Graph data prevails in a broad spectrum of application domains from social science, finance, to biological and medical sciences. In recent years, graph neural network (GNN) has emerged as a powerful tool for many graph learning problems such as node/graph classification, graph generation and recommendations. Despite the tremendous success of GNNs, there remain a lot of challenges such as over-squashing and limited expressivity. Many GNN issues are essentially connected to geometry and topology, e.g. curvature is related to over-squashing and counting cycles is related to the expressive power of GNNs. In this talk, I will introduce our recent works about how geometric and topological features can be used in GNNs to make them more powerful. In particular, we used Ricci curvature to better represent local structure information in graph convolution; we used persistent homology to enhance the performance of link prediction; and we explored cycle-based rule learning for knowledge graph reasoning.
Tengfei Ma is a staff research scientist in IBM T. J. Watson Research Center. Prior to moving to the US, he obtained his Ph.D. from the University of Tokyo and joined IBM Research Tokyo in 2015. Before that he got his master’s degree from Peking University and his bachelor degree from Tsinghua University. His research interests include machine learning, natural language processing and computational healthcare. Recently his research is mainly focused on graph neural networks and their applications. He is leading the internal project of deep learning on graphs which got the IBM outstanding research achievement in 2019, and he is also a co-PI for a DARPA project about wound healing. He has published over 50 papers in top AI conferences such as NeurIPS, ICLR, ICML, AAAI, EMNLP, and he is a recipient of the best paper award in ISWC 2021 research track. More details can be found in his homepage https://sites.google.com/site/matf0123.
YOU ONLY NEED TO REGISTER ONCE TO ATTEND THE ENTIRE SERIES – We will send you email announcements with details of the upcoming speakers.
Register in advance for this webinar: https://usc.zoom.us/webinar/register/WN__0VhakI6Q6i3JsasdmNWcA
After registering, you will receive an email confirmation containing information about joining the Zoom webinar.
The speaker approved to be recorded for this AI Seminar talk, it will be posted on our USC/ISI YouTube page within 1-2 business days: https://www.youtube.com/user/USCISI.
Host: Muhao Chen, POC: Alma Nava