Mathematics Colloquia and Seminars

Return to Colloquia & Seminar listing

Variational approximations provide an attractive computational alternative to MCMC-based strategies for approximating the posterior distribution in Bayesian inference. The Naive Mean-Field (NMF) approximation is the simplest version of this strategy—this approach approximates the posterior in KL divergence by a product distribution. There has been considerable progress recently in understanding the accuracy of NMF under structural constraints such as sparsity, but not much is known in the absence of such constraints. Moreover, in some high-dimensional settings, the NMF is expected to be grossly inaccurate, and advanced mean-field techniques (e.g. Bethe approximation) are expected to provide accurate approximations. We will present some recent work in understanding this duality in the context of high-dimensional regression. This is based on joint work with Sumit Mukherjee (Columbia) and Jiaze Qiu (Harvard University).