ニュース&特集
编者按：大语言模型（LLMs）在语言生成与基础推理中已展现出强大的能力，但它们在数学解题上的能力仍存在明显短板，尤其是难以兼顾复杂计算与定理证明。这背后的关键原因在于，现有模型普遍依赖于单一的推理范式（如自然语言、代码或符号推理），缺乏人类思考问题时那种灵活的推理能力。 为此，微软亚洲研究院与清华大学联合提出了“推理链”（Chain-of-Reasoning, CoR）框架，引入了自然语言、代码与…
One of the driving forces behind AI’s rapid progress is access to large-scale, high-quality data, essential to enable training models to continuously improve and perform reliably. But that well is running dry. As the supply of usable internet data shrinks,…
World models are a key concept in AI, used to simulate how agents behave in virtual environments and enable immersive, interactive experiences. They’re not only transforming game and media generation, they’re also opening new frontiers for using AI in complex,…
编者按：数据是人工智能发展的“动力燃油”，但如今其正面临“枯竭”的风险，这道“数据墙”成为制约大模型性能突破的关键瓶颈。在此背景下，合成数据技术应运而生。近期，微软亚洲研究院推出了一个可扩展的 SYNTHLLM 框架，能够生成多样化的合成数据，有效填补自然数据的空缺。此外，研究员们还发现并证实了合成数据的规模法则，为大模型使用合成数据进行训练与优化提供了科学依据。 人工智能在当今取得如此显著发展的…
世界模型（world models）是人工智能领域的一个重要概念，旨在通过模拟虚拟世界中主体行为的演变，实现高度逼真的互动体验。这种模型不仅可以为游戏和互动媒体的生成带来革命性的变化，也将为人工智能在复杂环境中的应用提供新的可能性。其中，生成式游戏（generative games）作为构建世界模型的关键途径，备受关注。例如，微软提出的 MUSE 能够用神经网络生成游戏《嗜血边缘（Bleeding…
Research Focus: Week of January 22, 2024
Welcome to Research Focus, a series of blog posts that highlights notable publications, events, code/datasets, new hires and other milestones from across the research community at Microsoft. Join Microsoft Research Forum (opens in new tab) for a continuous exchange of…
Research Focus: Week of January 8, 2024
| Zinan Lin, Jinyu Li, Bhaskar Mitra, Siân Lindley, Liang Wang, Nan Yang, と Furu Wei
Mixture-of-linear-experts for long-term time series forecasting; Weakly-supervised streaming multilingual speech model with truly zero-shot capability; KBFormer: Diffusion model for structured entity completion; Identifying risks of AI-mediated data access:
Research Focus: Week of October 23, 2023
In this issue: Kosmos-2.5: A Multimodal Literate Model; Can vine copulas explain complex relationships of weather variables; New system accelerates the adaptive training process; Structural inequalities and relational labor in the influencer industry.
Research Focus: Week of September 11, 2023
In this issue: Efficient polyglot analytics on semantic data aids query performance; generative retrieval for conversational question answering improves dialogue-based interfaces; a new tool uses ML to address capacity degradation in lithium-ion batteries.