nnScaler：重塑深度学习并行策略，大幅提升训练效率
作者：苗又山 编者按：深度学习技术已经在图像识别、语音识别、自然语言处理、搜索推荐等多个领域不断展现出巨大的应用价值。然而，随着模型规模的不断增大，深度学习模型的训练变得耗时且昂贵，设计最优的并行策略组合以提高其在多设备上的执行性能是目前该领域的一大挑战。在此背景下，微软亚洲研究院提出 nnScaler 技术，通过一套并行化原语和策略限定搜索的方法来寻求最佳的并行策略组合。这一尝试为寻求深度学习并...
Microsoft at ICML 2024: Innovations in machine learning
The competitive dynamics of AI agents and a method for learning and applying temporal action abstractions represent just some of Microsoft’s contributions to ICML 2024.
In the news | Monocle
Meet the Writers, Juan M Lavista Ferres on AI for Good
In Microsoft’s pioneering AI For Good Lab, data scientists and researchers’ use of artificial intelligence (AI) is helping to tackle disinformation, predict wildfires, track whales and even detect leprosy in vulnerable populations. But what are the dangers in AI being…
完全激活稀疏大模型，Q-Sparse突破LLM推理效能
作者：量子位 编者按：激活稀疏性是解决大语言模型（LLMs）在推理阶段出现的计算成本高、内存占用大等问题的有效方法，可以有效减少激活张量中激活元素的数量。然而该方法无法实现 LLMs 激活的完全稀疏性，从而限制了推理阶段的效率提升。 对此，微软亚洲研究院提出 Q-Sparse 实现了 LLMs 激活的完全稀疏性。该研究不仅揭示了包括推理优化规模法则（inference-optimal scalin...
Awards | MistyWest
Tusher Chakraborty named a Misties Top 20 Winner
Tusher Chakraborty was recognized for the contributions to enabling data-driven farming, influencing FCC to adopt regulations on IoT in TV White Spaces, and pioneering research in satellite-based IoT communications. The Misties Awards are for top 20 individual leaders who are…
Abstracts: July 18, 2024
| Gretchen Huizinga and Arindam Mitra
Senior Researcher Arindam Mitra introduces AgentInstruct. Using raw data sources, the automated multi-agent framework can create diverse, high-quality synthetic data at scale for the post-training of small and large language models.
In the news | Microsoft YouTube
Addressing food malnutrition with Microsoft AI | Microsoft Nonprofits X Amref Health Africa
Amref Health Africa partners with Microsoft AI for Good and Kenya's Ministry of Health to identify, address, and combat food malnutrition among children in Africa with the help of Microsoft AI technology.
Research Focus: Week of July 15, 2024
Advancing time series analysis with multi-granularity guided diffusion model; An algorithm-system co-design for fast, scalable MoE inference; What makes a search metric successful in large-scale settings; learning to solve PDEs without simulated data.
数据驱动模型提升电动汽车电池退化预测准确率
编者按：在全球向新能源转型的浪潮下，电动汽车的普及率正不断提升。然而，在享受电动汽车便利性的同时，你是否也在担忧电池的续航问题？电池的性能和寿命以及相应的监测、维护、回收等相关问题也同样困扰着电动汽车生产企业。而且如果废旧电池在回收、拆解和再利用的过程中处理不当，可能会对环境造成二次污染。 为了更有效地实现动力电池性能和寿命的精准预测，以及相应的对废旧动力电池的绿色回收和高效重复利用，微软亚洲研究...