Event Date
Event Date
Location
Remotely Presented via Zoom
Speaker: Anirban Bhattacharya, Associate Professor, Dept of Statistics, Texas A&M University
Title: "Statistical and algorithmic guarantees for variational Bayes"
Abstract: Variational Bayesian algorithms are popularly used in machine learning to approximate an intractable posterior distribution. In this talk, I shall discuss some recent efforts to establish frequentist properties of point estimates constructed from mean-field variational inference (VI), a commonly used class of VI algorithms. To that end, we develop a variational inequality to connect the Bayes risk under the variational approximation to the VI objective function. As an important upshot, we find that maximizing the evidence lower bound in VI has the effect of minimizing the Bayes risk within the variational density family. The variational inequality can be used to conclude that point estimates constructed from standard VI procedures converge at a near-optimal rate in a wide range of problems. In the second half of the talk, I shall discuss some ongoing work where we study the dynamics and convergence of coordinate descent algorithms for mean-field VI.
About the speaker: Anirban Bhattacharya is an Associate Professor in the Department of Statistics at Texas A&M University. His research interests lie at the intersection of theory, methods, and algorithms for large-scale modern Bayesian inference problems.
Seminar Date/Time: Thursday April 15, 4:10pm
This seminar will be delivered remotely via Zoom. To access the Zoom meeting for this seminar, please contact the instructor Professor Jairo Fùquene Patiño or Pete Scully (pscully@ucdavis.edu) for the meeting ID and password, stating your affiliation.