Dr. Kehinde Olobatuyi

Speaker: Dr. Kehinde Olobatuyi, postdoctoral researcher at the University of Victoria
Title: Hybrid Variational Algorithms for Approximate Hierarchical Bayesian Inference of High-Dimensional Data
Date: Friday,  November 18th
Time: 1:30 PM PST
Location: ASB 10900

Abstract: The Bayesian framework for machine learning allows the incorporation of prior knowledge into the system in a coherent manner which avoids overfitting problems but rather seeks to approximate the exact posterior and provides a principled basis for the selection of a model among alternative models. Unfortunately, the computation required in the Bayesian framework is usually intractable. This work provides a hybrid Variational Bayesian framework which approximates these intractable computations with latent variables by minimizing the Kullback-leibler divergence between the exact posterior and the approximate distribution. The work focuses on the problems of instability and intractability in VB-type algorithms and its inability to handle hierarchical Bayesian models due to problems of rigorous moment matching when the prior and likelihood distributions differ. The hybrid variational algorithm is used to estimate the parameters and hyperparameters of a hierarchical model for inverse problem.