Ax Xinyue Ma, Heelim Hong, Taegeon Um, Jongseop Lee, Seoyeong Choy, Woo-Yeon Lee, Myeongjae Jeon 3/3/2026

OrbitFlow: SLO-Aware Long-Context LLM Serving with Fine-Grained KV Cache Reconfiguration

OrbitFlow system dynamically reconfigures KV cache across CPU/GPU for long-context LLM serving with adaptive runtime memory management.

Ax Zixuan Ke, Yifei Ming, Austin Xu, Ryan Chin, Xuan-Phi Nguyen, Prathyusha Jwalapuram, Jiayu Wang, Semih Yavuz, Caiming Xiong, Shafiq Joty 3/3/2026

MAS-Orchestra: Understanding and Improving Multi-Agent Reasoning Through Holistic Orchestration and Controlled Benchmarks

MAS-Orchestra framework for orchestrating multi-agent systems with holistic reasoning and controlled benchmarks for evaluating agent coordination.

Ax Heewoong Noh, Gyoung S. Na, Namkyeong Lee, Chanyoung Park 3/3/2026

MSP-LLM: A Unified Large Language Model Framework for Complete Material Synthesis Planning

MSP-LLM unified framework for material synthesis planning using LLMs, addressing precursor identification and operation sequencing for materials discovery.

Ax Furong Jia, Ling Dai, Wenjin Deng, Fan Zhang, Chen Hu, Daxin Jiang, Yu Liu 3/3/2026

SpotAgent: Grounding Visual Geo-localization in Large Vision-Language Models through Agentic Reasoning

SpotAgent grounds vision-language models in geo-localization through agentic reasoning with grounded tool use for sparse and ambiguous visual cues.

Ax Jiaquan Zhang, Chaoning Zhang, Shuxu Chen, Xudong Wang, Zhenzhen Huang, Pengcheng Zheng, Shuai Yuan, Sheng Zheng, Qigan Sun, Jie Zou, Lik-Hang Lee, Yang Yang 3/3/2026

Learning Global Hypothesis Space for Enhancing Synergistic Reasoning Chain

Global Hypothesis Space method improves Chain-of-Thought reasoning by enabling holistic exploration of solution space rather than greedy step-by-step generation.

Ax Nanxu Gong, Haotian Li, Sixun Dong, Jianxun Lian, Yanjie Fu, Xing Xie 3/3/2026

To Think or Not To Think, That is The Question for Large Reasoning Models in Theory of Mind Tasks

Studies whether Large Reasoning Models transfer step-by-step inference benefits to Theory of Mind tasks assessing belief and intention inference.

Ax Tianxiang Xu, Jiayi Liu, Yixuan Tong, Jialu Xu, Yunqing Wei, Kaiwen Feng, PanPan Hou, Kangping Yin, Jiyuan Hu, Hao Zhou, Zhenxin Ma, Jian Xu, Guanjun Jiang 3/3/2026

Quark Medical Alignment: A Holistic Multi-Dimensional Alignment and Collaborative Optimization Paradigm

Multi-dimensional alignment framework for medical LLMs combining RLHF and RLVR paradigms for high-stakes question answering accuracy.

Ax Yiheng Shu, Saisri Padmaja Jonnalagedda, Xiang Gao, Bernal Jim\'enez Guti\'errez, Weijian Qi, Kamalika Das, Huan Sun, Yu Su 3/3/2026

REMem: Reasoning with Episodic Memory in Language Agent

REMem introduces episodic memory and spatiotemporal reasoning to language agents, enabling recollection and reasoning over interaction histories.

Ax Lu\'is Silva, Diogo Gon\c{c}alves, Catarina Farinha, Clara Matos, Lu\'is Ungaro 3/3/2026

Arbor: A Framework for Reliable Navigation of Critical Conversation Flows

Arbor framework decomposes decision tree navigation for LLMs in high-stakes workflows like healthcare triage, preventing instruction-following degradation.

Ax Sushant Mehta, Logan Ritchie, Suhaas Garre, Ian Niebres, Nick Heiner, Edwin Chen 3/3/2026

EnterpriseBench Corecraft: Training Generalizable Agents on High-Fidelity RL Environments

CoreCraft RL environment simulates enterprise customer support organization for training generalizable agentic AI with 2500+ entities and 23 tools.

Ax Shengtian Yang, Yu Li, Shuo He, Yewen Li, Qingpeng Cai, Peng Jiang, Lei Feng 3/3/2026

Phase-Aware Mixture of Experts for Agentic Reinforcement Learning

Phase-Aware Mixture of Experts architecture for RL-trained LLM agents, addressing simplicity bias in policy networks for complex task solving.

Ax Adrian Robert Minut, Hazem Dewidar, Iacopo Masi 3/3/2026

Spilled Energy in Large Language Models

Interprets LLM softmax layer as Energy-Based Model to track 'energy spills' correlating with factual errors and biases during decoding.

Ax Xuhui Ren, Shaokang Dong, Chen Yang, Qing Gao, Yunbin Zhao, Yongsheng Liu, Xinwei Geng, Xiang Li, Demei Yan, Yanqing Li, Chenhao Huang, Dingwei Zhu, Junjie Ye, Boxuan Yue, Yingnan Fu, Mengzhe Lv, Zezeng Feng, Boshen Zhou, Bocheng Wang, Xuanjing Huang, Yu-Gang Jiang, Tao Gui, Qi Zhang, Yunke Zhang 3/3/2026

MagicAgent: Towards Generalized Agent Planning

MagicAgent framework enables generalized agent planning across heterogeneous tasks using LLMs, addressing data scarcity and task conflicts.

Ax Dezhi Kong, Zhengzhao Feng, Qiliang Liang, Hao Wang, Haofei Sun, Changpeng Yang, Yang Li, Peng Zhou, Shuai Nie, Hongzhen Wang, Linfeng Zhou, Hao Jia, Jiaming Xu, Runyu Shi, Ying Huang 3/3/2026

ProactiveMobile: A Comprehensive Benchmark for Boosting Proactive Intelligence on Mobile Devices

ProactiveMobile benchmark evaluates proactive intelligence in mobile AI agents that anticipate user needs and autonomously initiate actions.

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

SideQuest: Model-Driven KV Cache Management for Long-Horizon Agentic Reasoning

SideQuest framework optimizes KV cache management for long-horizon agentic reasoning tasks requiring multi-hop reasoning over retrieved documents.

Ax Xiaoxi Li, Wenxiang Jiao, Jiarui Jin, Shijian Wang, Guanting Dong, Jiajie Jin, Hao Wang, Yinuo Wang, Ji-Rong Wen, Yuan Lu, Zhicheng Dou 3/3/2026

OmniGAIA: Towards Native Omni-Modal AI Agents

OmniGAIA benchmark for evaluating multi-modal AI agents combining vision, audio, and language with reasoning and tool usage capabilities.

Ax Aishik Sanyal 3/3/2026

ReCoN-Ipsundrum: An Inspectable Recurrent Persistence Loop Agent with Affect-Coupled Control and Mechanism-Linked Consciousness Indicator Assays

Agent architecture with recurrent persistence loop and affect proxy for machine consciousness research. Theoretical work on inspectable AI agents.

Ax Radha Sarma 3/3/2026

Agency and Architectural Limits: Why Optimization-Based Systems Cannot Be Norm-Responsive

Formal analysis demonstrating optimization-based systems and LLMs trained via RLHF cannot be norm-responsive agents without architectural changes.

Ax Xiang Li, Jiabao Gao, Sipei Lin, Xuan Zhou, Chi Zhang, Bo Cheng, Jiale Han, Benyou Wang 3/3/2026

Human or Machine? A Preliminary Turing Test for Speech-to-Speech Interaction

First Turing test for speech-to-speech conversational systems, evaluating 9 state-of-the-art systems against human participants.

Ax Dang Nguyen, Paymon Haddad, Eric Gan, Baharan Mirzasoleiman 3/3/2026

Changing the Training Data Distribution to Reduce Simplicity Bias Improves In-distribution Generalization

Study on modifying training data distribution to reduce simplicity bias and improve in-distribution generalization for neural networks.

Ax Eilam Shapira, Omer Madmon, Itamar Reinman, Samuel Joseph Amouyal, Roi Reichart, Moshe Tennenholtz 3/3/2026

GLEE: A Unified Framework and Benchmark for Language-based Economic Environments

GLEE framework and benchmark for evaluating LLM behavior in economic and strategic interactions requiring natural language communication.

Ax Wanrong Yang, Alberto Acuto, Yihang Zhou, Dominik Wojtczak 3/3/2026

A Survey for Deep Reinforcement Learning Based Network Intrusion Detection

Survey of deep reinforcement learning approaches for network intrusion detection systems reviewing DRL frameworks and recent applications.

Ax Yonatan Sverdlov, Yair Davidson, Nadav Dym, Tal Amir 3/3/2026

FSW-GNN: A Bi-Lipschitz WL-Equivalent Graph Neural Network

FSW-GNN graph neural network architecture achieving Weisfeiler-Leman equivalence with improved separation quality using bi-Lipschitz constraints.

Ax Hakan Aktas, Yigit Yildirim, Ahmet Firat Gamsiz, Deniz Bilge Akkoc, Erhan Oztop, Emre Ugur 3/3/2026

Neuro-Symbolic Skill Discovery for Conditional Multi-Level Planning

Neuro-symbolic architecture for discovering high-level action symbols from low-level skill demonstrations using neural networks and symbolic planning.

Ax Yuling Gu, Oyvind Tafjord, Hyunwoo Kim, Jared Moore, Ronan Le Bras, Peter Clark, Yejin Choi 3/3/2026

SimpleToM: Exposing the Gap between Explicit ToM Inference and Implicit ToM Application in LLMs

SimpleToM benchmark exposing gap between LLMs' explicit theory-of-mind reasoning and implicit application in predicting human behavior.

Ax Ruhan Wang, Yu Yang, Zhishuai Liu, Dongruo Zhou, Pan Xu 3/3/2026

Return Augmented Decision Transformer for Off-Dynamics Reinforcement Learning

Decision Transformer approach for offline reinforcement learning using return-conditioned supervised learning across different data distributions.

Ax Jingyu Xiao, Yuxuan Wan, Yintong Huo, Zixin Wang, Xinyi Xu, Wenxuan Wang, Zhiyao Xu, Yuhang Wang, Michael R. Lyu 3/3/2026

Interaction2Code: Benchmarking MLLM-based Interactive Webpage Code Generation from Interactive Prototyping

Interaction2Code benchmark for evaluating MLLMs on generating interactive webpage code from prototypes with dynamic user interactions.

Ax Jie Zhang, Xiangkui Cao, Zhouyu Han, Shiguang Shan, Xilin Chen 3/3/2026

Multi-PA: A Multi-perspective Benchmark on Privacy Assessment for Large Vision-Language Models

Multi-PA benchmark assessing privacy risks and preservation capabilities of large vision-language models across multiple dimensions.

Ax Ismail Khalfaoui-Hassani, Stefan Kesselheim 3/3/2026

Polynomial, trigonometric, and tropical activations

Study of polynomial, trigonometric, and tropical activation functions as alternatives to standard neural network activations.

Ax Jack Hong, Shilin Yan, Jiayin Cai, Xiaolong Jiang, Yao Hu, Weidi Xie 3/3/2026

WorldSense: Evaluating Real-world Omnimodal Understanding for Multimodal LLMs

WorldSense benchmark for evaluating multimodal LLMs on video understanding tasks requiring synchronized visual, audio, and text inputs.

Ax Shruti Joshi, Andrea Dittadi, S\'ebastien Lachapelle, Dhanya Sridhar 3/3/2026

Sparse Shift Autoencoders for Identifying Concepts from Large Language Model Activations

Sparse shift autoencoders for interpretability of LLM activations, enabling unsupervised discovery and identification of internal concepts.

Ax Tianjin Huang, Zhangyang Wang, Haotian Hu, Zhenyu Zhang, Gaojie Jin, Xiang Li, Li Shen, Jiaxing Shang, Tianlong Chen, Ke Li, Lu Liu, Qingsong Wen, Shiwei Liu 3/3/2026

GradientStabilizer:Fix the Norm, Not the Gradient

GradientStabilizer method addressing training instability from gradient-norm spikes in deep learning without requiring threshold tuning.

Ax Siming Huang, Yuliang Xu, Mingmeng Geng, Yao Wan, Dongping Chen 3/3/2026

Wikipedia in the Era of LLMs: Evolution and Risks

Framework analyzing LLM impact on Wikipedia content and page views with simulations exploring potential risks from automated contributions.

Ax Emmy Liu, Amanda Bertsch, Lintang Sutawika, Lindia Tjuatja, Patrick Fernandes, Lara Marinov, Michael Chen, Shreya Singhal, Carolin Lawrence, Aditi Raghunathan, Kiril Gashteovski, Graham Neubig 3/3/2026

Not-Just-Scaling Laws: Towards a Better Understanding of the Downstream Impact of Language Model Design Decisions

Meta-analysis of 92 open-source pretrained LLMs examining how architectural decisions and data curation impact model performance beyond just scale.

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

LLaVE: Large Language and Vision Embedding Models with Hardness-Weighted Contrastive Learning

Research on LLaVE multimodal embedding models using hardness-weighted contrastive learning. Improves distinguishing hard negatives in vision-language tasks.

Ax Kibum Kim, Sein Kim, Hongseok Kang, Jiwan Kim, Heewoong Noh, Yeonjun In, Kanghoon Yoon, Jinoh Oh, Julian McAuley, Chanyoung Park 3/3/2026

Token-Efficient Item Representation via Images for LLM Recommender Systems

Token-efficient item representation via images for LLM-based recommender systems, balancing efficiency and effectiveness in item encoding.

Ax Wenxuan Huang, Bohan Jia, Zijie Zhai, Shaosheng Cao, Zheyu Ye, Fei Zhao, Zhe Xu, Xu Tang, Yao Hu, Shaohui Lin 3/3/2026

Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models

Vision-R1 applies reinforcement learning to enhance reasoning capabilities in multimodal large language models, building on DeepSeek-R1-Zero.

Ax Zisheng Chen, Chunwei Wang, Runhui Huang, Hongbin Xu, Xiuwei Chen, Jun Zhou, Jianhua Han, Hang Xu, Xiaodan Liang 3/3/2026

SemHiTok: A Unified Image Tokenizer via Semantic-Guided Hierarchical Codebook for Multimodal Understanding and Generation

SemHiTok proposes a unified image tokenizer using semantic-guided hierarchical codebook for multimodal understanding and generation tasks.

Ax Zangwei Zheng, Xiangyu Peng, Yuxuan Lou, Chenhui Shen, Tom Young, Xinying Guo, Binluo Wang, Hang Xu, Hongxin Liu, Mingyan Jiang, Wenjun Li, Yuhui Wang, Anbang Ye, Gang Ren, Qianran Ma, Wanying Liang, Xiang Lian, Xiwen Wu, Yuting Zhong, Zhuangyan Li, Chaoyu Gong, Guojun Lei, Leijun Cheng, Limin Zhang, Minghao Li, Ruijie Zhang, Silan Hu, Shijie Huang, Xiaokang Wang, Yuanheng Zhao, Yuqi Wang, Ziang Wei, Yang You 3/3/2026

Open-Sora 2.0: Training a Commercial-Level Video Generation Model in $200k

Open-Sora 2.0 demonstrates training a commercial-grade video generation model for $200k, addressing cost-effectiveness in large-scale model training.

Ax Wenqiang Wang, Yijia Zhang, Zikai Zhang, Guanting Huo, Hao Liang, Shijie Cao, Ningyi Xu 3/3/2026

ROMA: a Read-Only-Memory-based Accelerator for QLoRA-based On-Device LLM

ROMA: specialized hardware accelerator for QLoRA-based on-device LLM inference with hybrid read-only-memory architecture for privacy-preserving deployment.

Ax Arsham Gholamzadeh Khoee, Shuai Wang, Yinan Yu, Robert Feldt, Dhasarathy Parthasarathy 3/3/2026

GateLens: A Reasoning-Enhanced LLM Agent for Automotive Software Release Analytics

GateLens: LLM agent framework for automotive software analytics combining reasoning with structured tabular data handling and ambiguity resolution capabilities.

Ax Chanhyuk Lee, Jiho Choi, Chanryeol Lee, Donggyun Kim, Seunghoon Hong 3/3/2026

AdaRank: Adaptive Rank Pruning for Enhanced Model Merging

AdaRank: adaptive rank selection method for SVD-based model merging that reduces cross-task interference and improves multi-task learning performance.

Ax Nan Zhang, Eugene Kwek, Yusen Zhang, Ngoc-Hieu Nguyen, Prasenjit Mitra, Rui Zhang 3/3/2026

When Reasoning Meets Compression: Understanding the Effects of LLMs Compression on Large Reasoning Models

Analysis of compression effects (quantization, distillation, pruning) on large reasoning model capabilities with mechanistic interpretation and performance benchmarking.

Ax Rachmad Vidya Wicaksana Putra, Avaneesh Devkota, Muhammad Shafique 3/3/2026

Scalable Multi-Task Learning through Spiking Neural Networks with Adaptive Task-Switching Policy for Intelligent Autonomous Agents

Spiking neural network approach for multi-task reinforcement learning in resource-constrained autonomous agents with adaptive task-switching to reduce interference.

Ax Kazuki Fujii, Yukito Tajima, Sakae Mizuki, Masaki Kawamura, Hinari Shimada, Taihei Shiotani, Koshiro Saito, Masanari Oi, Taishi Nakamura, Takumi Okamoto, Shigeki Ishida, Kakeru Hattori, Youmi Ma, Hiroya Takamura, Rio Yokota, Jun Sakuma, Naoaki Okazaki 3/3/2026

Rewriting Pre-Training Data Boosts LLM Performance in Math and Code

SwallowCode and MathShallow datasets with systematically rewritten pre-training data significantly improve LLM performance in code and math reasoning under open licenses.

Ax Philip Naumann, Jacob Kauffmann, Gr\'egoire Montavon 3/3/2026

Wasserstein Distances Made Explainable: Insights Into Dataset Shifts and Transport Phenomena

Explainability framework for Wasserstein distances analyzing dataset shifts and transport phenomena by decomposing distance contributions.

Ax Alberto Bassi, Marco Baity-Jesi, Aurelien Lucchi, Carlo Albert, Emanuele Francazi 3/3/2026

When Bias Meets Trainability: Connecting Theories of Initialization

Theoretical analysis connecting initialization bias and trainability in deep neural networks through mean field theory and gradient behavior characterization.

Ax Yuhui Wang, Rongyi Zhu, Ting Wang 3/3/2026

Self-Destructive Language Model

Defense mechanism against harmful fine-tuning attacks on LLMs by reducing model trainability on harmful data while maintaining safety and normal performance.

Ax Guangda Liu, Chengwei Li, Zhenyu Ning, Jing Lin, Yiwu Yao, Danning Ke, Minyi Guo, Jieru Zhao 3/3/2026

FreeKV: Boosting KV Cache Retrieval for Efficient LLM Inference

FreeKV: efficient KV cache retrieval method for long-context LLM inference that improves compression methods and reduces accuracy loss during context extension.

Ax Jingcong Liang, Siyuan Wang, Miren Tian, Yitong Li, Duyu Tang, Zhongyu Wei 3/3/2026

Not All Models Suit Expert Offloading: On Local Routing Consistency of Mixture-of-Expert Models

Analysis of expert offloading in Mixture-of-Experts LLMs on memory-constrained devices, examining local routing consistency and practical deployment challenges.