This session will concentrate on recent research developments in the area of likelihood-free Bayesian inference.
The term “likelihood-free” encompasses a wide range of situations, including when (a) the likelihood exists but is easier to sample from rather than to evaluate numerically, or (b) when the parameters of interest are linked to data through a loss function (not necessarily the likelihood function).
We will revisit tools for generalized Bayesian inference including (1) bootstrap-style samplers for loss-driven Bayes, (2) variational Bayes, and (3) ABC posterior sampling techniques which do not rely on the likelihoods to be tractable. The session will concentrate on some of the most recent developments in theory, methods as well as computation.
Session Chair: Veronika Rockova