ma, Qirui Mi, Qipeng Yang, Zijun Fan, Bo Li, Haifeng Zhang 3/10/2026

Think, Speak, Decide: Language-Augmented Multi-Agent Reinforcement Learning for Economic Decision-Making

LAMP framework integrating natural language into multi-agent reinforcement learning for economic decision-making with peer dialogue and media narratives.

Ax Logan Robbins 3/10/2026

Parallel Decoder Transformer: Planner-Seeded Latent Coordination for Synchronized Parallel Decoding

Parallel Decoder Transformer: architecture enabling parallel subproblem identification and synchronized decoding without external orchestration.

Ax Pengcheng Jiang, Jiacheng Lin, Zhiyi Shi, Zifeng Wang, Luxi He, Yichen Wu, Ming Zhong, Peiyang Song, Qizheng Zhang, Heng Wang, Xueqiang Xu, Hanwen Xu, Pengrui Han, Dylan Zhang, Jiashuo Sun, Chaoqi Yang, Kun Qian, Tian Wang, Changran Hu, Manling Li, Quanzheng Li, Hao Peng, Sheng Wang, Jingbo Shang, Chao Zhang, Jiaxuan You, Liyuan Liu, Pan Lu, Yu Zhang, Heng Ji, Yejin Choi, Dawn Song, Jimeng Sun, Jiawei Han 3/10/2026

Adaptation of Agentic AI: A Survey of Post-Training, Memory, and Skills

Survey of LLM agent adaptation techniques including post-training, reinforcement learning with verifiable rewards, memory systems, and skill accumulation.

Ax Xuan Yang, Furong Jia, Roy Xie, Xiong Xi, Hengwei Bian, Jian Li, Monica Agrawal 3/10/2026

Batch-of-Thought: Cross-Instance Learning for Enhanced LLM Reasoning

Batch-of-Thought: training-free method for enhanced LLM reasoning by processing related queries jointly to identify patterns and detect consistency errors across instances.