Microsoft Research · 36 records

dateidtitleauthorsvenue↑
2026-11-102611.10329SIRIS: Can LLMs Tell When Simulation Errors Matter in AI-Driven Systems ResearchJinwoo Hwang et al.arXiv12
2026-11-052610.41432Hybrid Computer Use: A Unified Programmatic, Semantic, and Visual Interface for General AgentsHaoqing Wang et al.arXiv14
2026-11-022610.39276MemSteer: Turning User Histories into Supervision for Agent Memory OptimizationJie Huang et al.arXiv8
2026-10-202610.22791Self-Evolving Coding Agents via Harness-Policy OptimizationZhepei Wei et al.arXiv54
2026-10-192610.26828Agent World Model in the Wild: Computer-Use Agent Environment SynthesisZhaoyang Wang et al.arXiv99
2026-10-192610.22732TokenEconomizer: Experience-Driven Black-Box Cost Control for Coding AgentsAaron J. Li et al.arXiv7
2026-10-152610.18097APT: Feedback-Driven Agentic Post-Training for Text-to-Image GenerationSirui Zhang et al.arXiv23
2026-10-072610.04933DiVeR: Decision-Critical Verifier Learning for VLA Test-Time ScalingSeongheon Park et al.arXiv16
2026-10-052610.04116CUAWright: A Minimal Unified Interface for Digital AgentsYadong Lu et al.arXiv6
2026-09-292609.35706Reinforcing Agentic Creativity in Scientific Ideation with Night SciencePriyanka Kargupta et al.arXiv5
2026-09-232609.26781Agensh: Scaling Organizational Intelligence to 1,024 AgentsZhihao Zhan et al.arXiv27
2026-09-222609.13231ShieldVLA: Feasibility-Aware Safety Alignment for Vision-Language-Action ModelsManan Tayal et al.arXiv16
2026-09-212609.20886BI-Agent and BI-Bench: Towards Automating End-to-End Business IntelligenceChuxuan Hu et al.arXiv26
2026-09-172609.18805ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE TasksJeonghye Kim et al.arXiv46
2026-09-152609.13058Expert-Space Exploration in MoE Reinforcement LearningHongyi He et al.arXiv6
2026-09-022609.01591StudentSim: Training LLM-based Student SimulatorsKe Yang et al.arXiv274
2026-08-172608.11612Dion3: Full-Stack Orthogonal UpdatesNoah Amsel et al.arXiv11
2026-08-142608.08888Full-bandwidth transformerXi Wang et al.arXiv15
2026-08-102608.02831Reinforcement Learning with Evolving Rubrics as Rewards for Audio ReasoningFangxu Yu et al.arXiv8
2026-08-032607.26246Weak-to-Strong On-Policy DistillationFangxu Yu et al.arXiv19
2026-07-312607.28074Echoverse: Deep, Evolving Environments for Training Computer-Use Agents at ScaleYash Pandya et al.arXiv11
2026-07-242607.20734LLMs Get Lost in Evolving User IntentJihoon Tack et al.arXiv22
2026-07-242607.04763Multi-Turn On-Policy Distillation with Prefix ReplayBaohao Liao et al.arXiv11
2026-07-212607.18110LLM-as-a-Coach: Experiential Learning for Non-Verifiable TasksTianzhu Ye et al.arXiv16
2026-07-202606.29538RESOURCE2SKILL: Distilling Executable Agent Skills from Human-Created Multimodal ResourcesYijia Fan et al.arXiv88
2026-07-022607.00466ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE ServingSangjin Choi et al.arXiv24
2026-06-292606.18953Object-Centric Residual RL for Zero-Shot Sim-to-Real VLA EnhancementKinam Kim et al.arXiv7
2026-06-112605.11212ReVision: Scaling Computer-Use Agents via Temporal Visual Redundancy ReductionAmirhossein Abaskohi et al.arXiv7
2026-06-102606.05922Retrospective Harness Optimization: Improving LLM Agents via Self-Preference over Trajectory RolloutsWenbo Pan et al.arXiv36
2026-06-092606.09828Latent Spatial Memory for Video World ModelsWeijie Wang et al.arXiv65
2026-05-262605.24517ECHO: Terminal Agents Learn World Models for FreeVaishnavi Shrivastava et al.arXiv7
2026-05-252605.23904SkillOpt: Executive Strategy for Self-Evolving Agent SkillsYifan Yang et al.arXiv219
2026-05-252605.23899From Raw Experience to Skill Consumption: A Systematic Study of Model-Generated Agent SkillsZisu Huang et al.arXiv26
2026-05-202605.15458Video Models Can Reason with Verifiable RewardsTinghui Zhu et al.arXiv4
2026-05-152605.15040Orchard: An Open-Source Agentic Modeling FrameworkBaolin Peng et al.arXiv20
2026-05-122605.10781Rebellious Student: Reversing Teacher Signals for Reasoning Exploration with Self-Distilled RLVRJeonghye Kim et al.arXiv15