Learning Model Reparametrizations: Implicit Variational Inference by Fitting MCMC distributions
This method addresses approximate inference for machine learning practitioners by offering a more flexible and easily applicable approach, though it appears incremental as it builds on existing variational and MCMC techniques.
The paper tackles the problem of approximate inference by introducing a new algorithm that combines reparametrization, MCMC, and variational methods to create a flexible implicit variational distribution, avoiding log density ratio computations and demonstrating applicability to banana-shaped distributions and variational autoencoders.
We introduce a new algorithm for approximate inference that combines reparametrization, Markov chain Monte Carlo and variational methods. We construct a very flexible implicit variational distribution synthesized by an arbitrary Markov chain Monte Carlo operation and a deterministic transformation that can be optimized using the reparametrization trick. Unlike current methods for implicit variational inference, our method avoids the computation of log density ratios and therefore it is easily applicable to arbitrary continuous and differentiable models. We demonstrate the proposed algorithm for fitting banana-shaped distributions and for training variational autoencoders.