Tag
Microsoft introduces 'Microsoft Research Focus', a new video/podcast digest covering AI agents, everyday AI usage patterns, BioEmu2's public preview, and weather forecasting research.
微软研究院及其合作者通过人类主导的智能体研究流程，解决了存在31年之久的KLS猜想，确立了对数凹分布的普适几何原理，并经独立研究者验证与机器形式化检验确认，为高维采样与随机算法效率理解扫清重大理论障碍。
一篇微软研究院等机构的论文提出 Self-HPO，一种交替优化 harness 与策略模型的递归自我改进框架，将 Qwen3.5-35B-A3B 在三个编码基准上的平均解决率从 57.8% 提升到 64.8%，且联合增益超过各自增益之和。
在 OCP 全球峰会上，MediaTek 现场演示了采用标准可插拔形态的 MicroLED 光链路，基于微软研究院的技术，用低成本 LED 替代铜线或激光进行多路并行数据传输，助力 AI 数据中心降低功耗并提升可靠性。
微软研究院发布 Magnetar，一个让 Rust、Go 和 C 程序直接生成零知识证明的通用系统，速度比现有同类系统快 10 到 39 倍，且仅依赖标准哈希函数实现抗量子安全，推动零知识证明走向主流应用。
微软研究院发布并发表于 Nature 的机器学习框架 Probabilistic Bias Correction (PBC)，通过学习现有天气模型的系统性误差并加以校正，显著提升两到六周的次季节天气预报准确性，在全球竞赛中夺冠，并已用于肯尼亚季风预报等实际场景。
Microsoft Research 介绍 StudentSim 研究系统，通过基于公开去标识化学习记录训练的个性化 AI 学生模拟器，在真实学习者参与之前评估并改进 AI 辅导工具，同时发布跨国际象棋、英语写作和数学的评估基准 StudentSimEval。
Microsoft researchers propose ActiveSaddler, a method that selects training tasks for agent harness auto-tuners based on unresolved failure patterns, boosting pass rates by 4.4 points on GAIA2 and 7.5 on Terminal-Bench 2.0 while cutting tuning costs sharply.
Microsoft's M365 Research introduces FOCUS, a training-free test-time method that compresses LLM agent interaction histories by retaining only the units that causally shape future decisions, cutting peak context by up to 48% and boosting task success by up to 8.9 points over full-history execution.
Microsoft Research introduces Quine, an early-stage AI system that creates a multimodal world model of biology to help scientists computationally explore complex hypotheses and prioritize experiments.
Researchers at Microsoft introduced DualGraph, an architecture that splits agent memory into Outline and Knowledge Graphs to enhance autonomous research, achieving unprecedented RACE scores and cost savings with Claude Opus 5.5 and GPT-6 Sol.
Microsoft Research's RetroChimera, a predictive model for retrosynthesis, is published in Nature, helping to accelerate chemical synthesis for advances in medicine, materials, and agriculture.
Microsoft Research presents FrogNano, a 4B coding agent trained exclusively via reinforcement learning with online task synthesis, achieving competitive performance without distillation from larger models.
Microsoft Research is announcing its undergraduate internship program for summer 2025, offering 12-week research experiences in labs across Redmond, New York City, and New England.
Undergraduate research intern Matheus Kunzler Maldaner explores agentic systems and shares his experience in the first episode of Microsoft Research's series on research careers.
Microsoft Research's paper on scalable GPU acceleration for scalar functions in analytical databases received an honorable mention at VLDB 2026, showcasing advancements in compilation and optimization techniques.
Agent Lightning v1.0 is a lightweight framework from Microsoft that connects agent harnesses to reinforcement learning training, improving Qwen3.5-9B's performance on SWE-bench Verified using modest compute.
This podcast episode from Microsoft Research discusses the dual role of AI in contributing to and addressing climate change, emphasizing the need to steer technology towards sustainable solutions.
Microsoft announces its participation in SIGCOMM 2026, featuring sessions on network traffic engineering, load balancing, adaptive photonic switching, and AI network simulation.
New research from Microsoft Research measures cross-lingual policy retention of tool-using agents by comparing action traces rather than final answers. After removing five confounds, four frontier models retain 71–73% of their action policy across languages, with code and 2.38M rollouts released.