Tengfei Ma
Affiliation
Stony Brook University
Talk Title
Latent Graph Learning for Time Series, Labels, and Concepts
Abstract
Graph Neural Networks have emerged as a powerful tool across numerous applications. While most graph neural networks assume the existence of known graph structures, explicit graph structures are often unavailable or suboptimal. Latent graph discovery (also known as graph structure learning), addresses this limitation by inferring meaningful unknown relational structures, thus bridging the gap between unstructured data and graph-based learning paradigms. In this talk, I will introduce some of our work on discovering latent graphs in various contexts, such as drug-drug-interaction prediction, time series analysis, and concept bottleneck models.
Bio
Tengfei Ma is an assistant professor in the Department of Biomedical Informatics, Stony Brook University. Before joining SBU, he was a staff research scientist in IBM T. J. Watson Research center, where he led the AI challenge of deep learning on graphs. He got his PhD from the University of Tokyo, MS from Peking University and BE from Tsinghua University, and also worked in IBM Research Tokyo for one year. His recent research was focused on deep graph learning, time series analysis and their application in the biomedical domain. Specifically, on graph learning his work includes scalability (such as FastGCN), generalizability, and combining geometry and topological information with GNNs. He has published over 60 papers in major AI conferences such as NeurIPS, ICLR, ICML, AAAI. He is a recipient of the best paper award in ISWC 2021 research track, and two IBM outstanding research awards.