Seminar| Institute of Mathematical Sciences
Time: Saturday, November 22th, 2025,10:00-11:00
Location: RS408, IMS
Speaker: Tianyu Xie, Peking University
Abstract: Reconstructing the evolutionary relationships among species, i.e., phylogenetic inference, hasbeen one of the central problems in computational biology. With a phylogenetic prior and evolutionary substitution likelihood model, this problem is formulated as Bayesian phylogenetic inference of the posterior distribution over phylogenetic trees. Previous approaches often leverages Monte-Carlo type approaches,e.g., MCMC, which can suffer from slow convergence and local mode trapping in practice. In this talk, we discuss how to integrate variational inference with deep learning as a powerful solution to Bayesian phylogenetic inference. Specifically, we develop an autoregressive probabilisitc model called ARTree and its accelerated version to modeling the tree topologies, and a semi-implicit hierarchical construction for the branch lengths. We also introduce representation learning for phylogenetic trees to provide high-resolution representations that are ready-to-use for downstream tasks. These deep learning approaches to Bayesian phylogenetic inference achieve state-of-the-art inference accuracies and inspire broader follow-up innovations.
【个人介绍】Tianyu Xie is currently a final-year PhD student at the School of Mathematical Sciences, Peking University, supervised by Prof. Cheng Zhang. Previously, he received B.S. in statistics and B.A. in economics from Peking University in 2021.His research interest is Bayesian inference as well as its applications in computational biology and generative models.His papers were published at top-tier conferences including NeurIPS, ICML, and ICLR, and SCI journals including Statistics & Computing.
He was awarded the National Scholarship, PKU Principal Scholarship, and PKU May-fourth Scholarship during his study in Peking University.His homepage is https://tyuxie.github.io/.