如何泛化AI的深度推理能力？
作者：系统智能组 编者按：大语言模型的推理能力一直是人工智能领域的研究热点，但传统依赖大规模数据和参数扩展的预训练方式在提升模型推理能力上逐渐遇到了瓶颈。微软亚洲研究院的最新研究关键计划步骤学习 CPL（Critical Plan Step Learning），旨在将强化学习扩展到更广泛、更复杂的问题场景，并取得了突破性进展。CPL 通过在自我生成的高层次抽象计划上进行强化学习，不仅提升了模型在数...
USENIX ATC 2024 best paper | How Microsoft is improving cloud AI infrastructure reliability
As cloud AI workloads grow in complexity and scale, maintaining high system reliability has become crucial. Traditional methods of ensuring system reliability, such as using redundant components, inadvertently introduce a new problem: subtle performance degradation, also known as “gray failures”.…
In the news | Microsoft on the Issues
Advancing biodiversity with AI
The health of our society is deeply intertwined with the health of our planet. While much of the global conversation around the environment focuses on the devastating impacts of climate change, it is crucial to recognize that climate and biodiversity…
刘海广：发挥“生物多样性”法则的力量，寻找科学的新答案
编者按：刘海广是一位典型的跨学科研究者，他的学术足迹遍布物理、生物、计算机科学和人工智能等多个领域，在跨学科的学习与研究中不断探索和突破。如今，作为微软研究院科学智能中心（Microsoft Research AI for Science）的首席研究员，刘海广在跨学科研究中取得了哪些成果？他和团队又将如何利用人工智能技术来加速科学研究进程，并推动科研成果向实际应用转化？ 从清华大学转学至香港浸会大...
In the news | Eye on AI (podcast)
Why AI Agents Are the Next Big Thing in Tech
In this episode of the Eye on AI podcast, we dive deep into the world of AI agents with Ece Kamar, VP of Research and Managing Director of AI Frontiers Lab at Microsoft. Ece shares her unique insights on the…
In the news | GeekWire
AI Dreams: Microsoft @ 50, Chapter 1
Since the early 1990s, the promise of AI has been a driving force at Microsoft Research, which has a track record of breakthroughs in speech recognition, computer vision, machine learning, and other research that continues to advance the state of…
In the news | Venture Beat
Microsoft’s Differential Transformer cancels attention noise in LLMs
Improving LLMs’ ability to retrieve in-prompt information can impact important applications such as retrieval-augmented generation (RAG) and in-context learning (ICL). Microsoft Research and Tsinghua University researchers have introduced Differential Transformer (Diff Transformer), a new LLM architecture that amplifies attention to relevant context while…
编者按：随着人工智能技术的快速发展，数据可视化日渐高效、智能。但自动化生成的图表是否可靠，成为了亟待解决的问题。微软亚洲研究院提出了 VisEval 评估框架，为这一挑战提供了解决方案，并因此荣获全球可视化领域顶尖的学术会议 IEEE VIS 2024 的最佳论文奖。通过高质量的数据集和可靠的自动化评估方法，VisEval 为数据可视化的未来发展提供了坚实的基础，助力数据可视化技术向更智能、更便捷...
In the news | LinkedIn
Microsoft gives $5M to UW Medicine to use AI to accelerate response to next pandemic (update)
David Baker received the Nobel Prize today for breakthrough work in protein design, on creating new proteins to serve as novel therapies and vaccines. Five years ago, we made the decision to go all-in with David. The goal was to…