Title: Sparse variational inference
Authors: Trevor Campbell - University of British Columbia (Canada) [presenting]
Abstract: The purpose is to cover recent work on Bayesian coresets (core of a dataset), a methodology for statistical inference via data compression. Coresets achieve compression by forming a small weighted subset of data that replaces the full dataset during inference, leading to significant computational gains with provably minimal loss in inferential quality. In particular, the talk will present methods for Bayesian coreset construction, from previously-developed subsampling, greedy, and sparse linear regression-based techniques to a novel algorithm based on sparse variational inference (VI). In contrast to past algorithms, sparse VI is fully automated, requiring only the dataset and probabilistic model specification as inputs. Empirical results will be shown which illustrate that despite requiring much less user input than past methods, sparse VI coreset construction provides state-of-the-art data summarization for Bayesian inference.