News & features
SYNTHLLM框架：突破人工智能“数据墙”瓶颈，为AI训练注入新动力
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…
MaaG：迈向世界模型，解决生成式游戏中的数值与时空一致性难题
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, and 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.
Research Focus: Week of August 14, 2023
In this issue: HyWay enables hybrid mingling; Auto-Tables transforms non-relational tables into standard relational forms; training dense retrievers to identify high-quality in-context examples for LLM; improving pronunciation assessment in CAPT.
Research Focus: Week of May 8, 2023
In this issue: Microsoft researchers win four more awards; AutoRXN automates calculations of molecular systems; LLM accelerator losslessly improves the efficiency of autoregressive decoding; a frequency domain approach to predict power system transients.
Research Focus: Week of November 7, 2022
Welcome to Research Focus, a new series of blog posts that highlights notable publications, events, code/datasets, new hires and other milestones from across the research community at Microsoft. Barun Patra, Saksham Singhal, Shaohan Huang, Zewen Chi, Li Dong, Furu Wei,…