【“星河Talk”第20期】纽约大学何河：理解大语言模型中的上下文学习｜活动预告

       10月24日，“上海人工智能实验室学术论坛”将举办第20期“星河Talk”暨第4期“AI安全与对齐”系列学术报告。本期邀请纽约大学助理教授何河，带来主题报告Understanding in-context learning in large language models。

     “星河Talk”不定期邀请全球顶尖学者，分享学术成果、探讨科技前沿。“AI安全与对齐”系列学术报告是上海人工智能实验室AI治理研究中心联合安远AI共同组织，探讨大模型安全与对齐领域的最新进展及关键突破。

“星河Talk”第20期

【活动详情】

主题：Understanding in-context learning in large language models

嘉宾：何河 纽约大学助理教授

主持人：马兴军 复旦大学与上海人工智能实验室双聘研究员、博士生导师

时间：北京时间 10月24日 上午10:00-11:00

参与方式：视频号直播

【讲座简介】

 In-context learning (ICL) is an emerging ability of large language models that allows them to adapt to new tasks without the need for parameter updates, even tasks involving complex reasoning. However, our understanding of how it works and how it develops is still in its early stages. One key question is whether LLMs “retrieve” learned tasks from in-context examples, or if they learn directly from these examples in real-time. A better understanding of these questions will assist us in controlling these models more effectively and aligning them with the intended task.

In this talk, the speaker will share recent work along these lines, with a focus on logical reasoning. Firstly, the speaker will describe a comprehensive evaluation revealing surprises and limitations in LLM's ability to reason in-context through chain-of-thought prompting. Following this, the methods to improve the stability and calibration of ICL will be discussed.

【本期嘉宾】

何河 纽约大学助理教授

Assistant Professor of Computer Science and Center for Data Science at New York University, and a visiting academic at AWS AI. Her current research focuses on improving robustness and truthfulness of NLP models, and building interactive systems that collaborate with humans. Before joining NYU in 2019, she obtained her PhD in 2016 from the University of Maryland, College Park, did a post-doc at Stanford, and spent one year at AWS. She is the recipient of a NeurIPS demonstration award and Larry S. Davis dissertation award. She has won research awards and grants from Google, Amazon, Cisco, Samsung, and Open Philanthropy.

【主持人】

马兴军 复旦大学与上海人工智能实验室双聘研究员、博士生导师

【上海人工智能实验室学术论坛】

“上海人工智能实验室学术论坛”分设“星河Talk”与“星启Talk”两个主题系列活动，将分别邀请全球顶尖教授和杰出青年学者作为嘉宾，线上线下分享学术成果、探讨科技前沿。更多精彩内容，敬请期待。