从直觉到“深度思考”：多维进化的大模型推理能力
编者按：尽管近年来人工智能的能力迅速增强，但在复杂的推理任务中仍存在不足。微软亚洲研究院的研究员们从多个角度对此展开研究，不断探索提升大模型推理能力的新途径。从利用蒙特卡洛树搜索模拟人类“深度思考”过程的 rStar-Math，到基于规则的强化学习方法 Logic-RL；从融合大语言模型数学直觉与符号方法的 LIPS，到提升自动形式化准确性的新框架；再到自动生成高质量、有监督数学数据的神经符号框架...
特征树驱动的数据合成框架，加速构建高质量模型基座
作者：工程与基础架构组 编者按：随着代码大模型能力的不断增强，高质量指令数据的构造成为释放其潜力的关键。然而，现有方法普遍依赖单一的代码片段作为构造种子，限制了数据的复杂度与多样性。近日，微软亚洲研究院联合厦门大学、清华大学提出全新特征树驱动的数据合成框架，通过建模代码语义层级关系，实现了对合成代码复杂度的精细控制，并支持从函数级到多文件级的多样任务生成。基于该框架训练得到的 EpiCoder 模...
Time-series data—measurements collected over time like stock prices or heart rates—plays a vital role in AI forecasting systems across industries. As these systems advance, the need for time-series data is increasing, especially synthetic data, which offers numerous advantages over real-world…
Math is more than a school subject—it's the engine behind scientific discovery, driving advances in everything from climate modeling to AI. At Microsoft Research Asia, senior researcher Xian Zhang is leading efforts to help AI move beyond surface-level pattern recognition toward…
Predicting and explaining AI model performance: A new approach to evaluation
| Lexin Zhou and Xing Xie
ADeLe, a new evaluation method, explains what AI systems are good at—and where they’re likely to fail. By breaking tasks into ability-based requirements, it has the potential to provide a clearer way to evaluate and predict AI model performance.
In the news | The Microsoft Cloud
Microsoft as customer zero: Empowering research teams with AI
Research has always been an integral part of Microsoft’s identity, driving our role as a global technology leader. Since 1991, Microsoft Research has dedicated itself to a fundamental research approach—advancing knowledge, deepening our understanding of the world, and exploring how…
In the news | Mayo Clinic
Peter Lee, Ph.D., president of Microsoft Research, elected to Mayo Clinic Board of Trustees
Peter Lee, Ph.D., president of Microsoft Research, was elected today to the Mayo Clinic Board of Trustees. Dr. Lee oversees Microsoft Research's 13 global laboratories, driving advancements in artificial intelligence, computer science, health and life sciences while fostering the development…
Abstracts: Heat Transfer and Deep Learning with Hongxia Hao and Bing Lv
| Gretchen Huizinga, Hongxia Hao, and Bing Lv
Silicon has long borne the burden of heat transfer in electronics, but in a post-Moore’s Law world, researchers like Hongxia Hao and Bing Lv are using AI to discover and design next-generation materials that exceed the limits of silicon’s thermal…
微软亚洲研究院发布社会责任人工智能白皮书，助力构建以人为本的AI系统
编者按：在 AI 技术不断突破与快速落地的今天，如何确保其发展方向与人类价值观协调一致，正成为科技界与社会各界亟需回答的关键问题。近日，微软亚洲研究院发布了《社会责任人工智能：研究的挑战与机遇》白皮书，系统梳理了 AI 在社会实践中的复杂挑战，并提出了包括技术、伦理、治理等多维度的十大关键研究问题。通过跨越计算机科学与社会科学的深度对话，该白皮书深入探讨了如何设计“负责任的人工智能”，推动实现技术...