challenge.json

{"papers":[{"i":3768,"pid":"61998","orid":"kpgURPRMGf","title":"The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models","authors":["Zanlin Ni","Shenzhi Wang","Yang Yue","Tianyu Yu","Weilin Zhao","Yeguo Hua","Tianyi Chen","Jun Song","YuCheng","Bo Zheng","Gao Huang"],"insts":["Tsinghua University","Department of Automation, Tsinghua University","Tsinghua University, Tsinghua University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=kpgURPRMGf","vs":"https://icml.cc/virtual/2026/poster/61998","arxiv":"2601.15165","award":"Outstanding Paper Award","alphaxiv":"2601.15165"},{"i":4146,"pid":"71132","orid":"71132","title":"High-accuracy sampling for diffusion models and log-concave distributions","authors":["Fan Chen","Sinho Chewi","Constantinos Daskalakis","Alexander Rakhlin"],"insts":["Massachusetts Institute of Technology","MIT"],"area":"Uncategorized","sub":"","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71132","arxiv":"2602.01338","award":"Outstanding Paper Award","alphaxiv":"2602.01338"},{"i":5341,"pid":"71065","orid":"71065","title":"The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception Probes","authors":["Mohammad Taufeeque","Stefan Heimersheim","Adam Gleave","Chris Cundy"],"insts":["FAR.AI","Google DeepMind"],"area":"Social Aspects","sub":"Alignment","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71065","arxiv":"2602.15515","award":"Outstanding Paper Honorable Mention","alphaxiv":"2602.15515"},{"i":2395,"pid":"71049","orid":"71049","title":"Motion Attribution for Video Generation","authors":["Xindi Wu","Despoina Paschalidou","Jun Gao","Antonio Torralba","Laura Leal-Taixé","Olga Russakovsky","Sanja Fidler","Jonathan Lorraine"],"insts":["Princeton University","NVIDIA","MIT"],"area":"Uncategorized","sub":"","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71049","arxiv":"2601.08828","award":"Outstanding Paper Honorable Mention","alphaxiv":"2601.08828"},{"i":5851,"pid":"62989","orid":"bA6BgSbaUi","title":"How much can language models memorize?","authors":["John Morris","Chawin Sitawarin","Narine Kokhlikyan","Chuan Guo","Edward Suh","Alexander Rush","Kamalika Chaudhuri","Saeed Mahloujifar"],"insts":["Cornell University","Anthropic","Facebook"],"area":"Uncategorized","sub":"","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=bA6BgSbaUi","vs":"https://icml.cc/virtual/2026/poster/62989","arxiv":"2505.24832","award":"Outstanding Paper Honorable Mention","alphaxiv":"2505.24832"},{"i":6588,"pid":"62241","orid":"iPjuUQbkfl","title":"A Random Matrix Perspective on the Consistency of Diffusion Models","authors":["Binxu Wang","Jacob A Zavatone-Veth","Cengiz Pehlevan"],"insts":["Harvard University"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=iPjuUQbkfl","vs":"https://icml.cc/virtual/2026/poster/62241","arxiv":"2602.02908","award":"Outstanding Paper Honorable Mention","alphaxiv":"2602.02908"},{"i":4513,"pid":"66206","orid":"5nNNVY8NW4","title":"To Grok Grokking: Provable Grokking in Ridge Regression","authors":["Mingyue Xu","Gal Vardi","Itay Safran"],"insts":["Purdue University","Weizmann Institute of Science","Ben-Gurion University of the Negev"],"area":"Theory","sub":"Learning Theory","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=5nNNVY8NW4","vs":"https://icml.cc/virtual/2026/poster/66206","arxiv":"2601.19791","award":"Outstanding Paper Honorable Mention","alphaxiv":"2601.19791"},{"i":5964,"pid":"64450","orid":"NUyt4uxzx0","title":"Chain-of-Thought Reasoning In The Wild Is Not Always Faithful","authors":["Iván Arcuschin","Jett Janiak","Robert Krzyzanowski","Senthooran Rajamanoharan","Neel Nanda","Arthur Conmy"],"insts":["Poseidon Research","ML Alignment & Theory Scholars","MATS(Neel/Nanda)"],"area":"Social Aspects","sub":"Accountability, Transparency, and Interpretability","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=NUyt4uxzx0","vs":"https://icml.cc/virtual/2026/poster/64450","arxiv":"2503.08679","alphaxiv":"2503.08679"},{"i":3396,"pid":"64287","orid":"P7RGcAOZZ3","title":"Geometry-Aware Dataset Condensation for Diffusion Model Training","authors":["Xiao Cui","Yulei Qin","Mo Zhu","Wengang Zhou","Hongsheng Li","Houqiang Li"],"insts":["University of Science and Technology of China","Tencent","Zhejiang University"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=P7RGcAOZZ3","vs":"https://icml.cc/virtual/2026/poster/64287","arxiv":"2606.05883","alphaxiv":"2606.05883"},{"i":5766,"pid":"62951","orid":"bWLfplRNzt","title":"ProtDBench: A Unified Benchmark of Protein Binder Design and Evaluation","authors":["Cong Liu","Milong Ren","Jiaqi Guan","Chengyue Gong","Jinyuan Sun","Xinshi Chen","Wenzhi Xiao"],"insts":["University of Amsterdam, University of Amsterdam","Institute of Computing Technology","ByteDance Inc."],"area":"Applications","sub":"Health / Medicine","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=bWLfplRNzt","vs":"https://icml.cc/virtual/2026/poster/62951","arxiv":"2605.04118","alphaxiv":"2605.04118"},{"i":2243,"pid":"64011","orid":"Rl2uQlCoQX","title":"SPEED-Bench: A Unified and Diverse Benchmark for Speculative Decoding","authors":["Talor Abramovich","Maor Ashkenazi","Izzy Putterman","Benjamin Chislett","Tiyasa Mitra","Bita Darvish Rouhani","Ran Zilberstein","Yonatan Geifman"],"insts":["NVIDIA, Tel Aviv University","NVIDIA","Technion"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=Rl2uQlCoQX","vs":"https://icml.cc/virtual/2026/poster/64011","arxiv":"2604.09557","alphaxiv":"2604.09557"},{"i":5788,"pid":"66362","orid":"4M5Kj2UqaM","title":"AgentSelect: Benchmark for Narrative Query-to-Agent Recommendation","authors":["Yunxiao Shi","Wujiang Xu","Tingwei Chen","Haoning Shang","Ling Yang","Yunfeng Wan","Zhuo Cao","Xing Zi","Dimitris Metaxas","Min Xu"],"insts":["University of Technology Sydney","Rutgers University","Alibaba Group"],"area":"Applications","sub":"Everything Else","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=4M5Kj2UqaM","vs":"https://icml.cc/virtual/2026/poster/66362","arxiv":"2603.03761","alphaxiv":"2603.03761"},{"i":4878,"pid":"60936","orid":"vCc2NAe0OS","title":"A Semantically Consistent Dataset for Data-Efficient Query-Based Universal Sound Separation","authors":["Kai Li","Jintao Cheng","Chang Zeng","Zijun Yan","Helin Wang","Zixiong Su","Bo Zheng","Xiaolin Hu"],"insts":["Tsinghua University","National Institute of Informatics","Johns Hopkins University"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=vCc2NAe0OS","vs":"https://icml.cc/virtual/2026/poster/60936","arxiv":"2601.22599","alphaxiv":"2601.22599"},{"i":43,"pid":"60687","orid":"xbAWn0w9kq","title":"DIYHealth Suite: Dataset, Model, and Benchmark for Health Management at Home","authors":["Changshuo Liu","Wu Junran","Zhongle Xie","Wenqiao Zhang","Kaiping Zheng","Jiaqi Zhu","Qingpeng Cai","Gene Anne Ooi","Marcus CJ Tan","Jianwei Yin","James Yip","Beng Chin Ooi"],"insts":["national university of singaore, National University of Singapore","National University of Singapore","Zhejiang University"],"area":"Applications","sub":"Health / Medicine","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=xbAWn0w9kq","vs":"https://icml.cc/virtual/2026/poster/60687","arxiv":"2606.07542","alphaxiv":"2606.07542"},{"i":523,"pid":"64918","orid":"If4X4W2HWx","title":"MemoryBench: A Benchmark for Memory and Continual Learning in LLM Systems","authors":["Qingyao Ai","Yichen Tang","Changyue Wang","Jianming Long","Weihang Su","Yiqun LIU"],"insts":["Tsinghua University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=If4X4W2HWx","vs":"https://icml.cc/virtual/2026/poster/64918","arxiv":"2510.17281","alphaxiv":"2510.17281"},{"i":3671,"pid":"60614","orid":"yU6X1XZl8t","title":"QuArch: A Benchmark for Evaluating LLM Reasoning in Computer Architecture","authors":["Shvetank Prakash","Andrew Cheng","Arya Tschand","Mark Mazumder","Varun Gohil","Jeffrey Ma","Jason Yik","Zishen Wan","Jessica A. Quaye","Elisavet Alvanaki","Avinash Kumar","Chandrashis Mazumdar","Tuhin Khare","Alexander Ingare","Ikechukwu Uchendu","Radhika Ghosal","Abhishek Tyagi","Chenyu Wang","Andrea Mattia Garavagno","Sarah Gu","Alice Guo","Grace Hur","Luca Carloni","Tushar Krishna","Ankita Nayak","Amir Yazdanbakhsh","Vijay Janapa Reddi"],"insts":["Harvard University","Harvard","Massachusetts Institute of Technology"],"area":"Applications","sub":"Everything Else","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=yU6X1XZl8t","vs":"https://icml.cc/virtual/2026/poster/60614","arxiv":"2510.22087","alphaxiv":"2510.22087"},{"i":1347,"pid":"61559","orid":"ov240fehF6","title":"MORE: A Multilingual Document Parsing Benchmark and Evaluation","authors":["Long Xu","Binghong Wu","TingHao YU","Hao Feng","zhenyuhuang","Haoqing Jiang","Yunhao Wang","Shuo Huang","feng zhang"],"insts":["Tencent Technology","Tencent Hunyuan","Tencent Hunyuan Team"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=ov240fehF6","vs":"https://icml.cc/virtual/2026/poster/61559","arxiv":"2607.02956","alphaxiv":"2607.02956"},{"i":4454,"pid":"66658","orid":"1ebAvNphi7","title":"PipeSD: An Efficient Cloud-Edge Collaborative Pipeline Inference Framework with Speculative Decoding","authors":["Yunhe Han","Yunqi Gao","Bing Hu","Mahdi Boloursaz Mashhadi","Yitong Duan","Pei Xiao","Yanfeng Zhang"],"insts":["Zhejiang University","University of Surrey","Zhongguancun Institute of Artificial Intelligence"],"area":"Deep Learning","sub":"Algorithms","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=1ebAvNphi7","vs":"https://icml.cc/virtual/2026/poster/66658","arxiv":"2605.13319","alphaxiv":"2605.13319"},{"i":5762,"pid":"63759","orid":"Tyv61ZKb9s","title":"SynerMedGen: Synergizing Medical Multimodal Understanding with Generation via Task Alignment","authors":["Weiren Zhao","DONG Yi","Cheng Chen"],"insts":["University of Hong Kong"],"area":"Applications","sub":"Health / Medicine","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=Tyv61ZKb9s","vs":"https://icml.cc/virtual/2026/poster/63759","alphaxiv":"2605.08724"},{"i":6750,"pid":"60806","orid":"wTgx7b2D9r","title":"FRIGID: Scaling Diffusion-Based Molecular Generation from Mass Spectra at Training and Inference Time","authors":["Montgomery Bohde","Hongxuan Liu","Mrunali Manjrekar","Magdalena Lederbauer","Shuiwang Ji","Runzhong Wang","Connor Coley"],"insts":["Texas A&M University","Massachusetts Institute of Technology","MIT"],"area":"Applications","sub":"Chemistry, Physics, and Earth Sciences","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=wTgx7b2D9r","vs":"https://icml.cc/virtual/2026/poster/60806","alphaxiv":"2604.16648"},{"i":174,"pid":"61140","orid":"tJP3FxzSPs","title":"video-SALMONN S: Memory-Enhanced Streaming Audio-Visual LLM","authors":["Guangzhi Sun","Yixuan Li","Xiaodong Wu","Yudong Yang","Wei Li","Zejun MA","Chao Zhang"],"insts":["University of Cambridge","Tsinghua University","Bytedance"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=tJP3FxzSPs","vs":"https://icml.cc/virtual/2026/poster/61140","alphaxiv":"2510.11129"},{"i":1462,"pid":"61269","orid":"rzBAQT2Fkg","title":"d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation","authors":["Yu-Yang Qian","Junda Su","Lanxiang Hu","Peiyuan Zhang","Zhijie Deng","Peng Zhao","Hao Zhang"],"insts":["Nanjing University","University of California, San Diego","Shanghai Jiao Tong University"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=rzBAQT2Fkg","vs":"https://icml.cc/virtual/2026/poster/61269","arxiv":"2601.07568","alphaxiv":"2601.07568"},{"i":1959,"pid":"64223","orid":"PoRigyDOcC","title":"Training–Inference Consistent Segmented Execution for Long-Context LLMs","authors":["Xianpeng Shang","Jiang Li","Zehua Duo","Qianyi Cai","Xiangdong Su"],"insts":["Inner Mongolia University","The Hong Kong University of Science and Technology"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=PoRigyDOcC","vs":"https://icml.cc/virtual/2026/poster/64223","alphaxiv":"2605.11744"},{"i":2362,"pid":"64360","orid":"OMdQJQwp26","title":"MedMosaic: A Challenging Large Scale Benchmark of Diverse Medical Audio","authors":["Harshit Rajgarhia","Shuubham Ojha","Asif Shaik","Akhil Pothanapalli","Rachuri Lokesh","Abhishek Mukherji","Prasanna Desikan"],"insts":["Centific","University of Maryland, College Park","centific"],"area":"Applications","sub":"Health / Medicine","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=OMdQJQwp26","vs":"https://icml.cc/virtual/2026/poster/64360","arxiv":"2605.00969","alphaxiv":"2605.00969"},{"i":2587,"pid":"63366","orid":"XjSd2CtV20","title":"VCG-Bench: Towards A Unified Visual-Centric Benchmark for Structured Generation and Editing","authors":["Xiaoyan Su","Peijie Dong","Zhenheng Tang","Song Tang","Yuyao Zhai","Kaitao Lin","Liang Chen","Gai Yuhang","Yuyu Luo","Qiang Wang","Xiaowen Chu"],"insts":["Shenzhen University","HKUST-GZ","The Hong Kong University of Science and Technology"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=XjSd2CtV20","vs":"https://icml.cc/virtual/2026/poster/63366","arxiv":"2605.15677","alphaxiv":"2605.15677"},{"i":3278,"pid":"63099","orid":"a86luANykT","title":"daVinci-Dev: Agent-native Mid-training for Software Engineering","authors":["Ji Zeng","Dayuan Fu","Tiantian Mi","Zhuang Yumin","Yaxing Huang","Xuefeng Li","Lyumanshan Ye","Muhang Xie","Qishuo Hua","Zhen Huang","Mohan Jiang","Hanning Wang","Shijie Xia","Yang Xiao","Jie Sun","Yunze Wu","Pengfei Liu"],"insts":["Shanghai Jiaotong University","SII; SJTU","Fudan University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=a86luANykT","vs":"https://icml.cc/virtual/2026/poster/63099","arxiv":"2601.18418","alphaxiv":"2601.18418"},{"i":3618,"pid":"62070","orid":"k7XzObg9Hy","title":"FOAM: Blocked State Folding for Memory-Efficient LLM Training","authors":["Ziqing Wen","Jiahuan Wang","ping luo","Dongsheng Li","Tao Sun"],"insts":["National University of Defense Technology","School of Computer Science, National University of Defense Technology","national university of defense technology"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=k7XzObg9Hy","vs":"https://icml.cc/virtual/2026/poster/62070","alphaxiv":"2512.07112"},{"i":4709,"pid":"64004","orid":"RpwnrBkht2","title":"FourTune: Towards Fully 4-Bit Efficient Post-Training for Diffusion Models","authors":["Bowen Xue","Zihan Min","Xingyang Li","Muyang Li","Yujun Lin","Zhekai Zhang","Haocheng Xi","Lvmin Zhang","Maneesh Agrawala","Jun-Yan Zhu","Song Han"],"insts":["University of Science and Technology of China","Tsinghua University","Massachusetts Institute of Technology"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=RpwnrBkht2","vs":"https://icml.cc/virtual/2026/poster/64004","alphaxiv":"2607.05711"},{"i":5243,"pid":"60697","orid":"xRVWftS3ES","title":"Memory is Reconstructed, Not Retrieved:  Graph Memory for LLM Agents","authors":["Shuo Ji","yibo li","Bryan Hooi"],"insts":["National University of Singapore"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=xRVWftS3ES","vs":"https://icml.cc/virtual/2026/poster/60697","arxiv":"2606.06036","alphaxiv":"2606.06036"},{"i":5286,"pid":"60972","orid":"uyRIOjFgOn","title":"Decouple Searching from Training: Scaling Data Mixing via Model Merging for Large Language Model Pre-training","authors":["Shengrui Li","Fei zhao","Kaiyan Zhao","Jieying Ye","Haifeng Liu","Fangcheng Shi","Zheyong Xie","Yao Hu","Shaosheng Cao"],"insts":["Xiaohongshu","The University of Tokyo","Beijing Institute of Technology"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=uyRIOjFgOn","vs":"https://icml.cc/virtual/2026/poster/60972","arxiv":"2602.00747","alphaxiv":"2602.00747"},{"i":5662,"pid":"64446","orid":"NWKaQIKoGp","title":"PlugMem: A Task-Agnostic Plugin Memory Module for LLM Agents","authors":["Ke Yang","Zixi Chen","Xuan He","Jize Jiang","Michel Galley","Chenglong Wang","Jianfeng Gao","Jiawei Han","Chengxiang Zhai"],"insts":["University of Illinois Urbana-Champaign","Tsinghua University","University of Illinois at Urbana-Champaign"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=NWKaQIKoGp","vs":"https://icml.cc/virtual/2026/poster/64446","arxiv":"2603.03296","alphaxiv":"2603.03296"},{"i":266,"pid":"62216","orid":"ic0AGRIkmY","title":"Evolutionary Generation of Multi-Agent Systems","authors":["Yuntong Hu","Matthew Trager","Yuting Zhang","Yi Zhang","Shuo Yang","Wei Xia","Stefano Soatto"],"insts":["Emory University","Amazon","Amazon Web Services"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=ic0AGRIkmY","vs":"https://icml.cc/virtual/2026/poster/62216","arxiv":"2602.06511","alphaxiv":"2602.06511"},{"i":268,"pid":"66192","orid":"5vufrrbi4N","title":"TextAtlas5M: A Large-Scale Dataset for Long Text Image Generation","authors":["Dongxing Mao","Alex Jinpeng Wang","weiming Han","Jiawei Zhang","Zhuobai Dong","Linjie Li","Lin Yiqi","Zhengyuan Yang","Libo Qin","Fuwei Zhang","Lijuan Wang","Min Li"],"insts":["Central South University","Wuhan University","Microsoft"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=5vufrrbi4N","vs":"https://icml.cc/virtual/2026/poster/66192","alphaxiv":"2502.07870"},{"i":338,"pid":"60755","orid":"ww57OvgpP9","title":"SimulCost: A Cost-Aware Benchmark and Toolkit for Automating Physics Simulations with LLMs","authors":["Yadi Cao","Sicheng Lai","Jiahe Huang","Yang Zhang","Zach Lawrence","Rohan Bhakta","Izzy Thomas","Mingyun Cao","Chung-Hao Tsai","Zihao Zhou","Yidong Zhao","Hao Liu","Alessandro Marinoni","Alexey Arefiev","Rose Yu"],"insts":["University of California San Diego","The Chinese University of Hongkong, Shenzhen","UC San Diego & Ai2"],"area":"Applications","sub":"Chemistry, Physics, and Earth Sciences","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=ww57OvgpP9","vs":"https://icml.cc/virtual/2026/poster/60755","arxiv":"2603.20253","alphaxiv":"2603.20253"},{"i":362,"pid":"62816","orid":"czTbPdmvtT","title":"LayerT2V: A Unified Multi-Layer Video Generation Framework","authors":["Guangzhao Li","Kangrui Cen","Baixuan Zhao","Yi Xin","Siqi Luo","Guangtao Zhai","Lei Zhang","Xiaohong Liu"],"insts":["Shanghai Jiao Tong University","OPPO Research Institute","Shanghai Jiaotong University"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=czTbPdmvtT","vs":"https://icml.cc/virtual/2026/poster/62816","alphaxiv":"2508.04228"},{"i":799,"pid":"65908","orid":"913g1X1YAW","title":"CATArena: Evaluating Evolutionary Capabilities of Code Agents via Iterative Tournaments","authors":["Lingyue Fu","Xin Ding","Linyue Pan","Yaoming Zhu","Shao Zhang","Lin Qiu","Weiwen Liu","Weinan Zhang","Xuezhi Cao","Xunliang Cai","Jiaxin Ding","Yong Yu"],"insts":["Shanghai Jiao Tong University","Shanghai Jiaotong University","Tsinghua University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=913g1X1YAW","vs":"https://icml.cc/virtual/2026/poster/65908","alphaxiv":"2510.26852"},{"i":1420,"pid":"63256","orid":"Yhc2PvgFyh","title":"Dual Latent Memory for Visual Multi-agent System","authors":["Xinlei Yu","Chengming Xu","Zhangquan Chen","Bo Yin","Cheng Yang","Yongbo He","Yihao Hu","Jiangning Zhang","Cheng Tan","Xiaobin Hu","Shuicheng YAN"],"insts":["National University of Singapore","Tencent","Tsinghua University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=Yhc2PvgFyh","vs":"https://icml.cc/virtual/2026/poster/63256","alphaxiv":"2602.00471"},{"i":2089,"pid":"63654","orid":"UxgMJP0JKY","title":"Fast Autoregressive Video Diffusion and World Models with Temporal Cache Compression and Sparse Attention","authors":["Dvir Samuel","Issar Tzachor","Matan Levy","Michael Green","Gal Chechik","Rami Ben-Ari"],"insts":["NVIDIA","OriginAI","Hebrew University"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=UxgMJP0JKY","vs":"https://icml.cc/virtual/2026/poster/63654","arxiv":"2602.01801","alphaxiv":"2602.01801"},{"i":2276,"pid":"66755","orid":"0bTEd4LpQr","title":"Numina-Lean-Agent: An Open and General Agentic Reasoning System for Formal Mathematics","authors":["Junqi Liu","Zihao Zhou","Zekai Zhu","Marco Dos Santos","Weikun He","jiawei liu","Yunzhou Xie","Junqiao Zhao","Qiufeng Wang","Lihong Zhi","Jia LI","Wenda Li"],"insts":["Academy of Mathematics and Systems Science, Chinese Academy of Science","Xi'an Jiaotong-Liverpool University","Tongji University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=0bTEd4LpQr","vs":"https://icml.cc/virtual/2026/poster/66755","arxiv":"2601.14027","alphaxiv":"2601.14027"},{"i":2734,"pid":"61638","orid":"oCNT5PcMSQ","title":"How Can I Publish My LLM Benchmark Without Giving the True Answers Away?","authors":["Takashi Ishida","Thanawat Lodkaew","Ikko Yamane"],"insts":["RIKEN / The University of Tokyo","The University of Tokyo","Ecole Nationale de la Statistique et de l'Analyse de l'information"],"area":"General Machine Learning","sub":"Evaluation","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=oCNT5PcMSQ","vs":"https://icml.cc/virtual/2026/poster/61638","arxiv":"2505.18102","alphaxiv":"2505.18102"},{"i":2828,"pid":"65809","orid":"9wYjjPydfe","title":"TIC-VLA: A Think-in-Control Vision-Language-Action Model for Robot Navigation in Dynamic Environments","authors":["Zhiyu Huang","Yun Zhang","Johnson Liu","Rui Song","Chen Tang","Jiaqi Ma"],"insts":["University of California, Los Angeles","University of Texas at Austin"],"area":"Applications","sub":"Robotics","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=9wYjjPydfe","vs":"https://icml.cc/virtual/2026/poster/65809","arxiv":"2602.02459","alphaxiv":"2602.02459"},{"i":3015,"pid":"66319","orid":"4jfuNNghPS","title":"FlashBlock: Attention Caching for Efficient Long-Context Block Diffusion","authors":["Zhuokun Chen","Jianfei Cai","Bohan Zhuang"],"insts":["Monash University","Zhejiang University"],"area":"Deep Learning","sub":"Attention Mechanisms","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=4jfuNNghPS","vs":"https://icml.cc/virtual/2026/poster/66319","arxiv":"2602.05305","alphaxiv":"2602.05305"},{"i":3111,"pid":"66296","orid":"4yzY0GFIJj","title":"Q-Sched: Pushing the Boundaries of Few-Step Diffusion Models with Quantization-Aware Scheduling","authors":["Natalia Frumkin","Diana Marculescu"],"insts":["The University of Texas at Austin"],"area":"Deep Learning","sub":"Algorithms","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=4yzY0GFIJj","vs":"https://icml.cc/virtual/2026/poster/66296","arxiv":"2509.01624","alphaxiv":"2509.01624"},{"i":4149,"pid":"60932","orid":"vGeNaFHdET","title":"EEG-FM-Bench: A Comprehensive Benchmark for the Systematic Evaluation and Diagnostic Analyses of EEG Foundation Models","authors":["Wei Xiong","Jiangtong Li","Jie Li","Kun Zhu","Changjun Jiang"],"insts":["Tongji University"],"area":"Applications","sub":"Neuroscience, Cognitive Science","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=vGeNaFHdET","vs":"https://icml.cc/virtual/2026/poster/60932","alphaxiv":"2508.17742"},{"i":4754,"pid":"63667","orid":"UnjxMTe57e","title":"PostTrainBench: Can LLM Agents Automate LLM Post-Training?","authors":["Ben Rank","Hardik Bhatnagar","Ameya Pandurang Prabhu","Shira Eisenberg","Karina Nguyen","Matthias Bethge","Maksym Andriushchenko"],"insts":["ELLIS Intitute Tübingen","MATS · Max Planck Institute for Intelligent Systems, Tübingen","University of Tübingen"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=UnjxMTe57e","vs":"https://icml.cc/virtual/2026/poster/63667","arxiv":"2603.08640","alphaxiv":"2603.08640"},{"i":4871,"pid":"64383","orid":"O88FCPAPAj","title":"Ranking Free RAG: Replacing Re-ranking with Selection in RAG for Sensitive Domains","authors":["Yash Saxena","Ankur Padia","Mandar Chaudhary","Kalpa Gunaratna","Srinivasan Parthasarathy","Manas Gaur"],"insts":["University of Maryland, Baltimore County","Liberty Mutual Insurance","eBay Inc."],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=O88FCPAPAj","vs":"https://icml.cc/virtual/2026/poster/64383","arxiv":"2505.16014","alphaxiv":"2505.16014"},{"i":5002,"pid":"61543","orid":"p5QSlnwume","title":"Rethinking Video Generation Model for the Embodied World","authors":["Yufan Deng","Zilin Pan","Hongyu Zhang","Xiaojie Li","Huruoqing","Yufei Ding","Yiming Zou","Yan Zeng","Zhou Daquan"],"insts":["Peking University, Bytedance Seed","Peking University","Tiktok"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=p5QSlnwume","vs":"https://icml.cc/virtual/2026/poster/61543","arxiv":"2601.15282","alphaxiv":"2601.15282"},{"i":332,"pid":"65980","orid":"8Fhq7QpYfI","title":"WF-Bench: A Benchmark for Neural-Network WaveFunction Expressivity and Scaling Laws","authors":["Lixing Zhang","Guijing Duan","Di Luo"],"insts":["University of California, Los Angeles","Renmin University of China","Tsinghua University"],"area":"Applications","sub":"Chemistry, Physics, and Earth Sciences","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=8Fhq7QpYfI","vs":"https://icml.cc/virtual/2026/poster/65980","alphaxiv":"2605.29683"},{"i":389,"pid":"65929","orid":"8nti23Zqkt","title":"EGG: An Expert-Guided Agent Framework for Kernel Generation","authors":["Yaochen Han","Ke Fan","Hongxu Jiang","Wanqi Xu","Weiyu Xie","Runhua Zhang","Chenhui Zhu","Yixiang Zhang"],"insts":["Beihang University","Shanghai Jiao Tong University","Beijing University of Aeronautics and Astronautics"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=8nti23Zqkt","vs":"https://icml.cc/virtual/2026/poster/65929","alphaxiv":"2606.26758"},{"i":394,"pid":"66489","orid":"3EcT46wsdc","title":"Efficiently Training Time-to-First-Spike Spiking Neural Networks from Scratch","authors":["Kaiwei Che","Wei Fang","Zhengyu Ma","Yifan Huang","Peng Xue","Li Yuan","Yonghong Tian"],"insts":["Peking University","Yale University","Peng Cheng Lab"],"area":"Applications","sub":"Neuroscience, Cognitive Science","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=3EcT46wsdc","vs":"https://icml.cc/virtual/2026/poster/66489","alphaxiv":"2410.23619"},{"i":749,"pid":"62868","orid":"cPDN8YlKdQ","title":"Beyond VLM-Based Rewards: Diffusion-Native Latent Reward Modeling","authors":["Gongye Liu","Bo Yang","Zhi Yida","Zhizhou Zhong","Lei Ke","Didan Deng","Han Gao","Yongxiang Huang","Kaihao Zhang","Hongbo Fu","Wenhan Luo"],"insts":["The Hong Kong University of Science and Technology","University of Science and Technology of China","Hong Kong University of Science and Technology"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=cPDN8YlKdQ","vs":"https://icml.cc/virtual/2026/poster/62868","arxiv":"2602.11146","alphaxiv":"2602.11146"},{"i":861,"pid":"66638","orid":"1tbhBSXcyX","title":"RelayCaching: Accelerating LLM Collaboration via Decoding KV Cache Reuse","authors":["Yingsheng Geng","Yuchong Gao","Weihong Wu","Guyue Liu","Jiang liu"],"insts":["Beijing University of Posts and Telecommunications","Tsinghua University","University of Electronic Science and Technology of China"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=1tbhBSXcyX","vs":"https://icml.cc/virtual/2026/poster/66638","arxiv":"2603.13289","alphaxiv":"2603.13289"},{"i":892,"pid":"66354","orid":"4P9cEcinYP","title":"Efficient, Property-Aligned Fan-Out Retrieval via RL-Amortized Diffusion","authors":["Pengcheng Jiang","Judith Li","Moonkyung Ryu","Lily Hu","Kun Su","Zhong Yi Wan","Liam Hebert","Hao Peng","Jiawei Han","Dima Kuzmin","Craig Boutilier"],"insts":["University of Illinois at Urbana-Champaign","Google Research","Research, Google"],"area":"Reinforcement Learning","sub":"Deep RL","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=4P9cEcinYP","vs":"https://icml.cc/virtual/2026/poster/66354","alphaxiv":"2603.06397"},{"i":1406,"pid":"61654","orid":"o3gN27ITWV","title":"Long-Context Modeling with Dynamic Hierarchical Sparse Attention for Memory-Constrained LLM Inference","authors":["Siheng Xiong","Joe Zou","Faramarz Fekri","Yae Jee Cho"],"insts":["Georgia Institute of Technology","University of California Berkeley","Google"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=o3gN27ITWV","vs":"https://icml.cc/virtual/2026/poster/61654","alphaxiv":"2510.24606"},{"i":1559,"pid":"60889","orid":"vaRFU0xKQa","title":"Med-SegLens: Latent-Level Model Diffing for Interpretable Medical Image Segmentation","authors":["Salma Ahmed","Emad Mohammed","Azam Bidgoli"],"insts":["Wilfrid Laurier University"],"area":"Applications","sub":"Health / Medicine","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=vaRFU0xKQa","vs":"https://icml.cc/virtual/2026/poster/60889","alphaxiv":"2602.10508"},{"i":1739,"pid":"62660","orid":"ea4sx1kxDz","title":"Speculative Coupled Decoding for Training-Free Lossless Acceleration of Autoregressive Visual Generation","authors":["Junhyuk So","Hyunho Kook","Chaeyeon Jang","Eunhyeok Park"],"insts":["POSTECH","Pohang University of Science and Technology"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=ea4sx1kxDz","vs":"https://icml.cc/virtual/2026/poster/62660","alphaxiv":"2510.24211"},{"i":1868,"pid":"63075","orid":"aJdgt8xDMy","title":"AVGen-Bench: A Task-Driven Benchmark for Multi-Granular Evaluation of Text-to-Audio-Video Generation","authors":["Ziwei Zhou","Zeyuan Lai","Rui Wang","Yifan Yang","Yuqing Yang","Qi Dai","Lili Qiu","Chong Luo"],"insts":["Fudan University","University of Science and Technology of China","Microsoft"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=aJdgt8xDMy","vs":"https://icml.cc/virtual/2026/poster/63075","arxiv":"2604.08540","alphaxiv":"2604.08540"},{"i":2029,"pid":"63517","orid":"WTziQZdpTV","title":"One Tool Is Enough: Reinforcement Learning of LLM Agents for Repository-Level Code Navigation","authors":["Zhaoxi Zhang","Yitong Duan","Yanzhi Zhang","Yiming Xu","Zhixiang Wang","Kun Liang","Yang Li","Jiahui Liang","Deguo Xia","Jizhou Huang","Jiyan He","Shuxin Zheng","Yunfang Wu"],"insts":["Peking University","Zhongguancun Institute of Artificial Intelligence","Beijing Zhongguancun Academy"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=WTziQZdpTV","vs":"https://icml.cc/virtual/2026/poster/63517","alphaxiv":"2512.20957"},{"i":2136,"pid":"63800","orid":"TbUbv5zqTF","title":"SALAAD: Sparse And Low-Rank Adaptation via ADMM for Large Language Model Inference","authors":["Hao Ma","Melis Ilayda Bal","Liang Zhang","Bingcong Li","Niao He","Melanie Zeilinger","Michael Muehlebach"],"insts":["ETH Zurich &                                   Max Planck Institute for Intelligent Systems","EPFL - EPF Lausanne","Department of Computer Science, ETHZ - ETH Zurich"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=TbUbv5zqTF","vs":"https://icml.cc/virtual/2026/poster/63800","alphaxiv":"2602.00942"},{"i":2241,"pid":"64608","orid":"Li5ki5Dopo","title":"Ambient Dataloops: Generative Models for Dataset Refinement","authors":["Adrian Rodriguez-Munoz","William Daspit","Adam Klivans","Antonio Torralba","Constantinos Daskalakis","Giannis Daras"],"insts":["Massachusetts Institute of Technology","University of Texas at Austin","MIT"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=Li5ki5Dopo","vs":"https://icml.cc/virtual/2026/poster/64608","arxiv":"2601.15417","alphaxiv":"2601.15417"},{"i":3074,"pid":"66488","orid":"3ExTD9F0u1","title":"TFRBench: A Reasoning Benchmark for Evaluating Forecasting Systems","authors":["Md Atik Ahamed","Mihir Parmar","Palash Goyal","Yiwen Song","Long T. Le","Qiang (Shaun) Cheng","Chun-Liang Li","Hamid Palangi","Jinsung Yoon","Tomas Pfister"],"insts":["University of Kentucky","Google"],"area":"Applications","sub":"Time Series","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=3ExTD9F0u1","vs":"https://icml.cc/virtual/2026/poster/66488","arxiv":"2604.05364","alphaxiv":"2604.05364"},{"i":3623,"pid":"61640","orid":"oBgLvd5YC6","title":"SimpleMem: Efficient Lifelong Memory for LLM Agents","authors":["Jiaqi Liu","Yaofeng Su","Peng Xia","Siwei Han","Zeyu Zheng","Cihang Xie","Mingyu Ding","Huaxiu Yao"],"insts":["Department of Computer Science, University of North Carolina at Chapel Hill","Fudan University","UNC-Chapel Hill"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=oBgLvd5YC6","vs":"https://icml.cc/virtual/2026/poster/61640","arxiv":"2601.02553","alphaxiv":"2601.02553"},{"i":4122,"pid":"65453","orid":"DVHpvumD60","title":"WarmServe: Enabling One-for-Many GPU Prewarming for Multi-LLM Serving","authors":["Chiheng Lou","Sheng Qi","Rui Kang","Yong Zhang","Chen Sun","pengcheng wang","Xuanzhe Liu","Xin Jin"],"insts":["Peking University","Huawei Technologies Ltd."],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=DVHpvumD60","vs":"https://icml.cc/virtual/2026/poster/65453","alphaxiv":"2512.09472"},{"i":4334,"pid":"64264","orid":"PLT2FKIs2c","title":"Light Up Your Face: A Physically Consistent Dataset and Diffusion Model for Face Fill-Light Enhancement","authors":["Jue Gong","Zihan Zhou","Jingkai Wang","Xiaohong Liu","Yulun Zhang","Xiaokang Yang"],"insts":["Shanghai Jiao Tong University","Shanghai Jiaotong University","Shanghai Jiao Tong University of China"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=PLT2FKIs2c","vs":"https://icml.cc/virtual/2026/poster/64264","alphaxiv":"2602.04300"},{"i":4616,"pid":"64125","orid":"QgoRoKIEEr","title":"Fast and Scalable Analytical Diffusion","authors":["Xinyi Shang","Peng Sun","Jingyu Lin","Zhiqiang Shen"],"insts":["University College London","Westlake University","Monash University"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=QgoRoKIEEr","vs":"https://icml.cc/virtual/2026/poster/64125","alphaxiv":"2602.16498"},{"i":4870,"pid":"64464","orid":"NMMmwSbzRx","title":"Stream RAG: Instant and Accurate Spoken Dialogue Systems with Streaming Tool Usage","authors":["Siddhant Arora","Haidar Khan","Kai Sun","Xin Dong","Sajal Choudhary","Seungwhan Moon","Xinyuan Zhang","Adithya Sagar","Surya Appini","Kaushik Patnaik","Sanat Sharma","Shinji Watanabe","Anuj Kumar","Ahmed A Aly","Yue Liu","Florian Metze","Zhaojiang Lin"],"insts":["Carnegie Mellon University","Rensselaer Polytechnic Institute","Meta"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=NMMmwSbzRx","vs":"https://icml.cc/virtual/2026/poster/64464","arxiv":"2510.02044","alphaxiv":"2510.02044"},{"i":4989,"pid":"63079","orid":"aIFgQusnPy","title":"Diffusion Bridge or Flow Matching? A Unifying Framework and Comparative Analysis","authors":["Kaizhen Zhu","Mokai Pan","Zhechuan Yu","Jingya Wang","Jingyi Yu","Ye Shi"],"insts":["ShanghaiTech University"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=aIFgQusnPy","vs":"https://icml.cc/virtual/2026/poster/63079","alphaxiv":"2509.24531"},{"i":5981,"pid":"64301","orid":"Oz335dV48X","title":"DFlash: Block Diffusion for Flash Speculative Decoding","authors":["Jian Chen","Yesheng Liang","Zhijian Liu"],"insts":["University of California, San Diego","NVIDIA"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=Oz335dV48X","vs":"https://icml.cc/virtual/2026/poster/64301","arxiv":"2602.06036","alphaxiv":"2602.06036"},{"i":6144,"pid":"66223","orid":"5f0gw9YpZC","title":"GUDA: Counterfactual Group-wise Training Data Attribution for Diffusion Models via Unlearning","authors":["Naoki Murata","Yuhta Takida","Chieh-Hsin Lai","Toshimitsu Uesaka","Bac Nguyen","Stefano Ermon","Yuki Mitsufuji"],"insts":["Sony AI","Sony Group Corporation","Stanford University"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=5f0gw9YpZC","vs":"https://icml.cc/virtual/2026/poster/66223","alphaxiv":"2601.22651"},{"i":6407,"pid":"63850","orid":"TBaUfO9znF","title":"On the Interplay of Pre-Training, Mid-Training, and RL on Reasoning Language Models","authors":["Charlie Zhang","Graham Neubig","Xiang Yue"],"insts":["UCSB","Carnegie Mellon University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=TBaUfO9znF","vs":"https://icml.cc/virtual/2026/poster/63850","arxiv":"2512.07783","alphaxiv":"2512.07783"},{"i":6441,"pid":"63497","orid":"Wfe1iJocjF","title":"FlashOptim: Memory Efficient Optimizers for Large-Scale Training","authors":["Jose Javier Gonzalez Ortiz","Abhay Gupta","Christopher Rinard","Davis Blalock"],"insts":["Databricks AI Research","Cerebras Systems, Inc","Massachusetts Institute of Technology"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=Wfe1iJocjF","vs":"https://icml.cc/virtual/2026/poster/63497","alphaxiv":"2602.23349"},{"i":6487,"pid":"63109","orid":"a3GdvuPItd","title":"Selecting Samples on Graphs: A Unified Dataset Pruning Framework for Lossless Training Acceleration","authors":["Dongyue Wu","Zilin Guo","Xiaoyu Li","Jiajia Liu","Jingdong Chen","Nong Sang","Changxin Gao"],"insts":["Huazhong University of Science and Technology","Ant Group"],"area":"Deep Learning","sub":"Algorithms","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=a3GdvuPItd","vs":"https://icml.cc/virtual/2026/poster/63109","alphaxiv":"2606.12913"},{"i":6674,"pid":"61469","orid":"prvGhNz39e","title":"Memory-Efficient LLM Pretraining via Minimalist Optimizer Design","authors":["Athanasios Glentis","Jiaxiang Li","Andi Han","Mingyi Hong"],"insts":["University of Minnesota","Facebook","University of Sydney"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=prvGhNz39e","vs":"https://icml.cc/virtual/2026/poster/61469","alphaxiv":"2506.16659"},{"i":46,"pid":"66431","orid":"3qX5RS8kpJ","title":"Dual-View Predictive Diffusion: Lightweight Speech Enhancement via Spectrogram-Image Synergy","authors":["Ke Xue","Rongfei Fan","Kai Li","Shanping Yu","Puning Zhao","Jianping An"],"insts":["Beijing Institute of Technology","Tsinghua University","SUN YAT-SEN UNIVERSITY"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=3qX5RS8kpJ","vs":"https://icml.cc/virtual/2026/poster/66431","alphaxiv":"2602.00568"},{"i":84,"pid":"61144","orid":"tI5CFbRhmV","title":"Rethinking Code Complexity Through the Lens of Large Language Models","authors":["Chen Xie","Yuling Shi","Xiaodong Gu","Beijun Shen"],"insts":["Shanghai Jiaotong University","Shanghai Jiao Tong University"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=tI5CFbRhmV","vs":"https://icml.cc/virtual/2026/poster/61144","alphaxiv":"2602.07882"},{"i":283,"pid":"62358","orid":"hJnZKtsDbe","title":"AgentExpt: Automating AI Experiment Design with LLM-based Resource Retrieval Agent","authors":["Yu Li","Lehui Li","Lin Chen","Qingmin Liao","Fengli Xu","Yong Li"],"insts":["Tsinghua University","Shandong University","Tsinghua Univeristy"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=hJnZKtsDbe","vs":"https://icml.cc/virtual/2026/poster/62358","alphaxiv":"2511.04921"},{"i":371,"pid":"62533","orid":"fl93PQfTT6","title":"DLLMQuant: A Post-Training Quantization Framework Tailored for Diffusion-Based Large Language Models","authors":["XUCHEN","Dawei Yang","Zhixuan Chen"],"insts":["Nanjing University of Aeronautics and Astronautics","Houmo AI","houmo"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=fl93PQfTT6","vs":"https://icml.cc/virtual/2026/poster/62533","alphaxiv":"2508.14090"},{"i":429,"pid":"65501","orid":"CzShhpY2qU","title":"Agent Primitives: Reuseable Latent Building Blocks for Multi-Agent Systems","authors":["Haibo Jin","Peng Kuang","Ye Yu","Xiaopeng Yuan","Haohan Wang"],"insts":["University of Illinois at Urbana-Champaign","UIUC"],"area":"Reinforcement Learning","sub":"Multi-agent","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=CzShhpY2qU","vs":"https://icml.cc/virtual/2026/poster/65501","alphaxiv":"2602.03695"},{"i":6068,"pid":"67042","orid":"ycj3XWCh6E","title":"Position: LLM Serving Needs Mathematical Optimization and Algorithmic Foundations, Not Just Heuristics","authors":["Zijie Zhou"],"insts":["The Hong Kong University of Science and Technology"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=ycj3XWCh6E","vs":"https://icml.cc/virtual/2026/poster/67042","alphaxiv":"2605.01280"},{"i":6034,"pid":"65267","orid":"FEmXFeqYNZ","title":"Success-Conditioning as Policy Improvement: The Optimization Problem Solved by Imitating Success","authors":["Daniel Russo"],"insts":["Columbia University"],"area":"Reinforcement Learning","sub":"Policy Search","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=FEmXFeqYNZ","vs":"https://icml.cc/virtual/2026/poster/65267","alphaxiv":"2601.18175"},{"i":3120,"pid":"66563","orid":"2azIa9tfl3","title":"Optimizing Rank for High-Fidelity Implicit Neural Representations","authors":["Julian McGinnis","Florian A. Hölzl","Suprosanna Shit","Florentin Bieder","Paul Friedrich","Mark Mühlau","bjoern menze","Daniel Rueckert","Benedikt Wiestler"],"insts":["Technical University of Munich","HPI, TUM","TUM"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=2azIa9tfl3","vs":"https://icml.cc/virtual/2026/poster/66563","alphaxiv":"2512.14366"},{"i":623,"pid":"71171","orid":"71171","title":"$\\tau^2$-Bench: Evaluating Conversational Agents in a Dual-Control Environment","authors":["Victor Barres","Honghua Dong","Soham Ray","Xujie Si","Karthik Narasimhan"],"insts":["Mercor","Huawei","Sierra.ai"],"area":"Deep Learning","sub":"Large Language Models","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71171","alphaxiv":"2506.07982"},{"i":3364,"pid":"71030","orid":"71030","title":"OPUS: Towards Efficient and Principled Data Selection in Large Language Model Pre-training in Every Iteration","authors":["Shaobo Wang","Xuan Ouyang","Tianyi Xu","Yuzheng Hu","Jialin Liu","Guo Chen","Tianyu Zhang","Junhao Zheng","Kexin Yang","Xingzhang Ren","Dayiheng Liu","Linfeng Zhang"],"insts":["Shanghai Jiao Tong University","Department of Computer Science, University of Wisconsin - Madison","Oregon State University"],"area":"Deep Learning","sub":"Foundation Models","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71030","alphaxiv":"2602.05400"},{"i":3758,"pid":"71142","orid":"71142","title":"WeDLM: Reconciling Diffusion Language Models with Standard Causal Attention for Fast Inference","authors":["Aiwei Liu","Minghua He","Shaoxun Zeng","Sijun Zhang","Linhao Zhang","Chuhan Wu","Wei Jia","Yuan Liu","Zhou Xiao","Jie Zhou"],"insts":["Wechat AI","Peking University","Tsinghua University"],"area":"Deep Learning","sub":"Large Language Models","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71142","alphaxiv":"2512.22737"},{"i":3817,"pid":"71034","orid":"71034","title":"VenusBench-Mobile: A Challenging and User-Centric Benchmark for Mobile GUI Agents with Capability Diagnostics","authors":["Yichen Gong","Zhuohan Cai","Sunhao Dai","Yuqi Zhou","Zhangxuan Gu","Changhua Meng","Shuheng Shen"],"insts":["Ant Group","Tsinghua University","Renmin University of China"],"area":"Deep Learning","sub":"Foundation Models","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71034","alphaxiv":"2604.06182"},{"i":3889,"pid":"71050","orid":"71050","title":"PhotoAgent: Exploratory Visual Aesthetic Planning with Large Vision Models","authors":["Mingde Yao","Zhiyuan You","King-Man Tam","Menglu Wang","Tianfan Xue"],"insts":["The Chinese University of Hong Kong","Institute of Science Tokyo","University of Science and Technology of China"],"area":"Reinforcement Learning","sub":"Multi-agent","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71050","alphaxiv":"2602.22809"},{"i":4212,"pid":"71069","orid":"71069","title":"Rex: A Family of Reversible Exponential (Stochastic) Runge-Kutta Solvers","authors":["Zander Blasingame","Chen Liu"],"insts":["AITHYRA","Clarkson University"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71069","alphaxiv":"2502.08834"},{"i":4241,"pid":"71038","orid":"71038","title":"Protein Fold Classification at Scale: Benchmarking and Pretraining","authors":["Dexiong Chen","Andrei Manolache","Mathias Niepert","Karsten Borgwardt"],"insts":["MPI of Biochemistry","Universität Stuttgart / Bitdefender","Universität Stuttgart"],"area":"Applications","sub":"Health / Medicine","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71038","alphaxiv":"2605.18552"},{"i":4242,"pid":"71037","orid":"71037","title":"Protein Autoregressive Modeling via Multiscale Structure Generation","authors":["Yanru Qu","Cheng-Yen Hsieh","Zaixiang Zheng","Ge Liu","Quanquan Gu"],"insts":["University of Illinois, Urbana Champaign","ByteDance Seed","ByteDance-Seed"],"area":"Applications","sub":"Chemistry, Physics, and Earth Sciences","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71037","alphaxiv":"2602.04883"},{"i":4587,"pid":"71154","orid":"71154","title":"Rare Event Analysis of Large Language Models","authors":["Jake McAllister Dorman","Edward Gillman","Dominic C Rose","Jamie Mair","Juan Garrahan"],"insts":["University of Nottingham"],"area":"Deep Learning","sub":"Large Language Models","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71154","alphaxiv":"2602.06791"},{"i":5412,"pid":"71077","orid":"71077","title":"RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies","authors":["Yinpei Dai","Hongze Fu","Jayjun Lee","Yuejiang Liu","Haoran Zhang","Jianing Yang","Chelsea Finn","Nima Fazeli","Joyce Chai"],"insts":["University of Michigan - Ann Arbor","EPFL","Meta"],"area":"Applications","sub":"Robotics","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71077","alphaxiv":"2603.04639"},{"i":229,"pid":"71029","orid":"71029","title":"Less is Enough: Synthesizing Diverse Data in Feature Space of LLMs","authors":["Zhongzhi Li","Xuansheng Wu","Yijiang Li","Lijie Hu","Ninghao Liu"],"insts":["University of Georgia","University of California, San Diego","MBZUAI"],"area":"General Machine Learning","sub":"Data","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71029","alphaxiv":"2602.10388"},{"i":663,"pid":"71153","orid":"71153","title":"Characterizing, Evaluating, and Optimizing Complex Reasoning","authors":["Haoran Zhang","Yafu Li","Zhi Wang","Zhilin Wang","Shunkai Zhang","Xiaoye Qu","Yu Cheng"],"insts":["Shanghai Jiao Tong University","Shanghai Artificial Intelligence Laboratory","Nanjing University"],"area":"Deep Learning","sub":"Large Language Models","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71153","alphaxiv":"2602.08498"},{"i":251,"pid":"63725","orid":"UGAP2F6FfV","title":"Holi-Spatial: Evolving Video Streams into Holistic 3D Spatial Intelligence","authors":["Yuanyuan Gao","Hao Li","Yifei Liu","Xinhao Ji","Yuning Gong","Yuanjun Liao","Fangfu Liu","Manyuan Zhang","Yuchen Yang","Dan Xu","Xue Yang","Huaxi Huang","Hongjie Zhang","Ziwei Liu","Xiao Sun","Dingwen Zhang","Zhihang Zhong"],"insts":["Hong Kong University of Science and Technology","Nanyang Technological University","Shanghai AI Laboratory"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=UGAP2F6FfV","vs":"https://icml.cc/virtual/2026/poster/63725","alphaxiv":"2603.07660"},{"i":567,"pid":"61957","orid":"lJpXXwhRRF","title":"VisionWebDev: A Hierarchical Benchmark for Visual Website Development with Agent Verification","authors":["Zehai He","Wenyi Hong","ZHEN YANG","Ziyang Pan","Mingdao Liu","Xiaotao Gu","Jie Tang"],"insts":["Tsinghua University","Tsinghua University, Tsinghua University","Zhipu AI"],"area":"General Machine Learning","sub":"Evaluation","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=lJpXXwhRRF","vs":"https://icml.cc/virtual/2026/poster/61957","alphaxiv":"2603.26648"},{"i":603,"pid":"63465","orid":"WwS8CTpUA6","title":"SceneSmith: Agentic Generation of Simulation-Ready Indoor Scenes","authors":["Nicholas Pfaff","Thomas Cohn","Sergey Zakharov","Rick Cory","Russ Tedrake"],"insts":["MIT","Massachusetts Institute of Technology","Toyota Research Institute"],"area":"Applications","sub":"Robotics","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=WwS8CTpUA6","vs":"https://icml.cc/virtual/2026/poster/63465","alphaxiv":"2602.09153"},{"i":651,"pid":"66055","orid":"7UEBX1KU1y","title":"Conditional Equivalence of DPO and RLHF: Assumptions, Failure Modes, and Provable Alignment","authors":["Yonggang Zhang","Zhiqin Yang","Wei Xue","Dong Fang","Bo Han","Yike Guo"],"insts":["Hong Kong Baptist University","HKUST","Hong Kong University of Science and Technology"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=7UEBX1KU1y","vs":"https://icml.cc/virtual/2026/poster/66055","alphaxiv":"2605.20834"},{"i":656,"pid":"65830","orid":"9kJQjx2B80","title":"ProcMEM: Learning Reusable Procedural Memory from Experience via Non-Parametric PPO for LLM Agents","authors":["QIRUI MI","Zhijian Ma","Mengyue Yang","Yisen Wang","Haoxuan Li","Haifeng Zhang","Jun Wang"],"insts":["Institute of Automation, Chinese Academy of Sciences, China","Institute of automation, Chinese academy of science, Chinese Academy of Sciences","University College London"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=9kJQjx2B80","vs":"https://icml.cc/virtual/2026/poster/65830","alphaxiv":"2602.01869"},{"i":708,"pid":"62679","orid":"ePFvXPdvhM","title":"PACT: Self-Evolving Physical Safety Alignment for Diffusion Policies in Embodied Manipulation","authors":["Lingxuan Wu","Zijian Zhu","Lizhong Wang","Chengyang Ying","Huayu Chen","Xiao Yang","Fangming Liu","Jun Zhu"],"insts":["Tsinghua University","Tsinghua University, Tsinghua University","Huazhong University of Science and Technology"],"area":"Applications","sub":"Robotics","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=ePFvXPdvhM","vs":"https://icml.cc/virtual/2026/poster/62679","alphaxiv":"2606.08414"},{"i":940,"pid":"66808","orid":"08tW615mgI","title":"Protein Autoregressive Modeling via Multiscale Structure Generation","authors":["Yanru Qu","Cheng-Yen Hsieh","Zaixiang Zheng","Ge Liu","Quanquan Gu"],"insts":["University of Illinois, Urbana Champaign","ByteDance Seed","ByteDance-Seed"],"area":"Applications","sub":"Chemistry, Physics, and Earth Sciences","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=08tW615mgI","vs":"https://icml.cc/virtual/2026/poster/66808","alphaxiv":"2602.04883"},{"i":1010,"pid":"60542","orid":"zAl9heLw4q","title":"Motion Attribution for Video Generation","authors":["Xindi Wu","Despoina Paschalidou","Jun Gao","Antonio Torralba","Laura Leal-Taixé","Olga Russakovsky","Sanja Fidler","Jonathan Lorraine"],"insts":["Princeton University","NVIDIA","MIT"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=zAl9heLw4q","vs":"https://icml.cc/virtual/2026/poster/60542","alphaxiv":"2601.08828"},{"i":1062,"pid":"62667","orid":"eXxFlOPTk4","title":"VideoKR: Towards Knowledge- and Reasoning-Intensive Video Understanding","authors":["Lin Fu","Zheyuan Yang","Yang Wang","Tingyu Song","Arman Cohan","Yilun Zhao"],"insts":["Zhejiang University","Tongji University","University of the Chinese Academy of Sciences"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=eXxFlOPTk4","vs":"https://icml.cc/virtual/2026/poster/62667","alphaxiv":"2606.05259"},{"i":1205,"pid":"61165","orid":"t73XUJvyQr","title":"Latent Laplace Diffusion for Irregular Multivariate Time Series","authors":["Zinuo You","Jin Zheng","John Cartlidge"],"insts":["University of Bristol"],"area":"Deep Learning","sub":"Sequential Models, Time series","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=t73XUJvyQr","vs":"https://icml.cc/virtual/2026/poster/61165","alphaxiv":"2605.19805"},{"i":1300,"pid":"66494","orid":"3BW15kSPfN","title":"h1: Bootstrapping LLMs to Reason over Longer Horizons via Reinforcement Learning","authors":["Alesia Ivanova","Sumeet Motwani","Jack Cai","Phil Torr","Riashat Islam","Shital Shah","Christian Schroeder de Witt","Charles London"],"insts":["University of Oxford","University of Wisconsin-Madison","Oxford"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=3BW15kSPfN","vs":"https://icml.cc/virtual/2026/poster/66494","alphaxiv":"2510.07312"},{"i":1336,"pid":"61499","orid":"pZNo1YWT5x","title":"Balancing Understanding and Generation in Discrete Diffusion Models","authors":["Yue Liu","Yuzhong Zhao","Zheyong Xie","Qixiang Ye","Jianbin Jiao","Yao Hu","Shaosheng Cao","Liu"],"insts":["University of the Chinese Academy of Sciences","Xiaohongshu","University of Chinese Academy of Sciences"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=pZNo1YWT5x","vs":"https://icml.cc/virtual/2026/poster/61499","alphaxiv":"2602.01362"},{"i":1456,"pid":"61834","orid":"mWxEAgz3xu","title":"Characterizing Agents in Production","authors":["Melissa Pan","Negar Arabzadeh","Riccardo Cogo","Yuxuan Zhu","Alexander Xiong","Lakshya A Agrawal","Huanzhi Mao","Emma Shen","Sid Pallerla","Liana Patel","Shu Liu","Tianneng Shi","Xiaoyuan Liu","Jared Davis","Emmanuele Lacavalla","Alessandro Basile","Shuyi Yang","Paul Castro","Daniel Kang","Koushik Sen","Dawn Song","Joseph E Gonzalez","Ion Stoica","Matei Zaharia","Marquita Ellis"],"insts":["University of California, Berkeley","Intesa Sanpaolo S.p.A.","University of Illinois Urbana Champaign"],"area":"General Machine Learning","sub":"Data","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=mWxEAgz3xu","vs":"https://icml.cc/virtual/2026/poster/61834","alphaxiv":"2512.04123"},{"i":1506,"pid":"61954","orid":"lK2o9OjoXf","title":"PLANTAIN: Plan-Answer Interleaved Reasoning","authors":["Anthony Liang","Jonathan Berant","Adam Fisch","Abhimanyu Goyal","Kalpesh Krishna","Jacob Eisenstein"],"insts":["University of Southern California","Google","Google DeepMind"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=lK2o9OjoXf","vs":"https://icml.cc/virtual/2026/poster/61954","alphaxiv":"2512.03176"},{"i":1647,"pid":"63474","orid":"Ws8swqL5ob","title":"PhotoAgent: Exploratory Visual Aesthetic Planning with Large Vision Models","authors":["Mingde Yao","Zhiyuan You","King-Man Tam","Menglu Wang","Tianfan Xue"],"insts":["The Chinese University of Hong Kong","Institute of Science Tokyo","University of Science and Technology of China"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=Ws8swqL5ob","vs":"https://icml.cc/virtual/2026/poster/63474","alphaxiv":"2602.22809"},{"i":1650,"pid":"62376","orid":"hCAEcqig2C","title":"Learning Randomized Reductions","authors":["Ferhat Erata","Orr Paradise","Thanos Typaldos","Timos Antonopoulos","ThanhVu Nguyen","Shafi Goldwasser","Ruzica Piskac"],"insts":["Yale University","EPFL","George Mason University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=hCAEcqig2C","vs":"https://icml.cc/virtual/2026/poster/62376","alphaxiv":"2412.18134"},{"i":1703,"pid":"62549","orid":"fdkSA4F0lN","title":"Weak Diffusion Priors Can Still Achieve Strong Inverse-Problem Performance","authors":["Jing Jia","Wei Yuan","Sifan Liu","Liyue Shen","Guanyang Wang"],"insts":[", Rutgers University","Rutgers University","Duke University"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=fdkSA4F0lN","vs":"https://icml.cc/virtual/2026/poster/62549","alphaxiv":"2601.22443"},{"i":1847,"pid":"66025","orid":"7pQIzVNctu","title":"Rex: A Family of Reversible Exponential (Stochastic) Runge-Kutta Solvers","authors":["Zander Blasingame","Chen Liu"],"insts":["AITHYRA","Clarkson University"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=7pQIzVNctu","vs":"https://icml.cc/virtual/2026/poster/66025","alphaxiv":"2502.08834"},{"i":1923,"pid":"65332","orid":"EeuLO2BjFN","title":"Maximum Likelihood Reinforcement Learning","authors":["Fahim Tajwar","Guanning Zeng","Yueer Zhou","Yuda Song","Daman Arora","Yiding Jiang","Jeff Schneider","Russ Salakhutdinov","Haiwen Feng","Andrea Zanette"],"insts":["Carnegie Mellon University","Zhejiang University","IIT Delhi"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=EeuLO2BjFN","vs":"https://icml.cc/virtual/2026/poster/65332","alphaxiv":"2602.02710"},{"i":2013,"pid":"62383","orid":"h7WBYYJF1Q","title":"Efficient Parallel Samplers for Recurrent-Depth Models","authors":["Jonas Geiping","Xinyu Yang","Guinan Su"],"insts":["ELLIS Institute &  Max-Planck Institute for Intelligent Systems, Tübingen AI Center","Carnegie Mellon University","Max Planck Institute for Intelligent Systems"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=h7WBYYJF1Q","vs":"https://icml.cc/virtual/2026/poster/62383","alphaxiv":"2510.14961"},{"i":2303,"pid":"62040","orid":"kR4iOTaAOJ","title":"ThunderAgent: A Fast, Simple, and Program-Aware Agentic Inference System","authors":["Hao Kang","Ziyang Li","Xinyu Yang","Weili Xu","Yinfang Chen","Junxiong Wang","Beidi Chen","Tushar Krishna","Chenfeng Xu","Simran Arora"],"insts":["Georgia Institute of Technology","Zhejiang University","Carnegie Mellon University"],"area":"General Machine Learning","sub":"Hardware and Software","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=kR4iOTaAOJ","vs":"https://icml.cc/virtual/2026/poster/62040","alphaxiv":"2602.13692"},{"i":2308,"pid":"64225","orid":"PnTXyTR2VG","title":"Multimodal Latent Language Modeling with Next-Token Diffusion","authors":["Yutao Sun","Hangbo Bao","Wenhui Wang","Zhiliang Peng","Li Dong","Shaohan Huang","Yaoyao Chang","Jianyong Wang","Furu Wei"],"insts":["Tsinghua University","Harbin Institute of Technology","Microsoft"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=PnTXyTR2VG","vs":"https://icml.cc/virtual/2026/poster/64225","alphaxiv":"2412.08635"},{"i":2311,"pid":"64240","orid":"PeFSCRulgy","title":"Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments","authors":["Siwei Wu","Yizhi Li","Yuyang Song","Wei Zhang","Yang Wang","Riza Batista-Navarro","Xian Yang","Mingjie Tang","Bryan Dai","Jian Yang","Chenghua Lin"],"insts":["University of Manchester","IQuest Research / University of Manchester","Sichuan University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=PeFSCRulgy","vs":"https://icml.cc/virtual/2026/poster/64240","alphaxiv":"2602.01244"},{"i":2315,"pid":"60900","orid":"vTp9JToZl9","title":"SplAttN: Bridging 2D and 3D with Gaussian Soft Splatting and Attention for Point Cloud Completion","authors":["Zhaoyang Li","Zhichao You","Tianrui Li"],"insts":["School of Computing and Artificial Intelligence, Southwest Jiaotong University","Southwest Jiaotong University"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=vTp9JToZl9","vs":"https://icml.cc/virtual/2026/poster/60900","alphaxiv":"2605.01466"},{"i":2341,"pid":"66443","orid":"3gCdh3u2GK","title":"Mind-Omni: A Unified Multi-Task Framework for Brain-Vision-Language Modeling via Discrete Diffusion","authors":["Yizhuo Lu","Changde Du","Qingyu Shi","Hang Chen","Jie Peng","Liuyun Jiang","Shuangchen Zhao","Huiguang He"],"insts":["CASIA","Institute of automation, Chinese academy of science, Chinese Academy of Sciences","Peking University"],"area":"Applications","sub":"Neuroscience, Cognitive Science","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=3gCdh3u2GK","vs":"https://icml.cc/virtual/2026/poster/66443","alphaxiv":"2605.29591"},{"i":2443,"pid":"61887","orid":"lwOoBzJykL","title":"LaST$_{0}$: Latent Spatio-Temporal Chain-of-Thought for Robotic Vision-Language-Action Model","authors":["Zhuoyang Liu","Jiaming Liu","Hao Chen","Jiale Yu","Ziyu Guo","Chengkai Hou","Xiangju Mi","Chenyang Gu","Renrui Zhang","Kun Wu","Zhengping Che","Jian Tang","Pheng Ann Heng","Shanghang Zhang"],"insts":["Peking University","Department of Computer Science and Engineering, The Chinese University of Hong Kong","CUHK"],"area":"Applications","sub":"Robotics","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=lwOoBzJykL","vs":"https://icml.cc/virtual/2026/poster/61887","alphaxiv":"2601.05248"},{"i":4154,"pid":"71122","orid":"71122","title":"Reward-free Alignment for Conflicting Objectives","authors":["Peter Chen","Xiaopeng Li","Xi Chen","Tianyi Lin"],"insts":["DAMO Academy, Alibaba Seattle // UC Berkeley","The Chinese University of Hong Kong, Shenzhen","NYU"],"area":"Deep Learning","sub":"Large Language Models","type":"Oral","spot":true,"or":"","vs":"https://icml.cc/virtual/2026/oral/71122","alphaxiv":"2602.02495"},{"i":6271,"pid":"65277","orid":"F9NDKf5oPy","title":"Learning Unmasking Policies for Diffusion Language Models","authors":["Metod Jazbec","Theo X. Olausson","Louis Béthune","Pierre Ablin","Michael Kirchhof","Joao Monteiro","Victor Guilherme Turrisi da Costa","Jason Ramapuram","Marco Cuturi"],"insts":["University of Amsterdam","MIT","Apple"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=F9NDKf5oPy","vs":"https://icml.cc/virtual/2026/poster/65277","alphaxiv":"2512.09106"},{"i":6451,"pid":"63432","orid":"XFTTezxLdU","title":"Procedural Pretraining: Warming Up Language Models with Abstract Data","authors":["Liangze Jiang","Zachary Shinnick","Anton Hengel","Hemanth Saratchandran","Damien Teney"],"insts":["EPFL","University of Adelaide","University of Adelaide/Australian Institute of Machine Learning"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=XFTTezxLdU","vs":"https://icml.cc/virtual/2026/poster/63432","alphaxiv":"2601.21725"},{"i":6794,"pid":"62561","orid":"fXr3uPr1G5","title":"Scaling Law for Quantization-Aware Training","authors":["Mengzhao Chen","Chaoyi Zhang","Jing Liu","Zeng","Zeyue Xue","Zhiheng Liu","Yunshui Li","Jin Ma","Jie Huang","zhou Xun","Ping Luo"],"insts":["University of Hong Kong","ByteDance Inc.","The University of Hong Kong"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=fXr3uPr1G5","vs":"https://icml.cc/virtual/2026/poster/62561","alphaxiv":"2505.14302"},{"i":2680,"pid":"65330","orid":"Efq2VvYk1o","title":"ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language Models","authors":["Long (Tony) Lian","Sida Wang","Felix Juefei-Xu","Tsu-Jui Fu","Xiuyu Li","Adam Yala","Trevor Darrell","Alane Suhr","Yuandong Tian","Xi Victoria Lin"],"insts":["UC Berkeley","FAIR at Meta","Google DeepMind"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=Efq2VvYk1o","vs":"https://icml.cc/virtual/2026/poster/65330","alphaxiv":"2512.07843"},{"i":718,"pid":"66806","orid":"09CSjVeDug","title":"Enhancing Reasoning for Diffusion LLMs via Distribution Matching Policy Optimization","authors":["Yuchen Zhu","Wei Guo","Jaemoo Choi","Petr Molodyk","Bo Yuan","Molei Tao","Yongxin Chen"],"insts":["Georgia Institute of Technology","Georgia Tech"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=09CSjVeDug","vs":"https://icml.cc/virtual/2026/poster/66806","alphaxiv":"2510.08233"},{"i":4134,"pid":"65361","orid":"ENuMNYCiV6","title":"Know More, Know Clearer: A Meta-Cognitive Framework for Knowledge Augmentation in Large Language Models","authors":["Hao Chen","Ye He","Yuchun Fan","Yukun Yan","Zhenghao Liu","Qingfu Zhu","Maosong Sun","Wanxiang Che"],"insts":["Harbin Institute of Technology","Northeastern University","Tsinghua University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=ENuMNYCiV6","vs":"https://icml.cc/virtual/2026/poster/65361","alphaxiv":"2602.12996"},{"i":3268,"pid":"64259","orid":"PQaxfoEcRc","title":"Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units","authors":["Jianhui Chen","Yuzhang Luo","Liangming Pan"],"insts":["Peking University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=PQaxfoEcRc","vs":"https://icml.cc/virtual/2026/poster/64259","alphaxiv":"2601.21996"},{"i":2455,"pid":"64619","orid":"LcswwEzzX7","title":"OXE-AugE: A Large-Scale Robot Augmentation of OXE for Scaling Cross-Embodiment Policy Learning","authors":["Guanhua Ji","Harsha Polavaram","Lawrence Yunliang Chen","Sandeep Bajamahal","Zehan Ma","Simeon Adebola","Chenfeng Xu","Ken Goldberg"],"insts":["University of Pennsylvania, University of Pennsylvania","University of California, Berkeley","Thinking Machines Lab"],"area":"Applications","sub":"Robotics","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=LcswwEzzX7","vs":"https://icml.cc/virtual/2026/poster/64619","alphaxiv":"2512.13100"},{"i":3816,"pid":"62831","orid":"coHiGZOFtS","title":"VenusBench-Mobile: A Challenging and User-Centric Benchmark for Mobile GUI Agents with Capability Diagnostics","authors":["Yichen Gong","Zhuohan Cai","Sunhao Dai","Yuqi Zhou","Zhangxuan Gu","Changhua Meng","Shuheng Shen"],"insts":["Ant Group","Tsinghua University","Renmin University of China"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=coHiGZOFtS","vs":"https://icml.cc/virtual/2026/poster/62831","alphaxiv":"2604.06182"},{"i":5907,"pid":"65015","orid":"Hm8OEDKpiO","title":"3ViewSense: Spatial and Mental Perspective Reasoning from Orthographic Views in Vision-Language Models","authors":["Shaoxiong Zhan","Yanlin Lai","Zheng Liu","Zijian Lin","Lin Hai","Xiaodong Cai","Shen Li","Wen Huang","Hai-Tao Zheng"],"insts":["Tsinghua University","Tsinghua University, Tsinghua University","Chongqing University"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=Hm8OEDKpiO","vs":"https://icml.cc/virtual/2026/poster/65015","alphaxiv":"2603.07751"},{"i":4765,"pid":"60920","orid":"vKWxArobP3","title":"Alignment-Guided Score Matching for Text-to-Image Alignment in Diffusion Models","authors":["Jaa-Yeon Lee","Yeobin Hong","Taesung Kwon","Jong Chul YE"],"insts":["KAIST","KAIST AI"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=vKWxArobP3","vs":"https://icml.cc/virtual/2026/poster/60920","alphaxiv":"2605.30038"},{"i":3115,"pid":"66554","orid":"2hQBG2ZlFb","title":"Beyond Log Likelihood: Probability-Based Objectives for Supervised Fine-Tuning across the Model Capability Continuum","authors":["Gaotang Li","Ruizhong Qiu","Xiusi Chen","Heng Ji","Hanghang Tong"],"insts":["University of Illinois Urbana-Champaign","University of Illinois, Urbana-Champaign"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=2hQBG2ZlFb","vs":"https://icml.cc/virtual/2026/poster/66554","alphaxiv":"2510.00526"},{"i":2491,"pid":"64728","orid":"KRSZj8z5Lr","title":"LiME: Lightweight Mixture of Experts for Efficient Multimodal Multi-task Learning","authors":["Md Kowsher","Haris Mansoor","Nusrat Prottasha","Ozlem Garibay","Victor Zhu","Zhengping Ji","Chen Chen"],"insts":["University of Central Florida","Coventry University","Stevens Institute of Technology"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=KRSZj8z5Lr","vs":"https://icml.cc/virtual/2026/poster/64728","alphaxiv":"2604.02338"},{"i":6196,"pid":"65791","orid":"A5venxSvpw","title":"MASPOB: Bandit-Based Prompt Optimization for Multi-Agent Systems with Graph Neural Networks","authors":["Zhi Hong","Qian Zhang","Jiahang Sun","Zhiwei Shang","Mingze Kong","Xiangyi Wang","Yao Shu","Zhongxiang Dai"],"insts":["The Chinese University of Hong Kong, Shenzhen","The University of Tokyo","The Chinese University of Hong Kong"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=A5venxSvpw","vs":"https://icml.cc/virtual/2026/poster/65791","alphaxiv":"2603.02630"},{"i":4262,"pid":"62428","orid":"giNBsVVFGt","title":"PRISM: Demystifying Retention and Interaction in Mid-Training","authors":["Bharat Runwal","Ashish Agrawal","Anurag Roy","Rameswar Panda"],"insts":["MIT-IBM Watson AI Lab, IBM Research","International Business Machines","IIT Kharagpur"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=giNBsVVFGt","vs":"https://icml.cc/virtual/2026/poster/62428","alphaxiv":"2603.17074"},{"i":3538,"pid":"60903","orid":"vSzRJyg6k0","title":"Reward-free Alignment for Conflicting Objectives","authors":["Peter Chen","Xiaopeng Li","Xi Chen","Tianyi Lin"],"insts":["DAMO Academy, Alibaba Seattle // UC Berkeley","The Chinese University of Hong Kong, Shenzhen","NYU"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=vSzRJyg6k0","vs":"https://icml.cc/virtual/2026/poster/60903","alphaxiv":"2602.02495"},{"i":5069,"pid":"65933","orid":"8m30ogkPk2","title":"RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies","authors":["Yinpei Dai","Hongze Fu","Jayjun Lee","Yuejiang Liu","Haoran Zhang","Jianing Yang","Chelsea Finn","Nima Fazeli","Joyce Chai"],"insts":["University of Michigan - Ann Arbor","EPFL","Meta"],"area":"Applications","sub":"Robotics","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=8m30ogkPk2","vs":"https://icml.cc/virtual/2026/poster/65933","alphaxiv":"2603.04639"},{"i":4274,"pid":"61483","orid":"pl10b6EQAN","title":"Skip a Layer or Loop It? Learning Program-of-Layers in LLMs","authors":["Ziyue Li","Yang Li","Tianyi Zhou"],"insts":["University of Maryland, College Park","University of North Carolina at Chapel Hill","Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=pl10b6EQAN","vs":"https://icml.cc/virtual/2026/poster/61483","alphaxiv":"2606.06574"},{"i":738,"pid":"64675","orid":"KyVlaw4BxE","title":"VideoFlexTok: Flexible-Length Coarse-to-Fine Video Tokenization","authors":["Andrei Atanov","Jesse Allardice","Roman Bachmann","Oğuzhan Fatih Kar","R Devon Hjelm","David Griffiths","Peter Fu","Amir Zamir","Afshin Dehghan"],"insts":["Swiss Federal Institute of Technology Lausanne (EPFL)","Apple","Microsoft Research / Mila"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=KyVlaw4BxE","vs":"https://icml.cc/virtual/2026/poster/64675","alphaxiv":"2604.12887"},{"i":3103,"pid":"63621","orid":"VMsumctGvg","title":"From Pixels to Tokens: A Systematic Study of Latent Action Supervision for Vision-Language-Action Models","authors":["Yihan Lin","Haoyang Li","Yang Li","Haitao Shen","Yihan Zhao","Chao Shao","Jing Zhang"],"insts":["Beihang University","Renmin University of China","Boss Zhipin"],"area":"Applications","sub":"Robotics","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=VMsumctGvg","vs":"https://icml.cc/virtual/2026/poster/63621","alphaxiv":"2605.04678"},{"i":3805,"pid":"65040","orid":"HSuU4xBmAv","title":"Geometry-Aware Decoding with Wasserstein-Regularized Truncation and Mass Penalties for Large Language Models","authors":["Arash Gholamidavoodi","Navid Rezazadeh","Seyed Davoudi","Pouya Pezeshkpour"],"insts":["Apple","University of California, Irvine","Islamic Azad University, Amol Branch"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=HSuU4xBmAv","vs":"https://icml.cc/virtual/2026/poster/65040","alphaxiv":"2602.10346"},{"i":2947,"pid":"61870","orid":"mDhyxu8WRb","title":"mHC: Manifold-Constrained Hyper-Connections","authors":["Zhenda Xie","Yixuan Wei","Huanqi Cao","Chenggang Zhao","Chengqi Deng","Jiashi Li","Damai Dai","Huazuo Gao","Mingyu Xu","Kuai Yu","Liang Zhao","Shangyan Zhou","Zhean Xu","Zhengyan Zhang","Wangding Zeng","Shengding Hu","Yuqing Wang","Jingyang Yuan","Lean Wang","Wenfeng Liang"],"insts":["DeepSeek AI","Hangzhou DeepSeek Artificial Intelligence Co., Ltd","Tsinghua University, Tsinghua University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=mDhyxu8WRb","vs":"https://icml.cc/virtual/2026/poster/61870","alphaxiv":"2512.24880"},{"i":3125,"pid":"65901","orid":"92oF5bU4cU","title":"Neural Thickets: Diverse Task Experts Are Dense Around Pretrained Weights","authors":["Yulu Gan","Phillip Isola"],"insts":["Massachusetts Institute of Technology","MIT"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=92oF5bU4cU","vs":"https://icml.cc/virtual/2026/poster/65901","alphaxiv":"2603.12228"},{"i":5921,"pid":"60719","orid":"xEgoeNrp8B","title":"PanoWorld-X: Generating Explorable Panoramic Worlds via Sphere-Aware Video Diffusion","authors":["Yuyang Yin","Hao-Xiang Guo","Fangfu Liu","Mengyu Wang","Hanwen Liang","Eric Li","Yikai Wang","Xiaojie Jin","Yao Zhao","Yunchao Wei"],"insts":["Beijing Jiaotong University","Department of Electronic Engineering, Tsinghua University","University of Toronto"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=xEgoeNrp8B","vs":"https://icml.cc/virtual/2026/poster/60719","alphaxiv":"2509.24997"},{"i":2655,"pid":"65206","orid":"FneePKFVHT","title":"PaperBanana: Automating Academic Illustration for AI Scientists","authors":["Dawei Zhu","Rui Meng","Yale Song","Xiyu Wei","Sujian Li","Tomas Pfister","Jinsung Yoon"],"insts":["Peking University","Google"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=FneePKFVHT","vs":"https://icml.cc/virtual/2026/poster/65206","alphaxiv":"2601.23265"},{"i":4419,"pid":"60609","orid":"yVFxjNzCQm","title":"PonderLM-2: Pretraining LLM with Latent Thoughts in Continuous Space","authors":["Boyi Zeng","He Li","Shixiang Song","Yixuan Wang","Zitong Wang","Ziwei He","Xinbing Wang","Zhouhan Lin"],"insts":["Shanghai Jiaotong University","Shanghai Jiao Tong University","SYSU & SJTU Lumia Lab"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=yVFxjNzCQm","vs":"https://icml.cc/virtual/2026/poster/60609","alphaxiv":"2509.23184"},{"i":6601,"pid":"62145","orid":"jPKqiaPTEd","title":"Protein Fold Classification at Scale: Benchmarking and Pretraining","authors":["Dexiong Chen","Andrei Manolache","Mathias Niepert","Karsten Borgwardt"],"insts":["MPI of Biochemistry","Universität Stuttgart / Bitdefender","Universität Stuttgart"],"area":"Applications","sub":"Chemistry, Physics, and Earth Sciences","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=jPKqiaPTEd","vs":"https://icml.cc/virtual/2026/poster/62145","alphaxiv":"2605.18552"},{"i":3133,"pid":"66583","orid":"2RJN5vDHG0","title":"Rare Event Analysis of Large Language Models","authors":["Jake McAllister Dorman","Edward Gillman","Dominic C Rose","Jamie Mair","Juan Garrahan"],"insts":["University of Nottingham"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=2RJN5vDHG0","vs":"https://icml.cc/virtual/2026/poster/66583","alphaxiv":"2602.06791"},{"i":1174,"pid":"61077","orid":"trn64znfNx","title":"Rotary Position Encodings for Graphs","authors":["Isaac Reid","Arijit Sehanobish","Cederik Höfs","Bruno Mlodozeniec","Leonhard Vulpius","Federico Barbero","Adrian Weller","Krzysztof Choromanski","Richard E Turner","Petar Veličković"],"insts":["University of Cambridge","Kensho Technologies/S&P Global","Attention Engineering"],"area":"Deep Learning","sub":"Graph Neural Networks","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=trn64znfNx","vs":"https://icml.cc/virtual/2026/poster/61077","alphaxiv":"2509.22259"},{"i":6139,"pid":"66243","orid":"5VgZUEpK6W","title":"Skill Neologisms: Towards Skill-based Continual Learning","authors":["Antonin Berthon","Nicolás Astorga","Mihaela van der Schaar"],"insts":["University of Cambridge","University of Cambridge and UCLA"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=5VgZUEpK6W","vs":"https://icml.cc/virtual/2026/poster/66243","alphaxiv":"2605.04970"},{"i":3693,"pid":"60925","orid":"vIZz7LvObC","title":"TokSuite: Measuring the Impact of Tokenizer Choice on Language Model Behavior","authors":["Gül Sena Altıntaş","Malikeh Ehghaghi","Brian Lester","Fengyuan Liu","Wanru Zhao","Marco Ciccone","Colin Raffel"],"insts":["University of Toronto","Vector Institute","University of North Carolina at Chapel Hill"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=vIZz7LvObC","vs":"https://icml.cc/virtual/2026/poster/60925","alphaxiv":"2512.20757"},{"i":6095,"pid":"66576","orid":"2UH01A9Za0","title":"Towards Efficient LLMs Annealing with Principled Sample Selection","authors":["Yuanjian Xu","Jianing Hao","Wanbo Zhang","Zhong Li","Guang Zhang"],"insts":["The Hong Kong Univerisity of Science and Technology (Guangzhou)","Fudan University","Microsoft Research Asia"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=2UH01A9Za0","vs":"https://icml.cc/virtual/2026/poster/66576","alphaxiv":"2605.31175"},{"i":5663,"pid":"64377","orid":"OC2z7iSQKa","title":"$\\tau^2$-Bench: Evaluating Conversational Agents in a Dual-Control Environment","authors":["Victor Barres","Honghua Dong","Soham Ray","Xujie Si","Karthik Narasimhan"],"insts":["Mercor","Huawei","Sierra.ai"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=OC2z7iSQKa","vs":"https://icml.cc/virtual/2026/poster/64377","alphaxiv":"2506.07982"},{"i":4760,"pid":"65626","orid":"BhahZSDowo","title":"Benchmarking at the Edge of Comprehension","authors":["Samuele Marro","Jialin Yu","Emanuele La Malfa","Oishi Deb","Jiawei Li","Yibo Yang","Ebey Abraham","Sunando Sengupta","Eric Sommerlade","Michael Wooldridge","Phil Torr"],"insts":["University of Oxford / Microsoft","University of Oxford","Department of Computer Science, University of Oxford"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=BhahZSDowo","vs":"https://icml.cc/virtual/2026/poster/65626","alphaxiv":"2602.14307"},{"i":2725,"pid":"62319","orid":"hgMZraPlSv","title":"Beyond Theorem Proving: Formulation, Framework and Benchmark for Formal Problem-Solving","authors":["Qi Liu","Xinhao Zheng","Renqiu Xia","Xingzhi Qi","Qinxiang Cao","Junchi Yan"],"insts":["Shanghai Jiao Tong University","Shanghai Jiaotong University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=hgMZraPlSv","vs":"https://icml.cc/virtual/2026/poster/62319","alphaxiv":"2505.04528"},{"i":2737,"pid":"65469","orid":"DK0kvnNelx","title":"FlexRank: Nested Low-Rank Knowledge Decomposition for Adaptive Model Deployment","authors":["Riccardo Zaccone","Stefanos Laskaridis","Marco Ciccone","Samuel Horváth"],"insts":["Politecnico di Torino","Amazon Science","Vector Institute"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=DK0kvnNelx","vs":"https://icml.cc/virtual/2026/poster/65469","alphaxiv":"2602.02680"},{"i":3416,"pid":"61962","orid":"lFkGJ60bGq","title":"How can embedding models bind concepts?","authors":["Arnas Uselis","Darina Koishigarina","Seong Joon Oh"],"insts":["University of Tübingen","KAIST & Cortiq"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=lFkGJ60bGq","vs":"https://icml.cc/virtual/2026/poster/61962","alphaxiv":"2605.31503"},{"i":6707,"pid":"61180","orid":"syG9I9ofd8","title":"Latent Collaboration in Multi-Agent Systems","authors":["Jiaru Zou","Xiyuan Yang","Ruizhong Qiu","Gaotang Li","Katherine Tieu","Pan Lu","Ke Shen","Hanghang Tong","Yejin Choi","Jingrui He","James Zou","Mengdi Wang","Ling Yang"],"insts":["Department of Computer Science, University of Illinois at Urbana-Champaign","University of Illinois Urbana-Champaign","Stanford University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=syG9I9ofd8","vs":"https://icml.cc/virtual/2026/poster/61180","alphaxiv":"2511.20639"},{"i":5693,"pid":"60628","orid":"yKcBAJMPXZ","title":"MetaphorVU: Towards Metaphorical Video Understanding","authors":["Zhuoqun Li","Boxi Cao","Guiping Jiang","Fangrui Lv","Ruotong Pan","Jianan Wang","Xiangyu Wu","Hongyu Lin","Yaojie Lu","Yong Du","Ruyin Jia","Liyan","Tingting Gao","Han Li","Xianpei Han","Le Sun"],"insts":["University of the Chinese Academy of Sciences","Institute of Software, Chinese Academy of Sciences","University of Science and Technology of China"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=yKcBAJMPXZ","vs":"https://icml.cc/virtual/2026/poster/60628","alphaxiv":"2605.25461"},{"i":5196,"pid":"66254","orid":"5PfEQzE9bf","title":"Midtraining Bridges Pretraining and Posttraining Distributions","authors":["Emmy Liu","Graham Neubig","Chenyan Xiong"],"insts":["Carnegie Mellon University","School of Computer Science, Carnegie Mellon University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=5PfEQzE9bf","vs":"https://icml.cc/virtual/2026/poster/66254","alphaxiv":"2510.14865"},{"i":5868,"pid":"61880","orid":"m1IRWFAMsa","title":"NorMuon: Making Muon more efficient and scalable","authors":["Zichong Li","Liming Liu","Chen Liang","Weizhu Chen","Tuo Zhao"],"insts":["Georgia Institute of Technology","Microsoft","Microsoft GenAI"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=m1IRWFAMsa","vs":"https://icml.cc/virtual/2026/poster/61880","alphaxiv":"2510.05491"},{"i":6072,"pid":"66813","orid":"06Nk3dJDMq","title":"Steer Like the LLM: Activation Steering that Mimics Prompting","authors":["Geert Heyman","Frederik Vandeputte"],"insts":["Nokia Bell Labs"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=06Nk3dJDMq","vs":"https://icml.cc/virtual/2026/poster/66813","alphaxiv":"2605.03907"},{"i":1761,"pid":"62732","orid":"ds3ZOevkwx","title":"Strategic Navigation or Stochastic Search? How Agents and Humans Reason Over Document Collections","authors":["Lukasz Borchmann","Jordy Van Landeghem","Michał Turski","Shreyansh Padarha","Ryan Kearns","Adam Mahdi","Niels Rogge","Clémentine Fourrier","Siwei Han","Huaxiu Yao","Artemis Llabrés","Yiming Xu","Dimosthenis Karatzas","Hao Zhang","Anupam Datta"],"insts":["Snowflake","Instabase","University of Oxford"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=ds3ZOevkwx","vs":"https://icml.cc/virtual/2026/poster/62732","alphaxiv":"2603.12180"},{"i":5760,"pid":"64049","orid":"RNuC8Nj6rD","title":"TD3B: Transition-Directed Discrete Diffusion for Allosteric Binder Generation","authors":["Hanqun CAO","Aastha Pal","Sophia Tang","Yinuo Zhang","Jingjie Zhang","Pheng Ann Heng","Pranam Chatterjee, PhD"],"insts":["The Chinese University of Hong Kong","University of Pennsylvania","national university of singaore, National University of Singapore"],"area":"Applications","sub":"Health / Medicine","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=RNuC8Nj6rD","vs":"https://icml.cc/virtual/2026/poster/64049","alphaxiv":"2605.09810"},{"i":3506,"pid":"65098","orid":"GnqHK8Ww98","title":"Teaching Models to Teach Themselves: Reasoning at the Edge of Learnability","authors":["Shobhita Sundaram","John Quan","Ariel Kwiatkowski","Kartik Ahuja","Yann Ollivier","Julia Kempe"],"insts":["MIT","DeepMind","Facebook"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=GnqHK8Ww98","vs":"https://icml.cc/virtual/2026/poster/65098","alphaxiv":"2601.18778"},{"i":5386,"pid":"62018","orid":"kcyOjXoUZu","title":"Walrus: A Cross-domain Foundation Model for Continuum Dynamics","authors":["Michael McCabe","Payel Mukhopadhyay","Tanya Marwah","Bruno Régaldo-Saint Blancard","François Rozet","Cristiana Diaconu","Lucas Meyer","Kaze Wong","Hadi Sotoudeh","Alberto Bietti","Irina Espejo","Rio Fear","Siavash Golkar","Tom Hehir","Keiya Hirashima","Geraud Krawezik","Francois Lanusse","Rudy Morel","Ruben Ohana","Liam Parker","Mariel Pettee","Jeff Shen","Kyunghyun Cho","Miles Cranmer","Shirley Ho"],"insts":["Google Research","University of Cambridge","Flatiron Institute"],"area":"Applications","sub":"Chemistry, Physics, and Earth Sciences","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=kcyOjXoUZu","vs":"https://icml.cc/virtual/2026/poster/62018","alphaxiv":"2511.15684"},{"i":5231,"pid":"64958","orid":"IMFgiWw4jd","title":"Characterizing, Evaluating, and Optimizing Complex Reasoning","authors":["Haoran Zhang","Yafu Li","Zhi Wang","Zhilin Wang","Shunkai Zhang","Xiaoye Qu","Yu Cheng"],"insts":["Shanghai Jiao Tong University","Shanghai Artificial Intelligence Laboratory","Nanjing University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=IMFgiWw4jd","vs":"https://icml.cc/virtual/2026/poster/64958","alphaxiv":"2602.08498"},{"i":6222,"pid":"65622","orid":"BkMSFFtm5M","title":"CVE-Factory: Scaling Expert-Level Agentic Tasks for Code Security Vulnerability","authors":["Xianzhen Luo","Jingyuan Zhang","Shiqi Zhou","JinYang Huang","Chuan Xiao","Qingfu Zhu","Zhiyuan Ma","YUE XING","Yang Yue","WencongZeng","Wanxiang Che"],"insts":["Harbin Institute of Technology","Kuaishou- 快手科技","Central South University"],"area":"Applications","sub":"Everything Else","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=BkMSFFtm5M","vs":"https://icml.cc/virtual/2026/poster/65622","alphaxiv":"2602.03012"},{"i":6494,"pid":"63078","orid":"aIH1jyU37z","title":"Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural Networks","authors":["Yoshihiro Maruyama"],"insts":["Kyoto University"],"area":"Deep Learning","sub":"Theory","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=aIH1jyU37z","vs":"https://icml.cc/virtual/2026/poster/63078","alphaxiv":"2012.06333"},{"i":6754,"pid":"60765","orid":"wsA8LgHU5U","title":"Learning to Theorize the World from Observation","authors":["Doojin Baek","Gyubin Lee","Junyeob Baek","Hosung Lee","Sungjin Ahn"],"insts":["Korea Advanced Institute of Science & Technology","KAIST"],"area":"Deep Learning","sub":"Other Representation Learning","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=wsA8LgHU5U","vs":"https://icml.cc/virtual/2026/poster/60765","alphaxiv":"2605.03413"},{"i":1509,"pid":"62251","orid":"iJwDylm93H","title":"Recovering Policy-Induced Errors: Benchmarking and Trajectory Synthesis for Robust GUI Agents","authors":["Tianpeng Bu","Xin Liu","Qihua Chen","Hao Jiang","Shurui Li","hongtao duan","Lu Jiang","lulu hu","Bin Yang","Minying Zhang"],"insts":["Alibaba Cloud Computing","Alibaba Group","University of Science and Technology of China"],"area":"Applications","sub":"Everything Else","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=iJwDylm93H","vs":"https://icml.cc/virtual/2026/poster/62251","alphaxiv":"2605.29447"},{"i":2859,"pid":"65886","orid":"97NEP1pyS3","title":"Reinforcement Learning with Evolving Rubrics for Deep Research","authors":["Rulin Shao","Akari Asai","Shannon Shen","Hamish Ivison","Varsha Kishore","Jingming Zhuo","Xinran Zhao","Molly Park","Samuel Finlayson","David Sontag","Tyler Murray","Sewon Min","Pradeep Dasigi","Luca Soldaini","Faeze Brahman","Scott Yih","Sherry Wu","Luke Zettlemoyer","Yoon Kim","Hannaneh Hajishirzi","Pang Wei Koh"],"insts":["University of Washington","CMU, Carnegie Mellon University","Massachusetts Institute of Technology"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=97NEP1pyS3","vs":"https://icml.cc/virtual/2026/poster/65886","alphaxiv":"2511.19399"},{"i":2955,"pid":"63692","orid":"UamxHbDR3p","title":"RelaxFlow: Text-Driven Amodal 3D Generation","authors":["Jiayin Zhu","Guoji Fu","Xiaolu Liu","Qiyuan He","Yicong Li","Angela Yao"],"insts":["National University of Singapore","Zhejiang University","NVIDIA"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=UamxHbDR3p","vs":"https://icml.cc/virtual/2026/poster/63692","alphaxiv":"2603.05425"},{"i":6204,"pid":"65737","orid":"AfqsNFzJcs","title":"Reward and Guidance through Rubrics: Promoting Exploration to Improve Multi-Domain Reasoning","authors":["Baolong Bi","Shenghua Liu","Yiwei Wang","Siqian Tong","Lingrui Mei","Yuyao Ge","Yilong Xu","Jiafeng Guo","Xueqi Cheng"],"insts":["University of Chinese Academy of Sciences","Institute of Computing Technology, Chinese Academy of Sciences","University of California, Merced"],"area":"Reinforcement Learning","sub":"Batch/Offline","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=AfqsNFzJcs","vs":"https://icml.cc/virtual/2026/poster/65737","alphaxiv":"2511.12344"},{"i":3987,"pid":"66066","orid":"7MlfE2Da2W","title":"SCALE: Self-uncertainty Conditioned Adaptive Looking and Execution for Vision-Language-Action Models","authors":["Hyeonbeom Choi","Daechul Ahn","Youhan Lee","Taewook Kang","Seongwon Cho","Jonghyun Choi"],"insts":["Seoul National University"],"area":"Applications","sub":"Robotics","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=7MlfE2Da2W","vs":"https://icml.cc/virtual/2026/poster/66066","alphaxiv":"2602.04208"},{"i":3471,"pid":"65757","orid":"ATpOQt9VVd","title":"Unifying Masked Diffusion Models with Various Generation Orders and Beyond","authors":["Chunsan Hong","Sanghyun Lee","Jong Chul YE"],"insts":["Korea Advanced Institute of Science and Technology","Korea Advanced Institute of Science & Technology","KAIST AI"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=ATpOQt9VVd","vs":"https://icml.cc/virtual/2026/poster/65757","alphaxiv":"2602.02112"},{"i":279,"pid":"65079","orid":"GyRMbsYFiG","title":"VGGT-Motion: Motion-Aware Calibration-Free Monocular SLAM for Long-Range Consistency","authors":["Zhuang Xiong","Chen Zhang","Qingshan Xu","Wenbing Tao"],"insts":["Huazhong University of Science and Technology","Nanyang Technological University"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=GyRMbsYFiG","vs":"https://icml.cc/virtual/2026/poster/65079","alphaxiv":"2602.05508"},{"i":3675,"pid":"61703","orid":"ncRRCG4BfP","title":"WestWorld: A Knowledge-Encoded Scalable Trajectory World Model for Diverse Robotic Systems","authors":["Yuchen Wang","Jiangtao Kong","Sizhe Wei","Xiaochang Li","Haohong Lin","Hongjue Zhao","Tianyi Zhou","Lu Gan","Huajie Shao"],"insts":["William & Mary","College of William and Mary","Georgia Tech"],"area":"Applications","sub":"Robotics","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=ncRRCG4BfP","vs":"https://icml.cc/virtual/2026/poster/61703","alphaxiv":"2603.14392"},{"i":187,"pid":"62568","orid":"fSnxuRhWoC","title":"End-to-End Autoregressive Image Generation with 1D Semantic Tokenizer","authors":["Wenda Chu","Bingliang Zhang","Jiaqi Han","Yizhuo Li","Linjie Yang","Yisong Yue","Qiushan Guo"],"insts":["California Institute of Technology","Tsinghua University","The University of Hong Kong"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=fSnxuRhWoC","vs":"https://icml.cc/virtual/2026/poster/62568","alphaxiv":"2605.00503"},{"i":5790,"pid":"64521","orid":"Mkal0hTCnh","title":"MEnvAgent: Scalable Polyglot Environment Construction for Verifiable Software Engineering","authors":["Chuanzhe Guo","Jingjing Wu","Sijun He","Yang Chen","Zhaoqi Kuang","Shilong Fan","Bingjin Chen","Siqi Bao","Jing Liu","Hua Wu","Qingfu Zhu","Wanxiang Che","Haifeng Wang"],"insts":["Harbin Institute of Technology","Baidu","Tencent"],"area":"Applications","sub":"Everything Else","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=Mkal0hTCnh","vs":"https://icml.cc/virtual/2026/poster/64521","alphaxiv":"2601.22859"},{"i":4684,"pid":"64302","orid":"OyPE1ganBR","title":"Stable-GFlowNet: Toward Diverse and Robust LLM Red-Teaming via Contrastive Trajectory Balance","authors":["Minchan Kwon","Sung Hyun Baek","Minseo Kim","Jaemyung Yu","Dongyoon Han","Junmo Kim"],"insts":["Korea Advanced Institute of Science & Technology","KAIST","NAVER AI Lab"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=OyPE1ganBR","vs":"https://icml.cc/virtual/2026/poster/64302","alphaxiv":"2605.00553"},{"i":4520,"pid":"63538","orid":"WJcFtJriqv","title":"UDM-GRPO: Stable and Efficient Group Relative Policy Optimization for Uniform Discrete Diffusion Models","authors":["Jiaqi Wang","Haoge Deng","Ting Pan","Yang Liu","Chengyuan Wang","Fan Zhang","Yonggang Qi","Xinlong Wang"],"insts":["Beijing University of Posts and Telecommunications","Alibaba Group","Zhejiang University"],"area":"Reinforcement Learning","sub":"Policy Search","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=WJcFtJriqv","vs":"https://icml.cc/virtual/2026/poster/63538","alphaxiv":"2604.18518"},{"i":1750,"pid":"65959","orid":"8VVbElV06v","title":"UniPercept: Towards Unified Perceptual-Level Image Understanding across Aesthetics, Quality, Structure, and Texture","authors":["Shuo Cao","Jiayang Li","Xiaohui Li","Yuandong Pu","Kaiwen Zhu","Yuanting Gao","Siqi Luo","Yi Xin","Qi Qin","Yu Zhou","Xiangyu Chen","Wenlong Zhang","Bin Fu","Yu Qiao","Yihao Liu"],"insts":["University of Science and Technology of China","Peking University","Shanghai Jiaotong University"],"area":"Applications","sub":"Computer Vision","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=8VVbElV06v","vs":"https://icml.cc/virtual/2026/poster/65959","alphaxiv":"2512.21675"},{"i":6780,"pid":"60582","orid":"ymHDVBwmta","title":"AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs","authors":["WenXiang Lin","HuangJunTao","ZhangLuHan","Lilaiyi","Xiang Bao","mengyang zhang","BingWang Wang","Shaohuai Shi"],"insts":["Harbin Institute of Technology, Shenzhen","Harbin Institute of Technology","Huawei Technologies Ltd."],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=ymHDVBwmta","vs":"https://icml.cc/virtual/2026/poster/60582","alphaxiv":"2605.00539"},{"i":4495,"pid":"66096","orid":"75AYDsndHP","title":"Autoregressive Boltzmann Generators","authors":["Danyal Rehman","Charlie Tan","Yoshua Bengio","Joey Bose","Alexander Tong"],"insts":["Mila - Quebec AI Institute","University of Oxford","Imperial College London / Mila"],"area":"Applications","sub":"Chemistry, Physics, and Earth Sciences","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=75AYDsndHP","vs":"https://icml.cc/virtual/2026/poster/66096","alphaxiv":"2606.27361"},{"i":3277,"pid":"67171","orid":"MqzZ9X6m7f","title":"Position: Sustainable Open-Source AI Requires Tracking the Cumulative Footprint of Derivatives","authors":["shaina raza","Iuliia Zarubiieva","Ahmed Radwan","Nathaniel Lesperance","Deval Pandya","Sedef Akinli Kocak","Graham Taylor"],"insts":["Vector Institute","University of Guelph","Ontario Tech University"],"area":"Applications","sub":"Energy","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=MqzZ9X6m7f","vs":"https://icml.cc/virtual/2026/poster/67171","alphaxiv":"2601.21632"},{"i":2929,"pid":"66094","orid":"76XSBLdBdg","title":"Simultaneous Speech-to-Speech Translation Without Aligned Data","authors":["Tom Labiausse","Romain Fabre","Yannick Estève","Alexandre Défossez","Neil Zeghidour"],"insts":["Kyutai","University of Avignon","Google"],"area":"Applications","sub":"Language, Speech and Dialog","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=76XSBLdBdg","vs":"https://icml.cc/virtual/2026/poster/66094","alphaxiv":"2602.11072"},{"i":2644,"pid":"65807","orid":"9wpwfSJCp9","title":"SleepLM: Natural-Language Intelligence for Human Sleep","authors":["Zongzhe Xu","Zitao Shuai","Eideen Mozaffari","Ravi Aysola","Rajesh Kumar","Yuzhe Yang"],"insts":["University of California, Los Angeles","UCLA Computer Science Department, University of California, Los Angeles","UCLA"],"area":"Applications","sub":"Health / Medicine","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=9wpwfSJCp9","vs":"https://icml.cc/virtual/2026/poster/65807","alphaxiv":"2602.23605"},{"i":3597,"pid":"62436","orid":"gc7Gg18ejz","title":"Stable Deep Reinforcement Learning via Isotropic Gaussian Representations","authors":["Ali Saheb pasand","Johan Obando-Ceron","Aaron Courville","Pouya Bashivan","Pablo Samuel Castro"],"insts":["mcgill/mila","Mila / Université de Montréal","University of Montreal"],"area":"Reinforcement Learning","sub":"Deep RL","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=gc7Gg18ejz","vs":"https://icml.cc/virtual/2026/poster/62436","alphaxiv":"2602.19373"},{"i":5804,"pid":"63337","orid":"XztRm216YS","title":"TimeRewarder: Learning Dense Reward from Passive Videos via Frame-wise Temporal Distance","authors":["Yuyang Liu","Chuan Wen","Yihang Hu","Dinesh Jayaraman","Yang Gao"],"insts":["IIIS, Tsinghua University","Shanghai Jiao Tong University","Tsinghua University"],"area":"Reinforcement Learning","sub":"Inverse","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=XztRm216YS","vs":"https://icml.cc/virtual/2026/poster/63337","alphaxiv":"2509.26627"},{"i":432,"pid":"62398","orid":"gz7hVnrRWq","title":"Advancing LLM Reasoning with Natural Language and Numerical Feedback","authors":["Xiaoying Zhang","Yipeng Zhang","Hao Sun","Kaituo Feng","Chaochao Lu","Chao Yang","Helen M Meng"],"insts":["The Chinese University of Hong Kong","University of Cambridge","Shanghai AI Laboratory "],"area":"Reinforcement Learning","sub":"Online","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=gz7hVnrRWq","vs":"https://icml.cc/virtual/2026/poster/62398","alphaxiv":"2506.03106"},{"i":5132,"pid":"60557","orid":"z1bSFIEexL","title":"Focusing Where Vision Matters: Selective Training for Large Vision Language Models via Visual Information Gain","authors":["Seulbi Lee","Sangheum Hwang"],"insts":["Seoul National University of Science and Technology"],"area":"Deep Learning","sub":"Foundation Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=z1bSFIEexL","vs":"https://icml.cc/virtual/2026/poster/60557","alphaxiv":"2602.17186"},{"i":269,"pid":"62950","orid":"bYDjZyqKxq","title":"MEMO: Memory-Augmented Model Context Optimization for Robust Multi-Turn Multi-Agent LLM Games","authors":["Yunfei Xie","Kevin Wang","Bobby Cheng","Jianzhu Yao","Zhizhou Sha","Alexander Duffy","Yihan Xi","Hongyuan Mei","Cheston Tan","Chen Wei","Pramod Viswanath","Zhangyang “Atlas” Wang"],"insts":["Rice University","University of Texas at Austin","A*STAR"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=bYDjZyqKxq","vs":"https://icml.cc/virtual/2026/poster/62950","alphaxiv":"2603.09022"},{"i":5387,"pid":"61392","orid":"qiZDlnvWTR","title":"NanoQuant: Efficient Sub-1-bit Quantization of Large Language Models","authors":["Hyochan Chong","Dongkyu Kim","Changdong Kim","Minseop Choi"],"insts":["Samsung Research","Samsung"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=qiZDlnvWTR","vs":"https://icml.cc/virtual/2026/poster/61392","alphaxiv":"2602.06694"},{"i":2004,"pid":"63403","orid":"XQlUqVCHJd","title":"TimeSpot: Benchmarking Geo-Temporal Understanding in Vision–Language Models in Real-World Settings","authors":["Azmine Toushik Wasi","Shahriyar Zaman Ridoy","Koushik Tonmoy","Kinga Tshering","S M Muhtasimul Hasan","Wahid Faisal","Tasnim Mohiuddin","Md Rizwan Parvez"],"insts":["Computational Intelligence and Operations Laboratory (CIOL)","Cognitive Agents and Interaction Lab, University of Dhaka","North South University"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=XQlUqVCHJd","vs":"https://icml.cc/virtual/2026/poster/63403","alphaxiv":"2603.06687"},{"i":4217,"pid":"64996","orid":"HwXyyvK7ZJ","title":"VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models","authors":["Rohit Saxena","Alessandro Suglia","Pasquale Minervini"],"insts":["University of Edinbugh","Heriot-Watt University","University College London"],"area":"Deep Learning","sub":"Robustness","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=HwXyyvK7ZJ","vs":"https://icml.cc/virtual/2026/poster/64996","alphaxiv":"2603.06148"},{"i":4095,"pid":"66212","orid":"5kTn1c3vtt","title":"Controlled LLM Training on Spectral Sphere","authors":["Tian Xie","Haoming Luo","Haoyu Tang","Hu Yiwen","Jason Liu","Qingnan Ren","Yang Wang","Xin Zhao","Rui Yan","Bing Su","Chong Luo","Baining Guo"],"insts":["Microsoft","Renmin University of China","None"],"area":"Optimization","sub":"Large Scale, Parallel and Distributed","type":"Poster","spot":true,"or":"https://openreview.net/forum?id=5kTn1c3vtt","vs":"https://icml.cc/virtual/2026/poster/66212","alphaxiv":"2601.08393"},{"i":813,"pid":"61924","orid":"ldCiNVFt8O","title":"d2: Improved Techniques for Training Reasoning Diffusion Language Models","authors":["Guanghan Wang","Gilad Turok","Yair Schiff","Marianne Arriola","Volodymyr Kuleshov"],"insts":["Cornell University","Department of Computer Science, Cornell University","Cornell Tech, Inception"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=ldCiNVFt8O","vs":"https://icml.cc/virtual/2026/poster/61924","alphaxiv":"2509.21474"},{"i":3829,"pid":"64102","orid":"QrC8OgQyOI","title":"Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models","authors":["Jinbin Bai","Yixuan Li","Yuchen Zhu","Yi Xin","Qingyu Shi","Aosong Feng","Xiaohong Liu","Molei Tao","Jianru Xue","Xiangtai Li","Ming-Hsuan Yang"],"insts":["National University of Singapore","Xi'an Jiaotong University","Georgia Institute of Technology"],"area":"Deep Learning","sub":"Generative Models and Autoencoders","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=QrC8OgQyOI","vs":"https://icml.cc/virtual/2026/poster/64102","alphaxiv":"2602.01842"},{"i":1277,"pid":"61345","orid":"r7uOjvZdzO","title":"From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models","authors":["Juncheng Wu","Hardy Chen","Haoqin Tu","Xianfeng Tang","Freda Shi","Hui Liu","Hanqing Lu","Cihang Xie","Yuyin Zhou"],"insts":["University of California, Santa Cruz","University of California Santa Cruz","Amazon"],"area":"Deep Learning","sub":"Large Language Models","type":"Poster","spot":false,"or":"https://openreview.net/forum?id=r7uOjvZdzO","vs":"https://icml.cc/virtual/2026/poster/61345","alphaxiv":"2605.20177"}],"claims":{"71029":[{"text":"Feature Activation Coverage metric measures data diversity in interpretable feature space of LLMs.","status":"unverified"},{"text":"Identifies shared interpretable feature space across model families (LLaMA, Mistral, Qwen) enabling cross-model knowledge transfer.","status":"unverified"},{"text":"FAC Synthesis consistently improves both data diversity and downstream performance on instruction following, toxicity detection, reward modeling, and behavior steering.","status":"unverified"}],"71030":[{"text":"OPUS outperforms industrial-level baselines on 30B-token pre-training of GPT-2 Large/XL on FineWeb and FineWeb-Edu.","status":"unverified"},{"text":"Continued pre-training of Qwen3-8B on SciencePedia: OPUS achieves superior performance using 0.5B tokens compared to full training with 3B tokens.","status":"unverified"}],"71034":[{"text":"Reveals large performance gaps of state-of-the-art mobile GUI agents relative to prior benchmarks.","status":"unverified"},{"text":"Diagnostic analysis shows failures dominated by deficiencies in perception and memory.","status":"unverified"},{"text":"Even strongest agents exhibit near-zero success under environment variations, highlighting brittleness in realistic settings.","status":"unverified"}],"71037":[{"text":"PAR is first multi-scale autoregressive framework for protein backbone generation via coarse-to-fine prediction","status":"unverified"},{"text":"Strong zero-shot generalization supports flexible human-prompted conditional generation and motif scaffolding without fine-tuning","status":"unverified"},{"text":"Favorable scaling behavior with high design quality from robust backbone generation addressing exposure bias","status":"unverified"}],"71038":[{"text":"TEDBench is large-scale non-redundant benchmark for protein fold classification from TED and Foldseek-clustered structures","status":"unverified"},{"text":"MiAE uses up to 90% masking ratio with SE(3)-invariant encoder and lightweight decoder outperforming supervised baselines","status":"unverified"},{"text":"Scales well and extends to experimental structures validating strong recipe for protein fold classification","status":"unverified"}],"71049":[{"text":"Motive-selected high-influence motion data improves motion smoothness and dynamic degree on VBench, achieving 74.1% human preference win rate over pretrained model","status":"unverified"}],"71050":[{"text":"PhotoAgent significantly outperforms existing methods in instruction faithfulness and visual quality across diverse editing scenarios","status":"unverified"},{"text":"Introduces UGC-Edit benchmark with 7,000 photos and learned aesthetic reward model for autonomous photo editing","status":"unverified"}],"71065":[{"text":"Demonstrates obfuscation emerges in realistic reward hacking coding environment via two strategies: obfuscated activations and obfuscated policies.","status":"unverified"},{"text":"Sufficiently high KL regularization and detector penalty reliably yield honest policies in white-box deception detector training.","status":"unverified"}],"71069":[{"text":"Proposes reversible exponential Runge-Kutta solvers enabling exact inversion without discretization error accumulation","status":"unverified"},{"text":"Improves sample quality of Boltzmann distributions with flow models and image generation capabilities with diffusion models","status":"unverified"},{"text":"Solvers overcome ODE domain limitations and improve stability with higher-order convergence","status":"unverified"}],"71077":[{"text":"Memory representation effectiveness is highly task-dependent with each design offering distinct advantages and limitations across different tasks","status":"unverified"},{"text":"RoboMME benchmark comprises 16 manipulation tasks with taxonomy evaluating temporal, spatial, object, and procedural memory","status":"unverified"},{"text":"14 memory-augmented VLA variants built on π₀.₅ backbone systematically explore different memory representations and integration strategies","status":"unverified"}],"71122":[{"text":"RACO provides convergence guarantees to Pareto-critical points respecting user-specified objective weights","status":"unverified"},{"text":"Clipping in conflict-averse gradient descent strictly improves convergence rate in two-objective setting","status":"unverified"},{"text":"Achieves better Pareto trade-offs than existing multi-objective alignment baselines across Qwen, Llama, and Gemma families","status":"unverified"}],"71132":[{"text":"Achieves δ-error in polylog(1/δ) steps for diffusion model sampling, an exponential improvement over previous results","status":"unverified"},{"text":"Under minimal data assumptions: complexity is Õ(d polylog(1/δ)) where d is data dimension","status":"unverified"},{"text":"First polylog(1/δ) complexity sampler for general log-concave distributions using only gradient evaluations","status":"unverified"}],"71142":[{"text":"WeDLM delivers up to 3x speedups on reasoning benchmarks while preserving quality of AR backbones","status":"unverified"},{"text":"Up to 10x speedups achieved in low-entropy generation regimes","status":"unverified"},{"text":"Built entirely on standard causal attention for prefix KV cache compatibility","status":"unverified"}],"71153":[{"text":"Selecting better reasoning at test time leads to up to 19.3% performance gain.","status":"unverified"},{"text":"During RL training, thinking rewards enhance reasoning and performance by up to 3.9% gain across diverse tasks.","status":"unverified"},{"text":"DAG-based pairwise evaluation method successfully captures complex reasoning structures for optimization.","status":"unverified"}],"71154":[{"text":"The paper presents an end-to-end framework for systematic analysis of rare events in LLMs spanning theory, efficient generation strategies, probability estimation, and error analysis","status":"unverified"},{"text":"The framework enables identification of behavior far from typical but highly significant during LLM inference and deployment","status":"unverified"}],"71171":[{"text":"τ²-Bench introduces Dec-POMDP telecom dual-control domain requiring both agent coordination and user guidance","status":"unverified"},{"text":"Agents exhibit significant performance drops when shifting from single-control to dual-control environments, highlighting user coordination challenges","status":"unverified"}],"kpgURPRMGf":[{"text":"JustGRPO achieves 89.1% accuracy on GSM8K with standard GRPO applied to diffusion LLMs","status":"unverified"},{"text":"Retains parallel decoding ability of diffusion LLMs while improving reasoning on math and coding","status":"unverified"},{"text":"Shows arbitrary order generation may limit reasoning potential, enabling solution coverage collapse","status":"unverified"}],"bA6BgSbaUi":[{"text":"GPT-style models have a capacity of approximately 3.6 bits per parameter","status":"unverified"},{"text":"Models memorize until their capacity fills, at which point unintended memorization decreases as models begin to generalize","status":"unverified"},{"text":"The method can estimate total memorization by completely eliminating generalization","status":"unverified"}],"iPjuUQbkfl":[{"text":"Develops random matrix theory framework quantifying how finite datasets shape expectation and variance of learned denoiser","status":"unverified"},{"text":"Shows sampling variability acts as renormalization of noise level through self-consistent relation σ² → κ(σ²)","status":"unverified"},{"text":"Theory sharply predicts behavior of linear diffusion models and validates predictions on UNet and DiT architectures","status":"unverified"}],"5nNNVY8NW4":[{"text":"Grokking stages proven to occur: model overfits early, poor generalization persists long after overfitting, then generalization error eventually becomes arbitrarily small.","status":"unverified"},{"text":"Grokking time can be amplified or eliminated in principled manner through proper hyperparameter tuning.","status":"unverified"},{"text":"Quantitative bounds on generalization delay scale with training hyperparameters in ridge regression with gradient descent and weight decay.","status":"unverified"}],"NUyt4uxzx0":[{"text":"Unfaithful CoT occurs on naturally worded, non-adversarial prompts without artificial biases or edited outputs.","status":"unverified"},{"text":"Production models show up to 13% unfaithfulness rates; frontier models are more faithful but none are entirely faithful, including thinking models.","status":"unverified"},{"text":"Models use subtly illogical reasoning to make speculative answers to hard math problems appear rigorously proven.","status":"unverified"}],"P7RGcAOZZ3":[{"text":"Approach achieves superior fidelity and distributional coverage in diffusion model training","status":"unverified"},{"text":"Preserves geometric structure necessary for effective diffusion model training through one-sided partial optimal transport","status":"unverified"}],"bWLfplRNzt":[{"text":"ProtDBench standardized evaluation framework with unified benchmark tasks, protocols, and success criteria for protein binder design","status":"unverified"},{"text":"Large wet-lab annotated dataset reveals substantial verifier-dependent bias and limited agreement in structure prediction model evaluation","status":"unverified"},{"text":"Benchmarks representative open-source generative binder design methods with throughput-aware metrics accounting for computational budget","status":"unverified"}],"Rl2uQlCoQX":[{"text":"SPEED-Bench quantifies how synthetic inputs overestimate real-world throughput in speculative decoding.","status":"unverified"},{"text":"SPEED-Bench identifies batch-size dependent optimal draft lengths and biases in low-diversity data across speculative decoding evaluation.","status":"unverified"}],"4M5Kj2UqaM":[{"text":"AgentSelect comprises 111,179 queries, 107,721 deployable agents, and 251,103 interaction records aggregated from 40+ sources.","status":"unverified"},{"text":"Models trained on AgentSelect transfer to public agent marketplace (MuleRun) with consistent gains on unseen agent catalog.","status":"unverified"}],"vCc2NAe0OS":[{"text":"Hive dataset (2.4k hours) enables models to achieve competitive accuracy with SAM-Audio trained on ~500x larger dataset","status":"unverified"},{"text":"Models trained on Hive exhibit remarkable zero-shot generalization on out-of-distribution evaluation benchmarks","status":"unverified"},{"text":"Semantic consistency synthesis protocol eliminates event co-occurrence in constructed dataset","status":"unverified"}],"xbAWn0w9kq":[{"text":"DIYHealthGPT delivers state-of-the-art performance over both general-purpose and medical-specific baselines on 11 home care tasks.","status":"unverified"},{"text":"DIYHealth-900K captures diverse real-world home care scenarios with heterogeneous multimodal data.","status":"unverified"},{"text":"DIYHealthGPT uses Hybrid Hyper Low-Rank Adaptation for adaptive foundation model fine-tuning on home care tasks.","status":"unverified"}],"If4X4W2HWx":[{"text":"MemoryBench provides user feedback simulation framework and comprehensive benchmark covering multiple domains, languages, and task types.","status":"unverified"},{"text":"Effectiveness and efficiency of state-of-the-art baselines are far from satisfying on continual learning tasks.","status":"unverified"}],"yU6X1XZl8t":[{"text":"QuArch provides 2,671 expert-validated QA pairs covering processor design, memory systems, and interconnection networks","status":"unverified"},{"text":"Frontier model accuracies range from 34% to 73% on advanced architecture questions, with wide gaps in higher-order reasoning","status":"unverified"},{"text":"Fine-tuning on QuArch yields up to 1.99x more area-efficient memory hierarchy designs and 40% more viable solutions","status":"unverified"}],"ov240fehF6":[{"text":"MORE is the first multilingual document parsing benchmark covering 149 languages, making it the most linguistically diverse benchmark.","status":"unverified"},{"text":"MORE extends evaluation beyond plain text to include complex structural elements such as code blocks, tables, and catalogs.","status":"unverified"},{"text":"All samples in MORE are curated from real-world documents via rigorous model-assisted, human-refined annotation pipeline.","status":"unverified"}],"1ebAvNphi7":[{"text":"Achieves 1.16×–2.16× speedup on cloud-edge collaborative inference with speculative decoding","status":"unverified"},{"text":"Reduces energy consumption by 14.3%–25.3% across real-world testbed scenarios","status":"unverified"}],"Tyv61ZKb9s":[{"text":"SynerMedGen achieves strong zero-shot performance across 22 medical image synthesis tasks with robust generalization to unseen datasets","status":"unverified"},{"text":"Outperforms state-of-the-art specialized medical image synthesis models and recent unified medical models","status":"unverified"},{"text":"SynerMed dataset with 1M paired synthesis samples and 2M generation-derived understanding instances supports unified training","status":"unverified"}],"wTgx7b2D9r":[{"text":"Surpasses 15% Top-1 accuracy on the challenging MassSpecGym benchmark for molecular generation.","status":"unverified"},{"text":"More than doubled the Top-1 accuracy of leading methods on NPLIB1 dataset.","status":"unverified"},{"text":"Achieves log-linear performance scaling with increasing inference-time compute.","status":"unverified"}],"tJP3FxzSPs":[{"text":"video-SALMONN S processes over 3-hour videos at 1 FPS and 360p resolution, outperforming strong non-streaming models under same memory budget","status":"unverified"},{"text":"video-SALMONN S outperforms streaming and non-streaming baselines by 3-7% on long video benchmarks","status":"unverified"},{"text":"video-SALMONN S achieves 15% absolute accuracy improvement over strong non-streaming models on ELViM benchmark","status":"unverified"}],"rzBAQT2Fkg":[{"text":"Achieves up to 10× speedup over vanilla LLaDA/Dream and up to 5× speedup over Qwen-2.5-7B without significant accuracy loss","status":"unverified"},{"text":"Introduces AUP (Accuracy Under Parallelism) metric that jointly measures accuracy and parallelism","status":"unverified"},{"text":"Uses pseudo-trajectory distillation to identify early-step token decoding and entropy-based multi-block decoding for inference","status":"unverified"}],"PoRigyDOcC":[{"text":"Training-consistent segment-level execution achieves approximately 6x lower peak prefill memory at 128K context compared to full-context attention with FlashAttention.","status":"unverified"},{"text":"Segment-level generation achieves performance comparable to full-context attention while maintaining competitive latency-memory trade-offs.","status":"unverified"}],"OMdQJQwp26":[{"text":"Even state-of-the-art model Gemini-2.5-pro achieves only 68.1% accuracy on MedMosaic medical audio question-answering.","status":"unverified"},{"text":"MedMosaic features 46,701 question-answer pairs spanning multiple question types for evaluating multi-hop reasoning.","status":"unverified"},{"text":"Benchmarking reveals persistent limitations in medical reasoning across all evaluated systems.","status":"unverified"}],"XjSd2CtV20":[{"text":"VCG-Bench dataset contains 1,449 diverse diagrams spanning 6 domains and 15 sub-domains with Diagram-as-Code paradigm using mxGraph XML.","status":"unverified"},{"text":"Current SOTA VLMs demonstrate challenges in structured fidelity and instruction compliance reflected in low mxGraph Execution Success Rates.","status":"unverified"},{"text":"Multi-dimensional metrics (mxGraph Execution Success Rate, Style Consistency Score) enable fine-grained evaluation of Vision-to-Code and Code-to-Code performance.","status":"unverified"}],"a86luANykT":[{"text":"Agent-native mid-training with daVinci-Dev achieves 56.1% resolution on SWE-Bench Verified with 32B models","status":"unverified"},{"text":"daVinci-Dev achieves 58.5% resolution on SWE-Bench Verified with 72B models, state-of-the-art for open training recipes","status":"unverified"}],"k7XzObg9Hy":[{"text":"FOAM achieves convergence rates equivalent to vanilla Adam under standard non-convex optimization settings.","status":"unverified"},{"text":"FOAM eliminates up to 90% of the memory overhead of optimizer states and accelerates convergence.","status":"unverified"},{"text":"FOAM is compatible with other memory-efficient optimizers and delivers performance matching or surpassing full-rank baselines.","status":"unverified"}],"RpwnrBkht2":[{"text":"FourTune matches quality of full-precision fine-tuning across customization, RL, and distillation tasks","status":"unverified"},{"text":"On FLUX.1-dev (12B), FourTune reduces memory overhead by 2.25× and increases end-to-end training throughput by 2.27× compared to BF16 LoRA","status":"unverified"}],"xRVWftS3ES":[{"text":"MRAgent with associative memory graph achieves up to 23% improvement on LoCoMo benchmark over strong baselines.","status":"unverified"},{"text":"Active reconstruction mechanism allows iterative exploration and pruning of memory retrieval paths based on accumulated reasoning evidence.","status":"unverified"}],"uyRIOjFgOn":[{"text":"DeMix breaks trade-off between sufficiency, accuracy and efficiency, obtaining optimal mixture with higher benchmark performance at lower search cost","status":"unverified"},{"text":"Decouples search from training costs via model merging to predict optimal data ratios without training proxy models","status":"unverified"}],"NWKaQIKoGp":[{"text":"PlugMem structures episodic memories into a knowledge-centric memory graph with propositional and prescriptive knowledge, enabling efficient memory retrieval over task-relevant knowledge.","status":"unverified"},{"text":"PlugMem consistently outperforms task-agnostic baselines and exceeds task-specific memory designs across three heterogeneous benchmarks while achieving highest information density.","status":"unverified"}],"ic0AGRIkmY":[{"text":"EvoMAS outperforms EvoAgent by +10.5 points on BBEH reasoning and +7.1 points on WorkBench.","status":"unverified"},{"text":"Reaches 79.1% on SWE-BenchVerified with Claude-4.5-Sonnet, matching the top of the leaderboard.","status":"unverified"},{"text":"Produces generated systems with higher executability and runtime robustness compared to code generation and rigid template approaches.","status":"unverified"}],"5vufrrbi4N":[{"text":"TextAtlas5M contains 5 million generated and collected images across diverse data types for long-text image rendering evaluation.","status":"unverified"},{"text":"Curates 4,000 human-improved test cases spanning four domains, forming one of most extensive benchmarks for text rendering.","status":"unverified"},{"text":"Training on TextAtlas5M consistently improves text rendering for both diffusion-based and autoregressive models.","status":"unverified"}],"ww57OvgpP9":[{"text":"Frontier LLMs achieve 46-64% success rate in single-round parameter tuning, dropping to 35-54% under high accuracy requirements","status":"unverified"},{"text":"Multi-round mode improves LLM success rates to 71-80%, but LLMs are 1.5-2.5x slower than traditional scanning","status":"unverified"},{"text":"SimulCost spans 2,916 single-round and 1,900 multi-round tasks across 12 simulators from fluid dynamics, solid mechanics, and plasma physics","status":"unverified"}],"czTbPdmvtT":[{"text":"LayerT2V produces full video, independent background layer, and multiple foreground RGB layers with alpha mattes","status":"unverified"},{"text":"LayerT2V substantially outperforms prior methods in visual fidelity, temporal consistency, and cross-layer coherence","status":"unverified"},{"text":"VidLayer is first large-scale dataset for multi-layer video generation","status":"unverified"}],"913g1X1YAW":[{"text":"Evolutionary potential is not strictly correlated with initial generation proficiency in code agents","status":"unverified"},{"text":"Current LLM agents struggle to concurrently leverage both peer-learning and self-reflection for effective performance gains","status":"unverified"}],"Yhc2PvgFyh":[{"text":"L²-VMAS improves average accuracy by 2.7-5.4% while reducing total token usage by 21.3-44.8%","status":"unverified"},{"text":"Method breaks the scaling wall where increasing agent turns degrades performance","status":"unverified"}],"UxgMJP0JKY":[{"text":"TempCache, AnnCA, and AnnSA methods achieve up to 5-10x end-to-end speedups while preserving near-identical visual quality.","status":"unverified"},{"text":"The framework maintains stable throughput and nearly constant peak GPU memory usage over long rollouts where prior methods progressively slow down.","status":"unverified"}],"0bTEd4LpQr":[{"text":"Numina-Lean-Agent solves all problems in Putnam 2025 (12/12), matching best closed-source system","status":"unverified"},{"text":"Successfully formalizes Brascamp-Lieb theorem by directly using general coding agent as formal math reasoner","status":"unverified"}],"oCNT5PcMSQ":[{"text":"Can accurately detect data contamination or hacking across wide range of benchmarks, models, and training methodologies","status":"unverified"},{"text":"Fully capable models should not surpass the Bayes accuracy; surpassing it signals leakage or gaming","status":"unverified"}],"9wYjjPydfe":[{"text":"TIC-VLA consistently outperforms prior VLA models in robot navigation with explicit latency-aware semantic reasoning.","status":"unverified"},{"text":"TIC-VLA maintains robust real-time control under multi-second reasoning latency in dynamic environments.","status":"unverified"},{"text":"TIC-VLA performs well both in simulation and on real robots for language-guided navigation.","status":"unverified"}],"4jfuNNghPS":[{"text":"FlashBlock achieves up to 1.44× higher token throughput and up to 1.6× reduction in attention time","status":"unverified"},{"text":"Cross-step attention redundancy within blocks enables reuse of stable attention outputs without modifying the diffusion process","status":"unverified"}],"4yzY0GFIJj":[{"text":"Q-Sched achieves 15.5% FID improvement over FP16 4-step Latent Consistency Model.","status":"unverified"},{"text":"Achieves 16.6% improvement over FP16 8-step Phased Consistency Model.","status":"unverified"},{"text":"Provides 4× reduction in model size while preserving single reusable checkpoint across bit-widths.","status":"unverified"}],"vGeNaFHdET":[{"text":"Multi-task learning acts as critical regularizer to mitigate overfitting in data-scarce EEG contexts","status":"unverified"},{"text":"Pre-training efficiency is limited by gradient conflicts between reconstruction objectives and downstream tasks","status":"unverified"},{"text":"Compact architectures with domain-specific inductive biases consistently outperform significantly larger models","status":"unverified"}],"UnjxMTe57e":[{"text":"Best agent achieves 21.5% accuracy on AIME vs 51.1% for official instruction-tuned models under bounded compute constraints","status":"unverified"},{"text":"GPT-5.1 Codex Max achieves 89% on BFCL with Gemma-3-4B vs 67% for official instruction-tuned model","status":"unverified"}],"O88FCPAPAj":[{"text":"METEORA achieves 21.05% higher precision and 13.41% higher recall than baselines","status":"unverified"},{"text":"METEORA increases F1 score from 0.10 to 0.44 under poisoning attacks, a 4.4x improvement","status":"unverified"},{"text":"METEORA improves downstream answer generation accuracy by 33.34% while reducing evidence volume by 80%","status":"unverified"}],"p5QSlnwume":[{"text":"RBench achieves 0.96 Spearman correlation with human judgment for evaluating robot-oriented video generation","status":"unverified"},{"text":"RoVid-X is largest open-source robotic dataset containing 4 million annotated video clips covering thousands of tasks","status":"unverified"}],"8Fhq7QpYfI":[{"text":"WF-Bench dataset spans multiple quantum matter regimes including topological states, Wigner crystals, and superconducting wavefunctions","status":"unverified"},{"text":"Empirical scaling laws characterize representability dependence on system size and key model parameters","status":"unverified"},{"text":"Unified dataset-driven framework enables consistent performance evaluation across neural network wavefunction architectures","status":"unverified"}],"8nti23Zqkt":[{"text":"EGG achieves 2.13x average speedup over PyTorch on real-world GPU kernel generation","status":"unverified"},{"text":"EGG outperforms existing agent-based and RL-based kernel generation approaches","status":"unverified"}],"3EcT46wsdc":[{"text":"TTFS SNN training framework achieves 99.48% on MNIST, 92.90% on Fashion-MNIST, 90.56% on CIFAR10, 70.27% on CIFAR100, 95.83% on DVS Gesture","status":"unverified"}],"cPDN8YlKdQ":[{"text":"DiNa-LRM outperforms existing diffusion-based reward baselines and achieves competitive performance compared to SOTA Vision-Language Model reward functions.","status":"unverified"},{"text":"Maintains substantially lower computational cost compared to VLM-based approaches.","status":"unverified"},{"text":"Improves preference optimization dynamics, enabling faster and more resource-efficient model alignment.","status":"unverified"}],"1tbhBSXcyX":[{"text":"RelayCaching achieves over 80% KV cache reuse and reduces TTFT by up to 4.7x compared to standard pipeline with negligible accuracy degradation","status":"unverified"},{"text":"Yields 3-6x end-to-end speedup across diverse collaborative LLM tasks spanning mathematical reasoning, general knowledge, and code generation","status":"unverified"}],"4P9cEcinYP":[{"text":"R4T improves retrieval quality over strong baselines on Polyvore and music playlist datasets.","status":"unverified"},{"text":"R4T reduces query-time fan-out latency by an order of magnitude compared to deploying RL-tuned LLMs.","status":"unverified"},{"text":"Diffusion-based generative retrieval enables efficient single-pass fan-out in embedding space with objective-aligned training targets.","status":"unverified"}],"o3gN27ITWV":[{"text":"DHSA maintains near-dense accuracy in highly sparse regimes, achieving 12-20% relative accuracy gains over Block Sparse Attention at comparable prefill cost.","status":"unverified"},{"text":"With memory-efficient tiled backend, DHSA delivers up to 10x prefill speedup at 128K context length on LLaMA-3.1-8B (4-bit) on a single 24GB GPU.","status":"unverified"}],"vaRFU0xKQa":[{"text":"Targeted latent-level interventions correct segmentation errors in 70% of failure cases and improve Dice score from 39.4% to 74.2% without retraining.","status":"unverified"},{"text":"Sparse autoencoders trained on SegFormer and U-Net identify stable backbone of shared representations while dataset shift is driven by differential reliance on population-specific latents.","status":"unverified"}],"ea4sx1kxDz":[{"text":"Speculative Coupled Decoding achieves up to 4.2x speedup in image generation and 13.6x speedup in video generation without any degradation.","status":"unverified"},{"text":"Requires only single-line modification to existing algorithm with almost zero overhead.","status":"unverified"}],"aJdgt8xDMy":[{"text":"AVGen-Bench reveals a pronounced gap between strong audio-visual aesthetics and weak semantic reliability in text-to-audio-video generation.","status":"unverified"},{"text":"Text-to-audio-video models show persistent failures in text rendering, speech coherence, physical reasoning, and universal breakdown in musical pitch control.","status":"unverified"}],"WTziQZdpTV":[{"text":"RepoNavigator's 7B model outperforms 14B baselines, and the 14B model surpasses 32B competitors on repository-level issue localization.","status":"unverified"},{"text":"RepoNavigator's 32B model exceeds closed-source models such as GPT-5 on most metrics for code navigation.","status":"unverified"}],"TbUbv5zqTF":[{"text":"SALAAD substantially reduces memory consumption during deployment while achieving performance comparable to ad-hoc methods","status":"unverified"},{"text":"A single SALAAD training run yields a continuous spectrum of model capacities enabling elastic deployment without retraining","status":"unverified"}],"Li5ki5Dopo":[{"text":"Ambient Dataloops achieve state-of-the-art performance in unconditional and text-conditional image generation.","status":"unverified"},{"text":"Ambient Dataloops achieve state-of-the-art performance in de novo protein design through iterative dataset refinement.","status":"unverified"}],"3ExTD9F0u1":[{"text":"Prompting LLMs with generated reasoning traces improves forecasting accuracy significantly compared to direct numerical prediction (e.g., 40.2% to 56.6%)","status":"unverified"},{"text":"Reasoning traces are causally effective for forecasting improvement and useful for evaluation","status":"unverified"},{"text":"Off-the-shelf LLMs consistently struggle with both reasoning and numerical forecasting across multiple domains","status":"unverified"}],"oBgLvd5YC6":[{"text":"SimpleMem achieves 26.4% average F1 improvement in LoCoMo benchmark.","status":"unverified"},{"text":"Reduces inference-time token consumption by up to 30x while maintaining superior performance.","status":"unverified"}],"DVHpvumD60":[{"text":"WarmServe reduces time-to-first-token (TTFT) by up to 50.8× compared to state-of-the-art autoscaling-based system.","status":"unverified"},{"text":"Supports up to 2.5× higher request throughput than GPU-sharing system.","status":"unverified"},{"text":"Model placement algorithm optimizes prewarming decisions to minimize cross-model interference with KV cache reservation strategy.","status":"unverified"}],"PLT2FKIs2c":[{"text":"Introduces LightYourFace-160K (LYF-160K), a large-scale paired dataset with 160,000 before-and-after pairs using physically consistent rendering.","status":"unverified"},{"text":"FiLitDiff is an efficient one-step model enabling fast, controllable, high-fidelity fill lighting at low computational cost while better preserving background illumination.","status":"unverified"}],"QgoRoKIEEr":[{"text":"Achieves 71× speedup on AFHQ while matching or exceeding full-scan baseline performance","status":"unverified"},{"text":"First successful scaling of analytical diffusion to ImageNet-1K","status":"unverified"},{"text":"Posterior Progressive Concentration phenomenon: effective support shrinks from global manifold to local neighborhood as SNR increases","status":"unverified"}],"NMMmwSbzRx":[{"text":"Stream RAG improves QA accuracy by over 20.0% absolute on AudioCRAG benchmark","status":"unverified"},{"text":"Stream RAG achieves state-of-the-art on SLUE-SQA benchmark and outperforms cascaded systems","status":"unverified"},{"text":"Stream RAG reduces latency by up to 57%","status":"unverified"}],"aIFgQusnPy":[{"text":"Diffusion Bridge has lower cost function than Flow Matching, guiding toward more stable and natural trajectories","status":"unverified"},{"text":"Flow Matching interpolation becomes increasingly ineffective when training data size is reduced","status":"unverified"}],"Oz335dV48X":[{"text":"DFlash achieves more than 6× lossless acceleration across range of models and tasks","status":"unverified"},{"text":"DFlash delivers up to 2.5× higher speedup than state-of-the-art EAGLE-3 speculative decoding method","status":"unverified"}],"5f0gw9YpZC":[{"text":"Identifies primary contributing groups more reliably than semantic similarity and gradient-based attribution methods","status":"unverified"},{"text":"Achieves ~100× speedup on CIFAR-10 over Leave-One-Group-Out retraining","status":"unverified"}],"TBaUfO9znF":[{"text":"RL produces true capability gains only when pre-training leaves sufficient headroom and RL data targets edge of competence","status":"unverified"},{"text":"Contextual generalization requires minimal yet sufficient pre-training exposure, after which RL reliably transfers","status":"unverified"},{"text":"Mid-training significantly enhances performance under fixed compute compared with RL only, demonstrating underexplored but central role","status":"unverified"}],"Wfe1iJocjF":[{"text":"FlashOptim reduces per-parameter memory by over 50% while preserving model quality and API compatibility.","status":"unverified"},{"text":"Reduces AdamW memory from 16 bytes to 7 bytes per parameter, or 5 bytes with gradient release.","status":"unverified"},{"text":"No measurable quality degradation on standard vision and language benchmarks including Llama-3.1-8B finetuning.","status":"unverified"}],"a3GdvuPItd":[{"text":"The unified graph-based dataset pruning framework admits a principled greedy solution with formal approximation guarantees.","status":"unverified"},{"text":"On ImageNet-1K with ResNet-50, the method reduces training time by over 40% without sacrificing accuracy.","status":"unverified"}],"prvGhNz39e":[{"text":"SCALE matches or exceeds Adam performance while using only 35-45% of total memory across LLaMA 60M-1B models.","status":"unverified"},{"text":"SCALE outperforms memory-efficient optimizers APOLLO and Muon in both perplexity and memory consumption for LLaMA 7B.","status":"unverified"}],"3qX5RS8kpJ":[{"text":"DVPD achieves state-of-the-art speech enhancement performance with only 35 of the parameters of SOTA lightweight model PGUSE.","status":"unverified"},{"text":"DVPD uses only 40% of the inference MACs compared to SOTA lightweight model PGUSE.","status":"unverified"},{"text":"DVPD uniquely exploits dual nature of spectrograms as both visual textures and frequency-domain representations.","status":"unverified"}],"tI5CFbRhmV":[{"text":"LM-CC correlates more strongly with LLM performance than traditional code complexity metrics","status":"unverified"},{"text":"Lowering LM-CC directly enhances LLM task performance on code understanding and generation","status":"unverified"},{"text":"Proposes code complexity metric from LLM perspective capturing nonlinearity of program semantics","status":"unverified"}],"hJnZKtsDbe":[{"text":"AgentExpt knowledge base links 108,825 accepted papers to their used baselines and datasets for comprehensive experimental design.","status":"unverified"},{"text":"Outperforms strongest baseline with +5.85% gains in Recall@20 and +7.90% in HitRate@10 for baseline and dataset recommendation.","status":"unverified"},{"text":"Collective perception-enhanced retriever and reasoning-augmented reranker produce refined rankings with interpretable justifications.","status":"unverified"}],"fl93PQfTT6":[{"text":"AWQ suffers 16% accuracy drop on LLADA under W4A4 post-training quantization","status":"unverified"},{"text":"DLLMQuant achieves over 10-point accuracy improvement on GSM8K for LLADA under 4-bit quantization","status":"unverified"}],"CzShhpY2qU":[{"text":"Agent Primitives-based MAS improve average accuracy by 12.0-16.5% over single-agent baselines","status":"unverified"},{"text":"Primitives-based MAS reduce token usage and inference latency by approximately 3-4x compared to text-based MAS while incurring only 1.3-1.6x overhead relative to single-agent inference","status":"unverified"}],"ycj3XWCh6E":[{"text":"Current LLM serving systems use general-purpose heuristics (join-shortest-queue, FIFO, LRU) that ignore distinctive structure of LLM inference","status":"unverified"},{"text":"Principled mathematical methods can match or exceed heuristic performance while providing theoretical guarantees across diverse workloads","status":"unverified"}],"FEmXFeqYNZ":[{"text":"Success conditioning exactly solves a trust-region optimization problem maximizing policy improvement subject to χ² divergence constraint","status":"unverified"},{"text":"Relative policy improvement, magnitude of policy change, and action-influence are exactly equal at every state","status":"unverified"}],"2azIa9tfl3":[{"text":"Up to 9 dB PSNR improvements over previous state-of-the-art implicit neural representations.","status":"unverified"},{"text":"Regulating network rank during training substantially improves high-frequency signal fidelity in vanilla MLPs.","status":"unverified"},{"text":"Using optimizers like Muon with high-rank, near-orthogonal updates consistently enhances INR architectures.","status":"unverified"}],"UGAP2F6FfV":[{"text":"Holi-Spatial-4M dataset contains 12K optimized 3DGS scenes, 1.3M 2D masks, 320K 3D bounding boxes, 320K instance captions, 1.2M 3D grounding instances, and 1.2M spatial QA pairs.","status":"unverified"},{"text":"Holi-Spatial significantly outperforms existing feed-forward and per-scene optimized methods on ScanNet, ScanNet++, and DL3DV datasets.","status":"unverified"},{"text":"Fine-tuning Vision-Language Models on spatial reasoning tasks using this dataset leads to substantial improvements in model performance.","status":"unverified"}],"lJpXXwhRRF":[{"text":"VisionWebDev comprises 193 tasks across 16 categories spanning static UI-to-code generation, interactive multi-page frontend reproduction, and long-horizon full-stack website development.","status":"unverified"},{"text":"State-of-the-art models still struggle on full-stack development, with substantial performance gaps across all task levels in the benchmark evaluation.","status":"unverified"}],"WwS8CTpUA6":[{"text":"SceneSmith generates 3-6x more objects than prior methods with <2% inter-object collisions and 96% of objects remaining stable under physics simulation.","status":"unverified"},{"text":"In a user study with 205 participants, it achieves 92% average realism and 91% average prompt faithfulness win rates against baselines.","status":"unverified"}],"7UEBX1KU1y":[{"text":"DPO equivalence with RLHF is conditional on the assumption that RLHF-optimal policies prefer human-preferred responses, which is frequently violated in practice.","status":"unverified"},{"text":"When the equivalence assumption fails, DPO optimizes relative advantage over the reference policy rather than absolute alignment with human preferences, leading to policies that decrease DPO loss while preferring dispreferred responses.","status":"unverified"},{"text":"CPO achieves state-of-the-art performance on standard benchmarks while providing provable alignment guarantees.","status":"unverified"}],"9kJQjx2B80":[{"text":"ProcMEM achieves superior reuse rates and significant performance gains with extreme memory compression.","status":"unverified"},{"text":"ProcMEM enables agents to learn procedural memory from interaction experiences without parameter updates.","status":"unverified"},{"text":"ProcMEM facilitates long-term autonomy through transparent accumulation, refinement, and reuse of procedural knowledge.","status":"unverified"}],"ePFvXPdvhM":[{"text":"PACT significantly reduces safety violations by 31.0% on average while improving task success by 30.7%","status":"unverified"},{"text":"Self-evolving post-training framework projects pretrained diffusion policies onto constraint-feasible regions without demonstration data or task rewards","status":"unverified"},{"text":"Incorporates curriculum progressively tightening constraints while maintaining theoretically bounded policy shift and monotone improvement","status":"unverified"}],"08tW615mgI":[{"text":"PAR presents first multi-scale autoregressive framework for protein backbone generation via coarse-to-fine next-scale prediction.","status":"unverified"},{"text":"PAR exhibits strong zero-shot generalization supporting flexible human-prompted conditional generation and motif scaffolding without fine-tuning.","status":"unverified"}],"zAl9heLw4q":[{"text":"Achieves 74.1% human preference win rate compared with pretrained base model using Motive-selected data.","status":"unverified"},{"text":"Improves both motion smoothness and dynamic degree on VBench benchmark.","status":"unverified"},{"text":"First framework to attribute motion rather than visual appearance in video generative models for data curation.","status":"unverified"}],"eXxFlOPTk4":[{"text":"VideoKR comprises 430K video reasoning examples over 126K newly collected CC-licensed expert-domain videos","status":"unverified"},{"text":"Models post-trained on VideoKR outperform prior post-training approaches on both general and knowledge-intensive video reasoning benchmarks","status":"unverified"}],"t73XUJvyQr":[{"text":"Latent Laplace Diffusion models target as low-dimensional latent trajectory enabling horizon-wide generation without step-by-step integration.","status":"unverified"},{"text":"Modal parameterization via learnable complex-conjugate poles allows direct evaluation over irregular timestamps.","status":"unverified"},{"text":"LLapDiff consistently outperforms baselines in long-horizon forecasting and supports missing-value imputation.","status":"unverified"}],"3BW15kSPfN":[{"text":"Curriculum training on synthetic compositions of 6th-grade math problems improves accuracy on competition-level benchmarks (GSM-Symbolic, MATH-500, AIME).","status":"unverified"},{"text":"The method achieves significantly higher long-horizon improvements than baselines even at high pass@k, and transfers substantially to diverse out-of-distribution ReasoningGym domains.","status":"unverified"}],"pZNo1YWT5x":[{"text":"XDLM surpasses UDLM by 5.4 points on zero-shot text benchmarks.","status":"unverified"},{"text":"XDLM achieves FID 54.1 in few-step image generation compared to MDLM's 80.8.","status":"unverified"},{"text":"XDLM achieves 15.0 MBPP in just 32 steps, effectively doubling baseline performance when scaled to 8B-parameter LLMs.","status":"unverified"}],"mWxEAgz3xu":[{"text":"First systematic study of 306 production LLM agents across 26 domains showing 68% execute at most 10 steps before human intervention","status":"unverified"},{"text":"Finds 70% of production agents rely on prompting rather than weight tuning, and 74% use primarily human evaluation","status":"unverified"},{"text":"Identifies reliability as the dominant challenge in production agent deployment","status":"unverified"}],"lK2o9OjoXf":[{"text":"PLANTAIN yields an ~6% improvement in pass@1 across several challenging math reasoning and coding benchmarks","status":"unverified"},{"text":"PLANTAIN reduces time-to-first-response by over 60% relative to think-then-answer baselines","status":"unverified"}],"Ws8swqL5ob":[{"text":"PhotoAgent formulates autonomous image editing as long-horizon decision-making problem using tree search and closed-loop execution","status":"unverified"},{"text":"Introduces UGC-Edit aesthetic evaluation benchmark with 7,000 photos and learned aesthetic reward model","status":"unverified"},{"text":"Significantly outperforms existing methods in instruction faithfulness and visual quality across diverse editing scenarios","status":"unverified"}],"hCAEcqig2C":[{"text":"Vanilla Bitween discovers RSRs for 43 of 80 functions (54%) in RSR-Bench, including first known reduction for sigmoid","status":"unverified"},{"text":"Agentic Bitween neuro-symbolic approach with LLM agents discovers RSRs for 64 of 80 functions (80%)","status":"unverified"},{"text":"Agentic Bitween outperforms pure neural baselines in both RSR discovery and verification accuracy","status":"unverified"}],"fdkSA4F0lN":[{"text":"Diffusion models trained on unrelated domains (e.g., bedrooms) can recover signals from different domains (e.g., human faces) when measurements are informative","status":"unverified"},{"text":"Weak diffusion priors succeed when high-dimensional measurements are provided, enabling Bayesian posterior concentration","status":"unverified"}],"7pQIzVNctu":[{"text":"Rex family of reversible exponential stochastic Runge-Kutta solvers enables precise inversion of ODE/SDE-based generative models","status":"unverified"},{"text":"Demonstrates utility in improving sample quality of Boltzmann distributions and image generation and editing capabilities of diffusion models","status":"unverified"}],"EeuLO2BjFN":[{"text":"MaxRL outperforms standard RL and GRPO with higher pass@1 and substantially improved pass@k","status":"unverified"},{"text":"Compute-indexed family of objectives interpolates between standard RL and exact maximum likelihood as compute increases","status":"unverified"}],"h7WBYYJF1Q":[{"text":"The diffusion forcing sampler achieves up to 5x speedup in generation for 3.5B recurrent-depth transformers without any tuning.","status":"unverified"},{"text":"Generation with the sampler is theoretically strictly more expressive than baseline autoregressive generation using the same time budget on modern hardware.","status":"unverified"}],"kR4iOTaAOJ":[{"text":"ThunderAgent achieves 1.5-3.6x throughput improvements in serving compared to state-of-the-art inference systems","status":"unverified"},{"text":"ThunderAgent achieves 1.8-3.9x improvements in RL rollout and up to 4.2x disk memory savings","status":"unverified"}],"PnTXyTR2VG":[{"text":"LatentLM surpasses Diffusion Transformers in both performance and scalability in image generation","status":"unverified"},{"text":"LatentLM outperforms state-of-the-art VALL-E 2 in text-to-speech while requiring 10 fewer decoding steps","status":"unverified"}],"PeFSCRulgy":[{"text":"TerminalTraj curated 32K Docker images and generated 50,733 verified terminal trajectories across eight domains","status":"unverified"},{"text":"TerminalTraj-32B achieves 35.30% on TB 1.0 and 22.00% on TB 2.0 with gains of up to 20% over backbone models","status":"unverified"}],"vTp9JToZl9":[{"text":"SplAttN achieves state-of-the-art performance on PCN and ShapeNet-55/34 point cloud completion benchmarks","status":"unverified"},{"text":"Counter-factual evaluation reveals SplAttN maintains robust dependency on visual cues while baselines degenerate into unimodal template retrievers","status":"unverified"}],"3gCdh3u2GK":[{"text":"Mind-Omni establishes state-of-the-art performance among multi-task unified brain-vision-language frameworks.","status":"unverified"},{"text":"Mind-Omni demonstrates performance competitive with, and at times superior to, larger specialized models.","status":"unverified"}],"lwOoBzJykL":[{"text":"Improves mean success rates by 13% on tabletop manipulation over prior SOTA VLA methods","status":"unverified"},{"text":"Improves mean success rates by 14% on mobile manipulation and 14% on dexterous hand manipulation","status":"unverified"},{"text":"Enables efficient reasoning through latent spatio-temporal representations with dual-system architecture","status":"unverified"}],"F9NDKf5oPy":[{"text":"Trains sampling procedures for diffusion language models via reinforcement learning","status":"unverified"},{"text":"Single-layer transformer policy matches SOTA heuristics on block generation while outperforming them in full-diffusion setting","status":"unverified"},{"text":"Trained policies improve both sample quality and token throughput","status":"unverified"}],"XFTTezxLdU":[{"text":"Front-loading as little as 0.1% procedural data significantly outperforms standard pretraining on C4, CodeParrot, and DeepMind-Math datasets.","status":"unverified"},{"text":"Procedural pretraining enables models to reach the same loss value with only 55%, 67%, and 86% of the original data depending on the dataset.","status":"unverified"},{"text":"On context recall (Needle-in-a-haystack), accuracy jumps from 10% to 98% when pretraining on Dyck sequences.","status":"unverified"}],"fXr3uPr1G5":[{"text":"Unified QAT scaling law models quantization error as function of model size, training data volume, and quantization group size.","status":"unverified"},{"text":"Quantization error decreases with model size but rises with more training tokens and coarser quantization granularity in W4A4 regime.","status":"unverified"},{"text":"Activation quantization error in FC2 layer caused by outliers is primary bottleneck of W4A4 QAT error; mixed-precision quantization addresses this.","status":"unverified"}],"Efq2VvYk1o":[{"text":"ThreadWeaver achieves 79.9% on AIME24 and 71.9% average performance on par with sequential reasoning models while delivering up to 1.53x speedup in token latency.","status":"unverified"},{"text":"Trained on Qwen3-8B, matches performance of cutting-edge sequential reasoning models.","status":"unverified"},{"text":"Establishes new Pareto frontier between accuracy and efficiency for parallel reasoning.","status":"unverified"}],"09CSjVeDug":[{"text":"DMPO achieves accuracy improvement of up to 54.3% over previously SOTA baselines and 66.41% over base model on reasoning benchmarks","status":"unverified"},{"text":"Distribution Matching Policy Optimization matches dLLM policy distribution to optimal, reward-tilted one through cross-entropy optimization","status":"unverified"},{"text":"Novel weight baseline subtraction technique enables effective training with small batch sizes","status":"unverified"}],"ENuMNYCiV6":[{"text":"Framework consistently outperforms strong baselines in knowledge-intensive tasks","status":"unverified"},{"text":"Uses internal cognitive signals to partition knowledge space into mastered, confused, and missing regions","status":"unverified"},{"text":"Cognitive consistency mechanism synchronizes subjective certainty with objective accuracy","status":"unverified"}],"PQaxfoEcRc":[{"text":"Removing small fractions of high-influence training samples significantly hinders emergence of targeted interpretable heads","status":"unverified"},{"text":"Highly repetitive structural data (LaTeX and HTML) acts as a catalyst that accelerates emergence of induction heads","status":"unverified"},{"text":"Interventions targeting induction head formation induce concurrent changes in in-context learning capability","status":"unverified"}],"LcswwEzzX7":[{"text":"OXE dataset is highly imbalanced with top four robot types accounting for over 85% of real data, risking overfitting to robot-scene combinations","status":"unverified"},{"text":"OXE-AugE augments OXE with 9 robot embodiments to provide 4.4 million trajectories, more than triple the original dataset","status":"unverified"},{"text":"Augmenting with diverse arms and grippers improves policy performance on augmented robots, unseen robots, and original robots under distribution shifts","status":"unverified"},{"text":"Fine-tuning OpenVLA and π₀ on OXE-AugE improves success rates by 24-45% on previously unseen robot-gripper combinations","status":"unverified"}],"coHiGZOFtS":[{"text":"Reveals large performance gaps of state-of-the-art mobile GUI agents relative to prior benchmarks.","status":"unverified"},{"text":"Diagnostic analysis shows failures dominated by deficiencies in perception and memory.","status":"unverified"},{"text":"Even strongest agents exhibit near-zero success under environment variations, highlighting brittleness in realistic settings.","status":"unverified"}],"Hm8OEDKpiO":[{"text":"Vision-Language Models fail on elementary spatial tasks like block counting due to spatial intelligence gap from missing view-consistent interface.","status":"unverified"},{"text":"3ViewSense significantly outperforms baselines on spatial reasoning benchmarks with consistent gains on occlusion-heavy counting.","status":"unverified"},{"text":"Framework improves stability and consistency of spatial descriptions offering path toward stronger spatial intelligence in multimodal systems.","status":"unverified"}],"vKWxArobP3":[{"text":"Achieves over 35% improvement in counting accuracy on GenEval benchmark by addressing over-counting and repetition failures","status":"unverified"},{"text":"Method is seamlessly applicable to existing diffusion backbones (SD1.5, SDXL, and SD3)","status":"unverified"}],"2hQBG2ZlFb":[{"text":"Near model-strong end, prior-leaning objectives downweighting low-probability tokens consistently outperform NLL.","status":"unverified"},{"text":"Toward model-weak end, NLL dominates objective behavior across conditions.","status":"unverified"},{"text":"Model-capability continuum is critical dimension governing when and why different probability-based objectives succeed or fail.","status":"unverified"}],"KRSZj8z5Lr":[{"text":"LiME achieves up to 4x fewer trainable parameters compared to corresponding MoE-PEFT baselines","status":"unverified"},{"text":"LiME achieves up to 29% faster training compared to MoE-PEFT baselines","status":"unverified"},{"text":"Theoretically, modulation approximates full expert-specific PEFT with bounded error","status":"unverified"}],"A5venxSvpw":[{"text":"MASPOB achieves state-of-the-art performance across diverse benchmarks while maintaining sample efficiency.","status":"unverified"},{"text":"Integrates Graph Neural Networks to capture topology-aware representations of prompt semantics in multi-agent systems.","status":"unverified"},{"text":"Reduces search complexity from exponential to linear through coordinate ascent decomposition.","status":"unverified"}],"giNBsVVFGt":[{"text":"27B tokens mid-training yields ~6-10 point gains on coding benchmarks and ~17-30 points on mathematical reasoning benchmarks","status":"unverified"},{"text":"RL applied on top of PRISM-mid-trained models produces further ~3-8 points on coding/math and ~17-20 points on science (GPQA-Diamond)","status":"unverified"},{"text":"Retention-aware mid-training is necessary intermediate step for reliable reasoning enhancement and RL scaling","status":"unverified"}],"vSzRJyg6k0":[{"text":"RACO framework provides convergence guarantees to Pareto-critical points respecting user-specified objective weights.","status":"unverified"},{"text":"Achieves better Pareto trade-offs than existing multi-objective baselines across Qwen 3, Llama 3, and Gemma 3 families.","status":"unverified"},{"text":"Resolves gradient conflicts via clipped variant of conflict-averse gradient descent.","status":"unverified"}],"8m30ogkPk2":[{"text":"RoboMME comprises 16 manipulation tasks under carefully designed taxonomy evaluating temporal, spatial, object, and procedural memory","status":"unverified"},{"text":"Develops suite of 14 memory-augmented VLA variants built on π0.5 backbone","status":"unverified"},{"text":"Shows that effectiveness of memory representations is highly task-dependent with distinct advantages across different tasks","status":"unverified"}],"pl10b6EQAN":[{"text":"Program-of-Layers enables skipping or looping pretrained layers to form customized programs for each input","status":"unverified"},{"text":"Consistently improves accuracy over standard inference and prior dynamic-depth methods, often executing fewer layers","status":"unverified"},{"text":"Gains persist under out-of-distribution evaluation, suggesting fixed-depth execution captures narrow subset of latent capacity","status":"unverified"}],"KyVlaw4BxE":[{"text":"VideoFlexTok represents videos with variable-length sequences of tokens structured in coarse-to-fine manner, allowing adaptation of token count to particular downstream needs.","status":"unverified"},{"text":"Achieves comparable generation quality (gFVD and ViCLIP Score) with a 10x smaller model (0.4B vs 3.6B) compared to 3D grid tokenizer approach.","status":"unverified"},{"text":"Enables long video generation of 10-second 81-frame videos with only 672 tokens, 8x fewer than comparable 3D grid tokenizer.","status":"unverified"}],"VMsumctGvg":[{"text":"Image-based latent actions benefit long-horizon reasoning while action-based latent actions excel at complex motor coordination.","status":"unverified"},{"text":"Directly supervising VLM with discrete latent action tokens yields most effective performance.","status":"unverified"},{"text":"Latent action supervision offers promising direction for vision-language-action training on mixed datasets.","status":"unverified"}],"HSuU4xBmAv":[{"text":"Top-W geometry-aware truncation consistently outperforms prior state-of-the-art decoding approaches with up to 33.7% improvement on four benchmarks","status":"unverified"},{"text":"Improves both accuracy-focused performance and creativity under judge-based open-ended evaluation across GSM8K, GPQA, AlpacaEval, and MT-Bench","status":"unverified"}],"mDhyxu8WRb":[{"text":"Restores identity mapping property of residual connections while maintaining performance gains of hyper-connections","status":"unverified"},{"text":"Incorporates rigorous infrastructure optimization to ensure efficiency and scalability","status":"unverified"},{"text":"Effective for training at scale with tangible performance improvements and superior scalability compared to HC","status":"unverified"}],"92oF5bU4cU":[{"text":"In large well-pretrained models, density of task-experts increases dramatically in neighborhood around pretrained weights.","status":"unverified"},{"text":"Sampling N parameter vectors, selecting top K, and ensembling via majority vote is competitive with PPO, GRPO, ES.","status":"unverified"},{"text":"Diverse specialists populate substantial fraction of neighborhood around pretrained weights in sufficiently trained models.","status":"unverified"}],"xEgoeNrp8B":[{"text":"Achieves explorable 3D scene video generation with maintained geometric consistency across camera movements.","status":"unverified"},{"text":"Sphere-aware attention mechanism captures true geometric adjacency by reprojecting features onto spherical surface.","status":"unverified"},{"text":"Superior performance in motion range, control precision, and visual quality for panoramic world generation.","status":"unverified"}],"FneePKFVHT":[{"text":"PaperBanana consistently outperforms leading baselines in faithfulness, conciseness, readability, and aesthetics for academic illustrations.","status":"unverified"},{"text":"Introduces PaperBananaBench comprising 292 test cases for methodology diagrams curated from NeurIPS 2025 publications.","status":"unverified"},{"text":"Method effectively extends to generation of high-quality statistical plots.","status":"unverified"}],"yVFxjNzCQm":[{"text":"A LM that generates one additional latent thought per token outperforms a standard model with double the parameters at identical inference cost","status":"unverified"},{"text":"PonderLM-2-Pythia-1.4B pretrained on 300B tokens significantly surpasses vanilla Pythia-2.8B trained on the same data on language modeling and downstream tasks","status":"unverified"}],"jPKqiaPTEd":[{"text":"Introduces TEDBench, large-scale non-redundant benchmark for protein fold classification from Encyclopedia of Domains and Foldseek-clustered AlphaFold structures","status":"unverified"},{"text":"Proposes Masked Invariant Autoencoders using extreme masking ratio up to 90% with SE(3)-invariant encoder","status":"unverified"},{"text":"MiAE scales well and outperforms supervised counterparts on TEDBench and experimental structures from CATH 4.4","status":"unverified"}],"2RJN5vDHG0":[{"text":"Framework provides systematic analysis of rare events in LLMs with practical implementation spanning theory, generation strategies, and probability estimation.","status":"unverified"},{"text":"Rare events are behavior far from typical but highly significant and become prominent in deployment despite being unobserved during development.","status":"unverified"}],"trn64znfNx":[{"text":"Wave-Induced Rotary Encodings (WIRE) rotate tokens based on the graph Laplacian spectrum to inject structural information into attention.","status":"unverified"},{"text":"WIRE recovers regular RoPE on grids and depends asymptotically on the graph effective resistance.","status":"unverified"},{"text":"WIRE is compatible with linear attention, unlike bias-based relative position encodings.","status":"unverified"}],"5VgZUEpK6W":[{"text":"Skill neologisms can improve model capabilities on specific skills without weight updates","status":"unverified"},{"text":"Independently trained skill neologisms can be composed zero-shot","status":"unverified"},{"text":"Provides scalable path towards skill-based continual learning","status":"unverified"}],"vIZz7LvObC":[{"text":"TokSuite releases fourteen pre-trained models with identical architecture, dataset, training budget, and initialization but different tokenizers.","status":"unverified"},{"text":"A multilingual robustness benchmark measures model performance under perturbations in English, Chinese, Farsi, Italian, and Turkish.","status":"unverified"}],"2UH01A9Za0":[{"text":"The SCP-based method achieves state-of-the-art results across multiple LLM model scales","status":"unverified"},{"text":"Sample selection balances global Hessian geometry with sample-wise gradient noise during LLM annealing","status":"unverified"}],"OC2z7iSQKa":[{"text":"Agents show significant performance drops when shifting from single-control to dual-control environments, revealing challenges in guiding user actions in shared, dynamic worlds.","status":"unverified"},{"text":"Fine-grained analysis of dual-control agents reveals distinct error patterns arising from reasoning versus communication/coordination failures.","status":"unverified"}],"BhahZSDowo":[{"text":"Critique-Resilient Benchmarking framework produces stable scores that correlate with external capability measures","status":"unverified"},{"text":"Reformulates benchmarking as adversarial generation-evaluation game enabling evaluation beyond full human comprehension","status":"unverified"}],"hgMZraPlSv":[{"text":"Solving is significantly harder than proving, highlighting alignment tax for constructive reasoning","status":"unverified"},{"text":"Introduces FPS framework with three benchmarks constructed via manual refactoring of over 1,000 problems","status":"unverified"}],"DK0kvnNelx":[{"text":"Enables train-once, deploy-everywhere paradigm with graceful cost-performance trade-off without training from scratch for each budget","status":"unverified"},{"text":"Extracts nested, importance-ordered components from pretrained models for adaptive deployment","status":"unverified"}],"lFkGJ60bGq":[{"text":"CLIP's binding function has high-complexity preventing shared binding mechanism that generalizes to unseen concept combinations","status":"unverified"},{"text":"In controlled transformer models, binding generalization emerges with sufficient data coverage using low-complexity multiplicative interactions","status":"unverified"}],"syG9I9ofd8":[{"text":"LatentMAS outperforms advanced single agents and text-based MAS baselines, achieving up to 14.6% higher accuracy across 9 benchmarks.","status":"unverified"},{"text":"Reduces output token usage by 70.8%–83.7% and provides 4–4.3× faster end-to-end inference while achieving lossless information exchange.","status":"unverified"}],"yKcBAJMPXZ":[{"text":"Current MLLMs struggle with accurate metaphorical video understanding, lagging far behind human level due to defective cross-domain mapping","status":"unverified"},{"text":"MetaphorBoost, an inference-time enhancement framework using metaphor knowledge graph augmentation, achieves consistent performance improvement on metaphorical video understanding","status":"unverified"}],"5PfEQzE9bf":[{"text":"Midtraining benefits largest for domains distant from general pretraining data such as code and math.","status":"unverified"},{"text":"Midtraining consistently outperforms continued pretraining on specialized data both in-domain and in mitigating forgetting.","status":"unverified"}],"m1IRWFAMsa":[{"text":"NorMuon achieves 21.74% better training efficiency than Adam on 1.1B pretraining","status":"unverified"},{"text":"NorMuon achieves 11.31% improvement over Muon while maintaining comparable memory footprint","status":"unverified"}],"06Nk3dJDMq":[{"text":"PSR models consistently outperform constant-coefficient interventions frequently used in the literature","status":"unverified"},{"text":"PSR models achieve performance close to or exceeding prompt steering while maintaining interpretability","status":"unverified"}],"ds3ZOevkwx":[{"text":"Humans show strong metacognitive calibration, adapting or abandoning failed strategies in document retrieval tasks.","status":"unverified"},{"text":"Frontier VLM agents often persist in unproductive loops with diminishing returns rather than demonstrating strategic reasoning.","status":"unverified"}],"RNuC8Nj6rD":[{"text":"TD3B sequence-based generative framework enables design of binders with specified agonist or antagonist behavior","status":"unverified"},{"text":"Directional transition control objective guides ligand generation for G protein-coupled receptors","status":"unverified"},{"text":"Decoupled from binding affinity and unattainable by equilibrium-based or inference-only guidance baselines","status":"unverified"}],"GnqHK8Ww98":[{"text":"Unlocks learning under sparse, binary rewards on datasets with 0/128 success rate using meta-RL for curriculum generation.","status":"unverified"},{"text":"Grounded rewards outperform intrinsic rewards, reliably avoiding instability and diversity collapse modes.","status":"unverified"}],"kcyOjXoUZu":[{"text":"Walrus outperforms prior foundation models on both short-term and long-term prediction horizons.","status":"unverified"},{"text":"Pretrained on 19 diverse scenarios spanning astrophysics, geoscience, rheology, plasma physics, acoustics, and classical fluids.","status":"unverified"},{"text":"Harmonic-analysis-based stabilization and load-balanced distributed training improve forecast stability and transfer performance.","status":"unverified"}],"IMFgiWw4jd":[{"text":"Selecting better reasoning at test time leads to up to 19.3% performance gain in large reasoning models.","status":"unverified"},{"text":"Using thinking rewards in reinforcement learning training enhances reasoning with up to 3.9% improvement across diverse tasks.","status":"unverified"}],"BkMSFFtm5M":[{"text":"CVE-Factory achieves 95% solution correctness and 96% environment fidelity when validated against human expert reproductions.","status":"unverified"},{"text":"Fine-tuned Qwen3-32B improves from 5.3% to 35.8% on LiveCVEBench with CVE-Factory training, surpassing Claude 4.5 Sonnet, with gains generalizing to Terminal Bench (12.5% to 31.3%).","status":"unverified"}],"aIH1jyU37z":[{"text":"Proves a universal approximation theorem for continuous order-equivariant maps","status":"unverified"},{"text":"Provides the first universal approximation theorem for sheaf neural networks","status":"unverified"}],"wsA8LgHU5U":[{"text":"System acquires the ability to construct theories represented as executable programs from observation alone.","status":"unverified"},{"text":"Enables explanation-driven generalization through program-based understanding of observations.","status":"unverified"}],"iJwDylm93H":[{"text":"RoTS-32B achieves state-of-the-art performance on OSWorld with a 47.4% success rate and 33.8% All-Pass@4 score","status":"unverified"},{"text":"The framework creates 800k high quality synthetic data via a tree-based pipeline that proactively discovers diverse error modes","status":"unverified"}],"97NEP1pyS3":[{"text":"DR Tulu-8B substantially outperforms existing open deep research agents by 15.6% over Tongyi DR on average.","status":"unverified"},{"text":"DR Tulu-8B matches or exceeds proprietary deep research agents by 0.7% over OpenAI DR while being 1000x cheaper per query.","status":"unverified"}],"UamxHbDR3p":[{"text":"Formalizes text-driven amodal 3D generation where text prompts steer completion of unseen regions while preserving input observation","status":"unverified"},{"text":"Proves relaxation mechanism is equivalent to applying low-pass filter on generative vector field","status":"unverified"},{"text":"Introduces ExtremeOcc-3D and AmbiSem-3D diagnostic benchmarks for evaluating amodal generation","status":"unverified"}],"AfqsNFzJcs":[{"text":"RGR-GRPO achieves +7.0% on mathematics, +5.4% on physics, +8.4% on chemistry, +6.6% on general reasoning vs verifiable online RL baseline.","status":"unverified"},{"text":"Maintains stable entropy fluctuations during off-policy training and achieves superior pass@k performance.","status":"unverified"},{"text":"Rubric-driven framework enables dense rewards and offline guidance across 14 benchmarks spanning multiple domains.","status":"unverified"}],"7MlfE2Da2W":[{"text":"SCALE improves state-of-the-art Vision-Language-Action models while maintaining single-pass efficiency without additional training, verifiers, or multiple forward passes.","status":"unverified"},{"text":"Joint modulation of visual perception and action based on self-uncertainty enables adaptive execution across varying conditions.","status":"unverified"}],"ATpOQt9VVd":[{"text":"OeMDM unifies masked diffusion models, autoregressive models, and block diffusion in a single framework by interpreting various generation orders.","status":"unverified"},{"text":"LoMDM jointly learns generation ordering and diffusion backbone through a single objective, enabling context-dependent ordering and outperforming discrete diffusion models.","status":"unverified"}],"GyRMbsYFiG":[{"text":"VGGT-Motion achieves state-of-the-art performance in zero-shot, long-range calibration-free monocular SLAM.","status":"unverified"},{"text":"Motion-aware submap construction and anchor-driven direct Sim(3) registration enable efficient and robust global consistency over kilometer-scale trajectories.","status":"unverified"},{"text":"Markedly improves trajectory accuracy and efficiency with linear complexity in submap-level pose graph optimization.","status":"unverified"}],"ncRRCG4BfP":[{"text":"WestWorld achieves significant improvements over competitive baselines in zero- and few-shot trajectory prediction","status":"unverified"},{"text":"The model demonstrates strong scalability across diverse robotic environments and improves performance on downstream model-based control for different robots","status":"unverified"}],"fSnxuRhWoC":[{"text":"End-to-end autoregressive model achieves state-of-the-art FID score of 1.48 without guidance on ImageNet 256×256","status":"unverified"}],"Mkal0hTCnh":[{"text":"MEnvAgent improves Fail-to-Pass rates by 8.6% while reducing time costs by 43% compared to baselines.","status":"unverified"},{"text":"MEnvData-SWE is the largest open-source polyglot dataset with 1,000 tasks across 10 languages with realistic verifiable Docker environments.","status":"unverified"}],"OyPE1ganBR":[{"text":"Stable-GFlowNet provides more stable training compared to GFN while maintaining optimal policy","status":"unverified"},{"text":"Demonstrates overwhelming attack performance and diversity in red-teaming activities","status":"unverified"}],"WJcFtJriqv":[{"text":"GenEval accuracy improves from 69% to 96% and PickScore increases from 20.46 to 23.81 in T2I tasks.","status":"unverified"},{"text":"OCR benchmark accuracy improves from 4% to 57%, achieving state-of-the-art performance in both continuous and discrete diffusion settings.","status":"unverified"}],"8VVbElV06v":[{"text":"UniPercept outperforms existing MLLMs on perceptual-level image understanding across aesthetics, quality, structure, and texture.","status":"unverified"},{"text":"Can serve as plug-and-play reward model for text-to-image generation improving image quality.","status":"unverified"}],"ymHDVBwmta":[{"text":"AGoQ reduces memory by up to 52% and achieves up to 1.34x improvement in training speed on 8B-32B LLaMA models compared to Megatron-LM, COAT, and DeepSpeed.","status":"unverified"},{"text":"Layer-aware activation quantization allocates bit-widths for activations achieving near 4-bit activation storage.","status":"unverified"},{"text":"Gradient quantization with 8-bit storage and precision-preserving 8-bit All-Reduce communication reduces memory and communication time.","status":"unverified"}],"75AYDsndHP":[{"text":"ArBG leads to significant improvements over flow-based models across all benchmarks, particularly in larger peptide systems such as 10-residue Chignolin.","status":"unverified"},{"text":"Robin 132M parameter model reduces zero-shot energy error by ~60% on 8-residue systems compared to previous state-of-the-art.","status":"unverified"}],"MqzZ9X6m7f":[{"text":"Data and Impact Accounting (DIA) standardizes carbon-and-water reporting metadata for AI model derivatives","status":"unverified"},{"text":"Tracking impacts across model lineages reveals aggregate ecosystem footprint currently invisible in open-source AI","status":"unverified"}],"76XSBLdBdg":[{"text":"Hibiki-Zero achieves state-of-the-art translation accuracy, latency, voice transfer and naturalness across five X-to-English simultaneous speech translation tasks","status":"unverified"},{"text":"Accomplishes this without word-level alignments between source and target speech, using only sentence-level aligned data","status":"unverified"}],"9wpwfSJCp9":[{"text":"SleepLM outperforms state-of-the-art in zero-shot and few-shot learning, cross-modal retrieval, and sleep captioning.","status":"unverified"},{"text":"Introduces sleep-text dataset comprising over 100K hours of data from more than 10,000 individuals.","status":"unverified"}],"gc7Gg18ejz":[{"text":"Isotropic Gaussian embeddings improve performance under non-stationarity while reducing representation collapse and neuron dormancy","status":"unverified"},{"text":"Sketched Isotropic Gaussian Regularization is simple and computationally inexpensive method for improving training stability","status":"unverified"},{"text":"Isotropic Gaussian embeddings induce stable tracking of time-varying targets and achieve maximal entropy under fixed variance budget","status":"unverified"}],"XztRm216YS":[{"text":"TimeRewarder achieves nearly perfect success in 9/10 Meta-World tasks with only 200,000 interactions per task.","status":"unverified"},{"text":"TimeRewarder outperforms previous methods and manually designed environment dense reward on both success rate and sample efficiency.","status":"unverified"}],"gz7hVnrRWq":[{"text":"Critique-GRPO achieves average Pass@1 improvements of +15.0-21.6% on various Qwen models and +7.3% on Llama-3.2-3B-Instruct through combined natural language and numerical feedback","status":"unverified"},{"text":"Critique-GRPO facilitates effective self-improvement through self-critiquing, achieving +16.7% Pass@1 improvement on AIME 2024","status":"unverified"}],"z1bSFIEexL":[{"text":"Visual Information Gain (VIG) metric measures reduction in prediction uncertainty provided by visual input.","status":"unverified"},{"text":"VIG-guided selective training improves visual grounding and mitigates language bias while requiring significantly reduced supervision.","status":"unverified"},{"text":"The method focuses exclusively on visually informative samples and tokens to achieve superior performance.","status":"unverified"}],"bYDjZyqKxq":[{"text":"MEMO raises mean win rate from 24.9% to 49.5% for GPT-4o-mini and from 21.7% to 44.3% for Qwen-2.5-7B using 2000 self-play games per task.","status":"unverified"},{"text":"Reduces run-to-run dispersion of end-to-end outcomes and yields more reliable rankings under prompt stratification.","status":"unverified"},{"text":"Achieves gains in negotiation games and imperfect-information settings while RL remains more effective in perfect-information games.","status":"unverified"}],"qiZDlnvWTR":[{"text":"NanoQuant is the first post-training quantization method to compress LLMs to binary and sub-1-bit levels.","status":"unverified"},{"text":"Compresses Llama2-70B by 25.8x in 13 hours on a single H100, enabling operation on consumer 8GB GPU.","status":"unverified"},{"text":"Achieves state-of-the-art accuracy at sub-1-bit compression rates, establishing new Pareto frontier in low-memory quantization.","status":"unverified"}],"XQlUqVCHJd":[{"text":"TimeSpot benchmark contains 1,455 ground-level images from 80 countries requiring geo-temporal reasoning","status":"unverified"},{"text":"State-of-the-art VLMs show consistently low performance, particularly for temporal inference tasks","status":"unverified"},{"text":"Includes spatial-temporal reasoning tasks probing physical plausibility and cue integration","status":"unverified"}],"HwXyyvK7ZJ":[{"text":"VLM-RobustBench spans 133 corrupted settings across 49 augmentation types with graded severities evaluating four VLM families","status":"unverified"},{"text":"Visual severity is weak predictor of difficulty; low-severity glass_blur reduces MMBench accuracy by ~8pp while geometric distortions reach 34pp","status":"unverified"},{"text":"Current VLMs are semantically strong but spatially fragile with resampling and geometric invariances key vulnerability","status":"unverified"}],"5kTn1c3vtt":[{"text":"SSO consistently outperforms AdamW and Muon across diverse architectures including Dense 1.7B, MoE 8B-A1B, and 200-layer DeepNet models.","status":"unverified"},{"text":"SSO enforces strict module-wise spectral constraints on both weights and updates, realizing fully μP-aligned optimization.","status":"unverified"},{"text":"Improved MoE router load balancing, suppressed outliers, and strictly bounded activations observed.","status":"unverified"}],"ldCiNVFt8O":[{"text":"d2 sets new state-of-the-art performance for DLMs on logical reasoning tasks (Countdown and Sudoku) and math reasoning benchmarks (GSM8K and MATH500)","status":"unverified"},{"text":"d2 significantly outperforms widely-used RL baselines when applied to popular diffusion language models","status":"unverified"}],"QrC8OgQyOI":[{"text":"LLaDA-S achieves best-of-N performance with substantially fewer function evaluations across four mathematical reasoning and code generation benchmarks.","status":"unverified"},{"text":"Performs efficiently across three discrete diffusion language models including LLaDA 8B and Dream 7B.","status":"unverified"}],"r7uOjvZdzO":[{"text":"Staged training achieves 1.5% higher reasoning accuracy with 20.8% shorter reasoning traces than merged training.","status":"unverified"},{"text":"Staged-training models achieve +5.2% on WeMath and +3.7% on RealWorldQA compared to base counterpart.","status":"unverified"}]},"areas":["Deep Learning","Applications","Uncategorized","General Machine Learning","Social Aspects","Theory","Reinforcement Learning","Optimization"]}