Learning Proposals for Probabilistic Programs with Inference Combinators
arXiv:2103.00668
Abstract
We develop operators for construction of proposals in probabilistic programs, which we refer to as inference combinators. Inference combinators define a grammar over importance samplers that compose primitive operations such as application of a transition kernel and importance resampling. Proposals in these samplers can be parameterized using neural networks, which in turn can be trained by optimizing variational objectives. The result is a framework for user-programmable variational methods that are correct by construction and can be tailored to specific models. We demonstrate the flexibility of this framework by implementing advanced variational methods based on amortized Gibbs sampling and annealing.
arXiv commentAccepted to UAI 2021
References in corpus (19)
- Auto-Encoding Variational Bayes
- Church: a language for generative models
- Markov Chain Monte Carlo and Variational Inference: Bridging the Gap
- Edward: A library for probabilistic modeling, inference, and criticism
- Gradient Estimation Using Stochastic Computation Graphs
- Scaling Exact Inference for Discrete Probabilistic Programs
- Automated Variational Inference in Probabilistic Programming
- Auto-Encoding Sequential Monte Carlo
- Denotational validation of higher-order Bayesian inference
- Simple, Distributed, and Accelerated Probabilistic Programming
- Deep Amortized Inference for Probabilistic Programs
- Nested Sequential Monte Carlo Methods
- Approximate Inference with Amortised MCMC
- Meta-Learning MCMC Proposals
- Revisiting Reweighted Wake-Sleep for Models with Stochastic Control Flow
- Improving Explorability in Variational Inference with Annealed Variational Objectives
- Composing inference algorithms as program transformations
- Functional Tensors for Probabilistic Programming
- Nested Variational Inference