In the news | OpenAI
Democratic Inputs to AI
Our nonprofit organization, OpenAI, Inc., is launching a program to award ten $100,000 grants to fund experiments in setting up a democratic process for deciding what rules AI systems should follow, within the bounds defined by the law.
Microsoft at ICML 2023: Discoveries and advancements in machine learning
| Kate Forster
Microsoft Research is proud to be a sponsor of ICML 2023! From audio classification to privacy estimation and more, explore conference highlights in our latest blog post.
如何评测一个大语言模型？
编者按：大型语言模型（Large language models, LLMs）因其在学术界和工业界展现出前所未有的性能而备受青睐。随着 LLMs 在研究和实际应用中被广泛使用，对其进行有效评测变得愈发重要。近期已有多篇论文围绕大模型的评测进行研究，但尚未有文章对评测的方法、数据、挑战等进行完整的梳理。日前，微软亚洲研究院的研究员们参与完成了介绍大模型评测领域的第一篇综述文章《A Survey on...
Collaborators: Gaming AI with Haiyan Zhang
| Gretchen Huizinga and Haiyan Zhang
For over a decade, Xbox has been leveraging AI to elevate gaming. Haiyan Zhang, GM of Gaming AI, explores the collaborations behind the work and the potential for generative AI to support better experiences for both players and game creators.
MABIM：多智能体强化学习算法的“炼丹炉”
编者按：现实世界中，许多问题和任务都是由多个参与者交互进行的，所以要想使用人工智能技术解决真实世界的问题，就需要更好地模拟这种复杂的环境，而这正是多智能体强化学习（MARL）的强项。早在2020年，微软亚洲研究院基于多智能体强化学习，推出了面向多行业横截面上的多智能体资源调度平台 MARO。 随着研究的深入，研究员们发现互动式的学习环境和测试平台对多智能体强化学习的发展至关重要。为此，近期微软亚洲...
Research Focus: Week of July 17, 2023
RetroRanker mitigates frequency bias in predictions of retrosynthesis models; new algorithm beats PPO on language tasks; DER dataset aids grid planning; improved PPML balances privacy & accuracy across shared data; ASL Citizen boosts sign language modeling.
作者：量子位 作为全新的神经网络架构，RetNet 同时实现了良好的扩展结果、并行训练、低成本部署和高效推理。这些特性将使 RetNet 有可能成为继 Transformer 之后大语言模型基础网络架构的有力继承者。 ——韦福如，微软亚洲研究院全球研究合伙人 以下内容经授权转载自公众号“量子位”，原文标题《Transformer后继有模！MSRA提出全新大模型基础架构：推理速度8倍提升，内存占用减...
Awards | USENIX OSDI
Daniel Berger receives OSDI 2023 Best Paper Award
Daniel Berger's paper, Enso: A Streaming Interface for NIC-Application Communication, was selected as a best paper by the OSDI'23 organizers for its quality and potential impact. Enso was also awarded for its open source artifacts.
Awards | ACM ASPLOS
Daniel Berger and Pantea Zardoshti receive ASPLOS 2023 Distinguished Paper Award
Our paper, Pond: CXL-Based Memory Pooling Systems for Cloud Platforms, was selected as a distinguished paper by the ASPLOS'23 organizers. Congrats to Daniel S. Berger and Pantea Zardoshti .