为什么你的LLMs玩不转外部知识？RAG分类学助你诊断！
作者：系统组（上海） 编者按：大语言模型在教育、医疗、金融等多领域的应用已展现出其不可忽视的价值。如何更好地结合外部数据，如何提升模型处理专业领域问题的可靠性，是大语言模型应用开发中值得不断思考的问题。针对此，微软亚洲研究院的研究员们提出了一种基于查询需求分层的 RAG 任务分类法，从显式事实、隐式事实、可解释的推理、隐式推理4个层级出发，直指大模型应用在不同认知处理阶段所面临的难点和定制化的解决...
In the news | LinkedIn Article
Juan Lavista Ferres at the United Nations General Assembly’s Summit of the Future
At Microsoft, we’ve seen how digital technology can drive progress toward the Sustainable Development Goals. But technology must be more than just powerful—it must be trustworthy, safe, and inclusive. It’s only when technology empowers everyone, especially the most vulnerable, that…
守护记忆：多模态大模型为认知障碍患者带来全新的训练方法
编者按：每年的9月21日是“世界阿尔茨海默病日”，阿尔茨海默病作为最常见的认知障碍，一直以来都备受关注。研究证明，科学的认知训练可以起到对该疾病的预防和延缓。为此，微软亚洲研究院与上海市精神卫生中心携手展开联合研究，基于微软 Azure OpenAI 服务中的多模态大模型，开发了利用多种模态数据（如语音、文字、图片、音乐等信息）的个性化认知训练框架，为认知障碍患者的认知训练带来了新的可能。 随着全...
金融模型、LLMs结构化剪枝、多模态对齐、AI芯片深度学习编译器
编者按：欢迎阅读“科研上新”栏目！“科研上新”汇聚了微软亚洲研究院最新的创新成果与科研动态。在这里，你可以快速浏览研究院的亮点资讯，保持对前沿领域的敏锐嗅觉，同时也能找到先进实用的开源工具。 本期，我们为大家精心挑选了4篇论文，内容涵盖金融市场元智能体模型、大语言模型的结构化剪枝、多模态感知中的模态对齐以及 AI 芯片的深度学习编译器，感兴趣的朋友快来看看吧！ 本期内容速览 01. 通过扩展式模态...
In the news | Communications of the ACM
AI Should Challenge, Not Obey
“How should we evaluate the legacy of Thomas Jefferson?,” asks a professor of American history. The reply: “The general consensus on Thomas Jefferson is that he was a complex and contradictory figure who championed the ideals of democracy, tolerance, and…
Eureka: Evaluating and understanding progress in AI
| Vidhisha Balachandran, Jingya Chen, Neel Joshi, Besmira Nushi, Hamid Palangi, Eduardo Salinas, Vibhav Vineet, James Woffinden-Luey, and Safoora Yousefi
How can we rigorously evaluate and understand state-of-the-art progress in AI? Eureka is an open-source framework for standardizing evaluations of large foundation models, beyond single-score reporting and rankings. Learn more about the extended findings.
In the news | The National
Microsoft to open its first Middle East ‘AI for Good Lab’ in Abu Dhabi
Plans to create a centre, founded and funded by Microsoft and G42, that will "identify, develop and advance best practices and industry standards for the responsible use of AI in the Middle East and the Global South" were unveiled. Microsoft,…
In the news | Live Science
‘Their capacity to emulate human language and thought is immensely powerful’: Far from ending the world, AI systems might actually save it
From disaster recovery to conservation and healthcare, plenty of AI projects will greatly benefit humanity, Microsoft experts Juan M. Lavista Ferres and William B. Weeks say. Over the last few years, artificial intelligence (AI) has been firmly in the world's spotlight, and…
星跃重洋 | 刘国栋：非典型理工男在微软亚洲研究院的科研“旅”记
结束一天的科研工作，中国科学院计算技术研究所博士生刘国栋走出微软大厦，抬头是繁星点点，耳机里装的是苏打绿乐队温暖的旋律。 通过“星跃计划”在微软亚洲研究院实习一年，刘国栋在微软亚洲研究院主管研究员苗又山、微软雷德蒙研究院高级研发工程师 Saeed Maleki 两位 mentor 的指导下，围绕加速深度学习模型的训练进行着科研探索。 恰如设立初衷，“星跃计划”在优秀人才与微软全球两大研究院的研究团...