Omar Rivasplata

On the asymptotic behaviour of $\alpha$-divergence generalized posteriors

We study the asymptotic behaviour of the solutions to the unconstrained generalized variational inference problem regularized by an $\alpha$-divergence. Under mild conditions, the minimizer is explicit up to a scalar normalizing constant, and we show that it admits a rescaling under which it converges in $L^1$ to a fix...