This research presents a way to improve the performance of Amortized Bayesian Inference (ABI) by using universal symmetries in the probabilistic joint model of parameters and data. The authors propose inverting Bayes’ theorem to estimate the marginal likelihood based on approximate representations of the joint model. The technique reduces approximation errors and accelerates the learning dynamics of conditional neural density estimators. The proposed method is applied to both a bimodal toy problem with an explicit likelihood and a realistic model with an implicit likelihood.
Publication date: 6 Oct 2023
Project Page: https://arxiv.org/pdf/2310.04395.pdf
Paper: https://arxiv.org/pdf/2310.04395
