Adversarial Variational Optimization of Non-Differentiable Simulators

By Gilles Louppe, Kyle Cranmer. 2017

72 citations recorded by the source corpus.

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.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Statistical and Nonlinear Physics, Generative Models, Probabilistic & Bayesian Methods

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Statistical and Nonlinear Physics

In Generative Models

In Probabilistic & Bayesian Methods

Citations in the corpus

Recurrent machines for likelihood-free inference (2018, methodology): “We also tested our model on a simpliﬁed simulator from particle physics, called Weinberg and introduced as a likelihood-free inference benchmark in (Louppe and Cranmer, 2017).”
Automatic Posterior Transformation for Likelihood-Free Inference (2019, background): “Some LFI meth-ods instead aim to infer a point or local estimate of the parameters (Pesah et al., 2018; Louppe & Cranmer, 2017; McCarthy et al., 2017).”
Likelihood-free MCMC with Amortized Approximate Ratio Estimators (2019, methodology): “Recent work in (Louppe et al., 2017) proposes a novel form of variational inference by introducing an adversary in combination with REINFORCE-estimates (Williams, 1992; Sutton et al.”
Synthetic Data for Deep Learning (2019, methodology): “[379], who provide an inference algorithm based on variational approximations for fitting the parameters of a domain-specific non-differentiable simulator.”
Meta-Sim2: Unsupervised Learning of Scene Structure for Synthetic Data Generation (2020, unknown): “[44] learn to optimize non-differentiable”
Meta-Sim: Learning to Generate Synthetic Datasets (2019, methodology): “[31] also attempt to optimize non-differentiable simulators using a variational upperbound of a GAN-like objective to produce samples representative of a target distribution.”
Mining gold from implicit models to improve likelihood-free inference (2018, methodology): “This poses a difﬁculty, making methods based on ∇ z x [42] and ∇ θ x inapplicable, which previously motivated techniques such as Adversarial Variational Optimization [32].”
Likelihood-free MCMC with Approximate Likelihood Ratios (2019, background): “Recent work in [45] proposes a novel form of variational inference by introducing an adversary in combination with REINFORCE -estimates [46, 47] to optimize a parameterized prior.”