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...
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.
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...
In the news | Microsoft on the Issues
Accelerating AI opportunity and climate solutions in Africa
Microsoft firmly believes that AI has tremendous potential to accelerate solutions to the various challenges societies are facing, including the climate crisis.
In the news | GeekWire
80 years after D-Day, Microsoft uses AI to enhance archival materials and keep memories alive
Microsoft is using artificial intelligence to help preserve the history of D-Day, 80 years after allied forces landed on the beaches of Normandy, France, and turned the tide of World War II.
In the news | Unlocked
The thread of memory
June 6, 1944, the beginning of the end: When Allied troops landed on the shores of Normandy 80 years ago, they made history with their courage and sacrifice, which contributed to the liberation of France from Nazi Germany. In The Thread…
Data-driven model improves accuracy in predicting EV battery degradation
Microsoft Research and Nissan Motor Corporation have collaborated to develop a machine learning model that improves the accuracy of predicting EV battery degradation by 80%. Learn how this collaboration supports long-term sustainability goals.
RUBICON: Evaluating conversations between humans and AI systems
| Param Biyani, Yasharth Bajpai, Arjun Radhakrishna, Gustavo Soares, and Sumit Gulwani
RUBICON evaluates AI-driven conversations and improves their quality by learning detailed domain-specific rubrics from minimal data. It gathers insights on AI assistant performance while maintaining user privacy and data security.