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.
In the news | AI for Good Insider
AI is not just transforming the digital world, it is increasingly being leveraged to solve some of the most pressing global issues. One such organization spearheading this charge is Microsoft (opens in new tab), through its AI for Good Research Lab. At…
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...
In the news | Microsoft Customer Stories
The oldest digital library and a leading distributor of free eBooks, Project Gutenberg wanted to make its collection more accessible to a broader community of members, including people who have visual impairments and those around the world who might not…
In the news | CBS News
Peter Lee spoke to CBS News chief medical correspondent, Dr. Jon LaPook, about the promise of AI to reducing tasks that are burdening healthcare professionals so they can focus on their patients.
尽管如今的 AI 模型已经具备了理解自然语言的能力,但科研人员并没有停止对模型的不断改善和理论探索。自然语言处理(NLP)领域的技术始终在快速变化和发展当中,酝酿着新的潮流和突破。 NLP 领域的顶级学术会议国际计算语言学年会 (Annual Meeting of the Association for Computational Linguistics,简称 ACL) 是关心 NLP 领域的研究...
7 月 8 日,由青海大学计算机技术与应用系、微软亚洲研究院、CCF 女工委共同举办的青海大学 - 微软 Ada Workshop在青海大学图书馆报告厅举办。活动邀请微软亚洲研究院资深学术合作经理孙丽君女士、中国科学技术大学信息学院教授陈雪锦女士、青海大学计算机技术与应用系教师吴利女士、青海民族大学教授林倩女士等嘉宾,从女性 IT 工作者角度分享心路历程与成功经验。活动由微软亚洲研究院学术合作经理...
| Payod Panda
Because headphones rank among the most popular wearables in the market, we have an exciting opportunity to expand their capabilities through integrating existing sensors with supplementary ones to enable a wide variety of experiences that go beyond traditional audio control.
编者按:编译器在传统计算科学中一直是一个重要的研究课题。在人工智能技术快速发展和广泛应用的今天,人工智能模型需要部署在多样化的计算机硬件架构上。同时,训练和部署大型人工智能模型时又对硬件性能有着更高的要求,有时还需根据硬件定制化代码。这些都对人工智能时代的编译器提出了新的更高的要求。 为了适应迅速发展的人工智能模型和加速硬件的需求,微软亚洲研究院以设计和构建具有高度灵活性、高效性、可扩展的 AI...