Doubly Semi-Implicit Variational Inference
arXiv:1810.02789
Abstract
We extend the existing framework of semi-implicit variational inference (SIVI) and introduce doubly semi-implicit variational inference (DSIVI), a way to perform variational inference and learning when both the approximate posterior and the prior distribution are semi-implicit. In other words, DSIVI performs inference in models where the prior and the posterior can be expressed as an intractable infinite mixture of some analytic density with a highly flexible implicit mixing distribution. We provide a sandwich bound on the evidence lower bound (ELBO) objective that can be made arbitrarily tight. Unlike discriminator-based and kernel-based approaches to implicit variational inference, DSIVI optimizes a proper lower bound on ELBO that is asymptotically exact. We evaluate DSIVI on a set of problems that benefit from implicit priors. In particular, we show that DSIVI gives rise to a simple modification of VampPrior, the current state-of-the-art prior for variational autoencoders, which improves its performance.
References in corpus (18)
- Adam: A Method for Stochastic Optimization
- Estimating divergence functionals and the likelihood ratio by convex risk minimization
- Markov Chain Monte Carlo and Variational Inference: Bridging the Gap
- Bayesian Compression for Deep Learning
- Auxiliary Deep Generative Models
- Learning in Implicit Generative Models
- Hierarchical Implicit Models and Likelihood-Free Variational Inference
- Variational Inference using Implicit Distributions
- VAE with a VampPrior
- Operator Variational Inference
- Gradient Estimators for Implicit Models
- Structured Stochastic Variational Inference
- Copula variational inference
- Semi-Implicit Variational Inference
- Overpruning in Variational Bayesian Neural Networks
- Variational Gaussian Copula Inference
- The Deep Weight Prior
- Kernel Implicit Variational Inference