学术讲座

时间：

2023年11月29日（周三）14:20-16:30

地点：立德楼801智慧教室

邀请人：

黄文炳 中国人民大学高瓴人工智能学院准聘副教授

报告一

主讲人姓名：荣钰腾讯 AI Lab专家研究员

主讲人简介：

Yu Rong is a principal researcher of AI for Science Center in Tencent AI Lab. He received this B.E. degree from Sun Yat-sen University, Guangzhou, China in 2012 and the Ph.D. degree from The Chinese University of Hong Kong in 2016. He joined Tencent AI Lab in June 2017. In Tencent AI Lab, he is working on developing the novel and efficient graph learning algorithms to tackle real-world challenges, particularly in the realm of social networks and AI for Science. He has published over 50 papers on machine learning and data mining top conferences and has got the champion of NeurIPS 2022 Open Catalyst Challenge as the team leader. In addition to research activities, he has successfully organized several academic events such as Deep Graph Learning tutorials at KDD 2020 and TheWebConf 2020.

报告题目：

Geometric Graph Learning for Protein Mutation Prediction and Dynamical Simulation

报告摘要：

Geometric machine learning aims to generalize neural networks or other machine learning models to non-Euclidean domains such as graphs or manifolds. Recently, Geometric Graph Neural Networks (GGNNs) have become powerful tools for modeling graph data with geometric information and have demonstrated their potential to advance scientific fields, such as particle motion modeling, molecular dynamics, protein folding, etc. The pivotal challenge in modeling geometric information is how to incorporate principles from geometry, topology, and other mathematical fields, such as geometric symmetry and equivariance, to improve the generalization ability of current GNN models, which can understand complex data form scientific domain. In this talk, I will introduce our explorations of employing GGNNs for tackling issues relating to Protein Mutation Prediction and long-term physical simulations, and discuss the future directions of this field.

报告二

主讲人姓名：

李佳 香港科技大学（广州）副研究员

主讲人简介：

李佳，香港科技大学(广州) 数据科学与分析学域 助理教授，港科大广州-创邻图数据实验室联合主任。李佳博士毕业于香港中文大学。他在工业界有多年的数据挖掘工作经历，曾供职于Google和腾讯(微信支付风控)，主持了腾讯第一代大数据风控系统建设，获得2015年度腾讯技术突破一等奖。其研究目前主要为图数据的大模型，异常检测，图神经网络以及基于图数据的药物发现和医疗健康。他以第一作者或者通讯作者在人工智能与数据挖掘领域顶级会议与期刊发表二十多篇CCF-A论文，如Nature Communications, NeurIPS, SIGKDD, ICML, TPAMI等。他的工作获得2023年数据挖掘顶会SIGKDD Best Research Paper Award，是中国大陆首次获得该荣誉。

报告题目：

图数据预训练与提示学习

报告摘要：

大型语言模型的最新进展催生了 OpenAI 的 ChatGPT 等高效模型，这些模型的发展主要集中在文本数据的预训练范式和下游提示学习当中。本演讲将涵盖图数据的预训练和提示学习的一些最新进展。首先会介绍图预训练范式中基于对比学习和基于重建范式的路线之争。针对基于重建范式的图数据预训练，重点介绍基于图反卷积的解码模型WGDN。针对图数据提示学习，重点介绍基于子图模式的提示学习，并就如何提升图神经网络多任务学习做重点展开。最后，我们将展望图数据大模型未来发展，并讨论它带来的机遇和挑战。