题    目：Robust and Scalable Bayesian Dimension Reduction

主讲人：Wang Liangliang 副教授

单    位：Simon Fraser University

时    间：2025年5月6日 15:00

地    点：金明综合楼7101

摘    要：Dimension reduction plays a central role in modern statistical analysis, particularly when dealing with high-dimensional or complex structured data. In this talk, I present recent advances in Bayesian dimension reduction methods that improve robustness and computational efficiency. First, I introduce a robust Bayesian functional principal component analysis (RB-FPCA) framework for analyzing functional data, even when observations are sparse or contaminated by outliers. By leveraging skew-elliptical distributions, RB-FPCA accurately captures principal modes of variation and offers improved estimation of the covariance structure. Next, I present a generalized Bayesian multidimensional scaling (GBMDS) framework that enables flexible embeddings of multivariate data under non-Gaussian error structures and user-defined dissimilarity measures. A key innovation in both models is the use of annealed Sequential Monte Carlo (ASMC) for posterior inference, which overcomes limitations of traditional MCMC by enhancing posterior exploration and providing accurate marginal likelihood estimates for model comparison. Applications to environmental, biological, and synthetic datasets demonstrate that these methods offer superior performance in robustness, scalability, and uncertainty quantification compared to existing approaches.

简    介：王亮亮教授，加拿大英属哥伦比亚大学（University of British Columbia）统计学博士，西蒙菲莎大学（Simon Fraser University）副教授。王亮亮教授长期从事贝叶斯统计、机器学习以及生物统计的科学研究；担任多个国际会议的组委会成员；在国际和国内著名的统计和机器学习杂志上发表学术论文六十余篇；主持和参与过多个国家基金项目，如加拿大统计科学研究所（CANSSI）项目、加拿大自然科学与工程研究理事会（NSERC）项目等。