北京雁栖湖应用数学研究院

孙明明 , 王雅晴

孙明明

2024年10月17日 14:00 至 16:00

A3-1-301

Zoom 230 432 7880 (BIMSA)

Inspired by the nervous system of worms, MIT researchers proposed the model of Liquid Neural Networks (LNN) in 2006. In the years that followed, they developed a theoretical framework for LNNs, proving their superior theoretical properties and achieving successful applications in time series modeling, differential equation solving, and bioinformatics. Recently, they introduced Liquid Foundation Models (LFMs), a general AI model based on LNNs. LFMs are large neural networks built upon the theories of dynamical systems, signal processing, and numerical linear algebra. This unique combination allows LFMs to leverage decades of theoretical advances in these fields. The development from LNNs to LFMs illustrates a path from natural phenomena to mathematical models and eventually to AI products. This journey might offer us significant insights.

孙明明，北京雁栖湖湖应用数学研究院研究员。研究方向为自然语言处理、认知计算与世界模型。曾任百度研究院主任研究员，百度认知计算实验室负责人。曾负责百度的搜索推荐系统和百度大语言模型文心一言的研发工作，并领导了认知自然语言处理和自动驾驶技术的创新项目。获 2023 世界互联网大会领先科技奖，百度骄傲特别奖，百度最高奖等奖项。