NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:7847
Title:Variational Bayesian Optimal Experimental Design

This paper provides a novel, sound, and seemingly practical method for Bayesian optimal experimental design. The basic idea is to create upper and lower bounds on the information gain analogously to existing ideas to upper and lower-bound the marginal likelihood in traditional VI. This is an important problem, and the reviewers were unanimous that the ideas had merit.