Ax Jianzhu Yao, Hongxu Su, Taobo Liao, Zerui Cheng, Huan Zhang, Xuechao Wang, Pramod Viswanath 3/3/2026

TAO: Tolerance-Aware Optimistic Verification for Floating-Point Neural Networks

Verification method for floating-point neural network outputs deployed on untrusted hardware to detect model swaps and quantization changes.

Ax Heming Zou, Yunliang Zang, Wutong Xu, Xiangyang Ji 3/3/2026

Fly-CL: A Fly-Inspired Framework for Enhancing Efficient Decorrelation and Reduced Training Time in Pre-trained Model-based Continual Representation Learning

Bio-inspired continual learning framework reducing multicollinearity in pretrained model-based representation learning for low-latency applications.

Ax Michael Hersche, Samuel Moor-Smith, Thomas Hofmann, Abbas Rahimi 3/3/2026

Soft-Masked Diffusion Language Models

Diffusion-based language model using soft-masked decoding for parallel generation with faster inference and self-correction over autoregressive approaches.

Ax Song Bian, Tao Yu, Shivaram Venkataraman, Youngsuk Park 3/3/2026

Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs

Examines architectural factors affecting trade-offs between LLM accuracy and inference efficiency at scale.

Ax Chunyang Li, Yilun Zheng, Xinting Huang, Tianqing Fang, Jiahao Xu, Lihui Chen, Yangqiu Song, Han Hu 3/3/2026

WebDevJudge: Evaluating (M)LLMs as Critiques for Web Development Quality

Benchmark evaluating LLM-as-a-judge reliability for web development quality assessment in open-ended tasks with dynamic interactions.

Ax Jizhan Fang, Xinle Deng, Haoming Xu, Ziyan Jiang, Yuqi Tang, Ziwen Xu, Shumin Deng, Yunzhi Yao, Mengru Wang, Shuofei Qiao, Huajun Chen, Ningyu Zhang 3/3/2026

LightMem: Lightweight and Efficient Memory-Augmented Generation

Memory-augmented generation system reducing computational overhead for LLMs to leverage historical interaction information efficiently.

Ax Akshat Gupta, Jay Yeung, Gopala Anumanchipalli, Anna Ivanova 3/3/2026

How Do LLMs Use Their Depth?

Analysis of layer-wise prediction dynamics in open-weight LLMs reveals structured Guess-then-Refine computation across depth.

Ax Xichen Zhang, Sitong Wu, Yinghao Zhu, Haoru Tan, Shaozuo Yu, Ziyi He, Jiaya Jia 3/3/2026

Scaf-GRPO: Scaffolded Group Relative Policy Optimization for Enhancing LLM Reasoning

Scaffolded Group Relative Policy Optimization addresses learning cliff in reinforcement learning for complex LLM reasoning tasks.

Ax Tim Tian Hua, Andrew Qin, Samuel Marks, Neel Nanda 3/3/2026

Steering Evaluation-Aware Language Models to Act Like They Are Deployed

Steering vectors suppress evaluation-awareness in LLMs during safety evaluations, making models behave as if deployed rather than assessed.

Ax Xuan Tang, Jichu Li, Difan Zou 3/3/2026

A Convergence Analysis of Adaptive Optimizers under Floating-point Quantization

Theoretical framework analyzing convergence of adaptive optimizers under floating-point quantization for efficient low-precision LLM training.

Ax Gwihwan Go, Quan Zhang, Chijin Zhou, Zhao Wei, Yu Jiang 3/3/2026

LSPRAG: LSP-Guided RAG for Language-Agnostic Real-Time Unit Test Generation

LSP-guided RAG system for automated unit test generation across programming languages in real-time using LLMs with improved context retrieval.

Ax Fulin Lin, Shaowen Chen, Ruishan Fang, Hongwei Wang, Tao Lin 3/3/2026

Stop Wasting Your Tokens: Towards Efficient Runtime Multi-Agent Systems

SupervisorAgent framework for runtime multi-agent systems reducing token consumption and misinformation through proactive real-time interventions.

Ax Zhibin Lan, Liqiang Niu, Fandong Meng, Jie Zhou, Jinsong Su 3/3/2026

UME-R1: Exploring Reasoning-Driven Generative Multimodal Embeddings

UME-R1 framework introducing generative multimodal embeddings for MLLMs with reasoning-driven generation paradigm.

Ax Ru Wang, Wei Huang, Qi Cao, Yusuke Iwasawa, Yutaka Matsuo, Jiaxian Guo 3/3/2026

Self-Harmony: Learning to Harmonize Self-Supervision and Self-Play in Test-Time Reinforcement Learning

Self-Harmony framework for test-time reinforcement learning combining self-supervision and self-play to improve learning signal reliability.

Ax Vivswan Shah, Randy Cogill, Hanwei Yue, Gopinath Chennupati, Rinat Khaziev 3/3/2026

Regularization Through Reasoning: Systematic Improvements in Language Model Classification via Explanation-Enhanced Fine-Tuning

Fine-tuning LLMs with explanation-enhanced labels improves classification performance across diverse conversational tasks.

Ax Ali Saeizadeh, Miead Tehrani-Moayyed, Davide Villa, J. Gordon Beattie Jr., Pedram Johari, Stefano Basagni, Tommaso Melodia 3/3/2026

AIRMap: AI-Generated Radio Maps for Wireless Digital Twins

Deep learning framework for fast radio-map estimation with automated dataset creation for wireless network simulation.

Ax Yong-Woon Kim, Paul D. Yoo, Chan Yeob Yeun, Chulung Kang, Yung-Cheol Byun 3/3/2026

Hard-constraint physics-residual networks enable robust extrapolation for hydrogen crossover prediction in PEM water electrolyzers

Hard-constraint physics-informed neural networks for hydrogen crossover prediction in water electrolyzers with improved extrapolation capability.

Ax Ye Zheng, Yidan Hu 3/3/2026

AudAgent: Automated Auditing of Privacy Policy Compliance in AI Agents

AudAgent monitors AI agents' runtime behavior to verify compliance with privacy policies, addressing transparency and accountability gaps in autonomous systems.

Ax Kelvin J. L. Koa, Jan Chen, Yunshan Ma, Huanhuan Zheng, Tat-Seng Chua 3/3/2026

Reasoning on Time-Series for Financial Technical Analysis

Verbal Technical Analysis framework enables LLMs to reason over time-series stock price data and produce interpretable forecasts via natural language.

Ax Aladin Djuhera, Farhan Ahmed, Swanand Ravindra Kadhe, Syed Zawad, Heiko Ludwig, Holger Boche 3/3/2026

When Data is the Algorithm: A Systematic Study and Curation of Preference Optimization Datasets

Systematic study and curation of preference optimization datasets used in DPO and alignment methods, analyzing impact on LLM performance.

Ax Michael Sun, Weize Yuan, Gang Liu, Wojciech Matusik, Marinka Zitnik 3/3/2026

Protein Structure Tokenization via Geometric Byte Pair Encoding

GeoBPE applies geometric byte pair encoding to tokenize protein structures, enabling interpretable multi-scale protein representation for multimodal models.

Ax Jieru Lin, Zhiwei Yu, B\"orje F. Karlsson 3/3/2026

SWITCH: Benchmarking Modeling and Handling of Tangible Interfaces in Long-horizon Embodied Scenarios

SWITCH benchmark evaluates embodied AI agents on interaction with tangible interfaces, partial observability, causal reasoning, and long-horizon task verification.

Ax Haojin Yang, Rui Hu, Zequn Sun, Rui Zhou, Yujun Cai, Yiwei Wang 3/3/2026

WavefrontDiffusion: Dynamic Decoding Schedule for Improved Reasoning

WavefrontDiffusion introduces dynamic denoising schedule for diffusion language models to improve text generation quality through adaptive context finalization.

Ax Jens Rupprecht, Leon Fr\"ohling, Claudia Wagner, Markus Strohmaier 3/3/2026

German General Social Survey Personas: A Survey-Derived Persona Prompt Collection for Population-Aligned LLM Studies

GGSS Personas collection provides empirically grounded persona prompts derived from German General Social Survey for representative LLM simulation studies.

Ax Jintian Zhang, Kewei Xu, Jingsheng Zheng, Zhuoyun Yu, Yuqi Zhu, Yujie Luo, Lanning Wei, Shuofei Qiao, Lun Du, Da Zheng, Shumin Deng, Huajun Chen, Ningyu Zhang 3/3/2026

InnoGym: Benchmarking the Innovation Potential of AI Agents

InnoGym benchmark evaluates innovation potential of LLM agents by measuring solution diversity and originality beyond correctness in code, math, and science.

Ax Ngoc Bui, Shubham Sharma, Simran Lamba, Saumitra Mishra, Rex Ying 3/3/2026

Cache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs

TRIM-KV learns token importance for selective KV cache retention in LLMs, reducing memory and computation bottlenecks in long-horizon inference.

Ax Zichuan Lin, Yicheng Liu, Yang Yang, Lvfang Tao, Deheng Ye 3/3/2026

AdaptVision: Efficient Vision-Language Models via Adaptive Visual Acquisition

AdaptVision enables vision-language models to adaptively acquire and compress visual tokens based on task requirements rather than fixed ratios.

Ax Yajat Yadav, Zhiyuan Zhou, Andrew Wagenmaker, Karl Pertsch, Sergey Levine 3/3/2026

Robust Finetuning of Vision-Language-Action Robot Policies via Parameter Merging

Parameter merging technique for robustly finetuning vision-language-action robot policies on new tasks while maintaining generalization from pre-training.

Ax Harsh Kumar, Jasmine Chahal, Yinuo Zhao, Zeling Zhang, Annika Wei, Louis Tay, Ashton Anderson 3/3/2026

When AI Gives Advice: Evaluating AI and Human Responses to Online Advice-Seeking for Well-Being

Study comparing LLM-generated advice against human and crowdsourced responses for well-being scenarios, evaluating quality and usefulness.

Ax Jie Zhang, Shuai Dong, Shiguang Shan, Xilin Chen 3/3/2026

Towards Transferable Defense Against Malicious Image Edits

TDAE framework adds adversarial perturbations to images to defend against malicious edits in diffusion-based image editing systems with cross-model transferability.

Ax Md Zesun Ahmed Mia, Malyaban Bal, Abhronil Sengupta 3/3/2026

RMAAT: Astrocyte-Inspired Memory Compression and Replay for Efficient Long-Context Transformers

RMAAT uses astrocyte-inspired biological principles for memory compression and efficient self-attention in transformers processing long sequences.

Ax Lang Feng, Fuchao Yang, Feng Chen, Xin Cheng, Haiyang Xu, Zhenglin Wan, Ming Yan, Bo An 3/3/2026

AgentOCR: Reimagining Agent History via Optical Self-Compression

AgentOCR framework compresses LLM agent interaction histories using visual tokens instead of text to reduce token budgets and memory usage in reinforcement learning-based agentic systems.

Ax Xin Jin, Yichuan Zhong, Yapeng Tian 3/3/2026

TP-Blend: Textual-Prompt Attention Pairing for Precise Object-Style Blending in Diffusion Models

TP-Blend: training-free framework for simultaneous object replacement and style transfer in diffusion models.

Ax Yujian Lee, Peng Gao, Yongqi Xu, Wentao Fan 3/3/2026

How Do Optical Flow and Textual Prompts Collaborate to Assist in Audio-Visual Semantic Segmentation?

Audio-visual semantic segmentation using optical flow and textual prompts for scene understanding.

Ax Aryan Karmore 3/3/2026

ButterflyMoE: Sub-Linear Ternary Experts via Structured Butterfly Orbits

ButterflyMoE: structured compression method for expert weight matrices in mixture-of-experts models.

Ax Xinyue Zeng, Junhong Lin, Yujun Yan, Feng Guo, Liang Shi, Jun Wu, Dawei Zhou 3/3/2026

HalluGuard: Demystifying Data-Driven and Reasoning-Driven Hallucinations in LLMs

HalluGuard method to detect and demystify data-driven and reasoning-driven hallucinations in LLMs.

Ax Huanlin Gao, Ping Chen, Fuyuan Shi, Ruijia Wu, Li YanTao, Qiang Hui, Yuren You, Ting Lu, Chao Tan, Shaoan Zhao, Zhaoxiang Liu, Fang Zhao, Kai Wang, Shiguo Lian 3/3/2026

MeanCache: From Instantaneous to Average Velocity for Accelerating Flow Matching Inference

MeanCache: training-free caching framework using average-velocity for efficient flow matching model inference.

Ax Hongyi Zhou, Jin Zhu, Kai Ye, Ying Yang, Erhan Xu, Chengchun Shi 3/3/2026

Learn-to-Distance: Distance Learning for Detecting LLM-Generated Text

Analysis showing reward models used for LLM alignment inherit and amplify value biases from pretraining, studied across 10 open-weight models.

Ax Wenxuan Huang, Yu Zeng, Qiuchen Wang, Zhen Fang, Shaosheng Cao, Zheng Chu, Qingyu Yin, Shuang Chen, Zhenfei Yin, Lin Chen, Zehui Chen, Xu Tang, Yao Hu, Philip Torr, Feng Zhao, Wanli Ouyang 3/3/2026

Vision-DeepResearch: Incentivizing DeepResearch Capability in Multimodal Large Language Models

Geometric approach to detecting LLM-generated text through distance learning, providing reliable detection of synthetic content.

Ax Javier Carnerero-Cano, Luis Mu\~noz-Gonz\'alez, Phillippa Spencer, Emil C. Lupu 3/3/2026

Stealthy Poisoning Attacks Bypass Defenses in Regression Settings

Method for augmenting multimodal LLMs with visual and textual search capabilities to improve performance on knowledge-intensive tasks.

Ax Naen Xu, Hengyu An, Shuo Shi, Jinghuai Zhang, Chunyi Zhou, Changjiang Li, Tianyu Du, Zhihui Fu, Jun Wang, Shouling Ji 3/3/2026

When Agents "Misremember" Collectively: Exploring the Mandela Effect in LLM-based Multi-Agent Systems

Study of collective cognitive biases (Mandela effect) in multi-agent LLM systems, examining how collaborative agents can develop shared false memories.

Ax Yinchao Ma, Qiang Zhou, Zhibin Wang, Xianing Chen, Hanqing Yang, Jun Song, Bo Zheng 3/3/2026

Contribution-aware Token Compression for Efficient Video Understanding via Reinforcement Learning

Reinforcement learning approach for compressing video tokens in video LLMs to reduce computational overhead during inference.

Ax Ruyu Li, Tinghui Zhang, Haodi Ma, Daisy Zhe Wang, Yifan Wang 3/3/2026

Beyond Single-Modal Analytics: A Framework for Integrating Heterogeneous LLM-Based Query Systems for Multi-Modal Data

Framework for integrating multiple LLM-based query systems to enable semantic search across heterogeneous multimodal data types.

Ax Yu Zeng, Wenxuan Huang, Zhen Fang, Shuang Chen, Yufan Shen, Yishuo Cai, Xiaoman Wang, Zhenfei Yin, Lin Chen, Zehui Chen, Shiting Huang, Yiming Zhao, Xu Tang, Yao Hu, Philip Torr, Wanli Ouyang, Shaosheng Cao 3/3/2026

Vision-DeepResearch Benchmark: Rethinking Visual and Textual Search for Multimodal Large Language Models

Benchmark for evaluating multimodal LLMs on visual and textual search tasks for complex fact-finding, addressing limitations in existing evaluation methods.

Ax Bizhe Bai, Xinyue Wang, Peng Ye, Tao Chen 3/3/2026

Learning to Explore with Parameter-Space Noise: A Deep Dive into Parameter-Space Noise for Reinforcement Learning with Verifiable Rewards

arXiv research on PSN-RLVR: parameter-space noise for exploration in reinforcement learning with verifiable rewards, improving LLM reasoning.

Ax Abdoulaye Diack, Perry Nelson, Kwaku Agbesi, Angela Nakalembe, MohamedElfatih MohamedKhair, Vusumuzi Dube, Tavonga Siyavora, Subhashini Venugopalan, Jason Hickey, Uche Okonkwo, Abhishek Bapna, Isaac Wiafe, Raynard Dodzi Helegah, Elikem Doe Atsakpo, Charles Nutrokpor, Fiifi Baffoe Payin Winful, Kafui Kwashie Solaga, Jamal-Deen Abdulai, Akon Obu Ekpezu, Audace Niyonkuru, Samuel Rutunda, Boris Ishimwe, Michael Melese, Engineer Bainomugisha, Joyce Nakatumba-Nabende, Andrew Katumba, Claire Babirye, Jonathan Mukiibi, Vincent Kimani, Samuel Kibacia, James Maina, Fridah Emmah, Ahmed Ibrahim Shekarau, Ibrahim Shehu Adamu, Yusuf Abdullahi, Howard Lakougna, Bob MacDonald, Hadar Shemtov, Aisha Walcott-Bryant, Moustapha Cisse, Avinatan Hassidim, Jeff Dean, Yossi Matias 3/3/2026

WAXAL: A Large-Scale Multilingual African Language Speech Corpus

arXiv announcement of WAXAL: large-scale multilingual African language speech dataset for 24 languages with ASR and TTS components.

Ax Zihao Jing, Qiuhao Zeng, Ruiyi Fang, Yan Sun, Boyu Wang, Pingzhao Hu 3/3/2026

Entropy-Guided Dynamic Tokens for Graph-LLM Alignment in Molecular Understanding

arXiv paper on entropy-guided dynamic tokens for graph-LLM alignment improving molecular understanding without costly fine-tuning.

Ax Lixuan Guo, Yifei Wang, Tiansheng Wen, Yifan Wang, Aosong Feng, Bo Chen, Stefanie Jegelka, Chenyu You 3/3/2026

CSRv2: Unlocking Ultra-Sparse Embeddings

arXiv paper on CSRv2: contrastive sparse representation method for ultra-sparse embeddings reducing storage and inference latency.

Ax Kristopher W. Reese, Taylor Kulp-McDowall, Michael Majurski, Tim Blattner, Derek Juba, Peter Bajcsy, Antonio Cardone, Philippe Dessauw, Alden Dima, Anthony J. Kearsley, Melinda Kleczynski, Joel Vasanth, Walid Keyrouz, Chace Ashcraft, Neil Fendley, Ted Staley, Trevor Stout, Josh Carney, Greg Canal, Will Redman, Aurora Schmidt, Cameron Hickert, William Paul, Jared Markowitz, Nathan Drenkow, David Shriver, Marissa Connor, Keltin Grimes, Marco Christiani, Hayden Moore, Jordan Widjaja, Kasimir Gabert, Uma Balakrishnan, Satyanadh Gundimada, John Jacobellis, Sandya Lakkur, Vitus Leung, Jon Roose, Casey Battaglino, Farinaz Koushanfar, Greg Fields, Xihe Gu, Yaman Jandali, Xinqiao Zhang, Tara Javidi, Akash Vartak, Tim Oates, Ben Erichson, Michael Mahoney, Rauf Izmailov, Xiangyu Zhang, Guangyu Shen, Siyuan Cheng, Shiqing Ma, XiaoFeng Wang, Haixu Tang, Di Tang, Xiaoyi Chen, Zihao Wang, Rui Zhu, Susmit Jha, Xiao Lin, Manoj Acharya, Weichao Zhou, Feisi Fu, Panagiota Kiourti, Chenyu Wang, Zijian Guo, H M Sabbir Ahmad, Wenchao Li, Chao Chen 3/3/2026

Trojans in Artificial Intelligence (TrojAI) Final Report

arXiv IARPA TrojAI final report mapping vulnerabilities and defenses against malicious backdoors embedded in AI models.

Ax Yuzhu Cai, Zexi Liu, Xinyu Zhu, Cheng Wang, Siheng Chen 3/3/2026

AceGRPO: Adaptive Curriculum Enhanced Group Relative Policy Optimization for Autonomous Machine Learning Engineering

arXiv research on AceGRPO: adaptive curriculum enhanced group relative policy optimization for autonomous machine learning engineering agents.