Adversarial Variational Optimization of Non-Differentiable Simulators

TL;DR

This paper introduces Adversarial Variational Optimization (AVO), a likelihood-free inference method that enables parameter estimation for complex, non-differentiable simulators by minimizing divergence between simulated and observed data distributions.

Contribution

The paper proposes AVO, a novel likelihood-free inference algorithm that combines adversarial training with variational optimization for non-differentiable simulators.

Findings

AVO effectively estimates parameters for both discrete and continuous data simulators.

The method minimizes the Jensen-Shannon divergence between simulated and real data distributions.

AVO outperforms existing likelihood-free inference techniques in various scenarios.

Abstract

Complex computer simulators are increasingly used across fields of science as generative models tying parameters of an underlying theory to experimental observations. Inference in this setup is often difficult, as simulators rarely admit a tractable density or likelihood function. We introduce Adversarial Variational Optimization (AVO), a likelihood-free inference algorithm for fitting a non-differentiable generative model incorporating ideas from generative adversarial networks, variational optimization and empirical Bayes. We adapt the training procedure of generative adversarial networks by replacing the differentiable generative network with a domain-specific simulator. We solve the resulting non-differentiable minimax problem by minimizing variational upper bounds of the two adversarial objectives. Effectively, the procedure results in learning a proposal distribution over simulator parameters, such that the JS divergence between the marginal distribution of the synthetic data and the empirical distribution of observed data is minimized. We evaluate and compare the method with simulators producing both discrete and continuous data.