Automatic Posterior Transformation for Likelihood-Free Inference
This addresses the challenge of likelihood-free inference for researchers and practitioners using complex simulators, offering incremental improvements in flexibility and efficiency over existing methods.
The paper tackles the problem of performing Bayesian inference on stochastic simulators with intractable likelihoods by introducing automatic posterior transformation (APT), a sequential neural posterior estimation method that allows for arbitrary, dynamically updated proposals and is compatible with flow-based density estimators, resulting in a more flexible, scalable, and efficient approach that can handle high-dimensional data like time series and images.
How can one perform Bayesian inference on stochastic simulators with intractable likelihoods? A recent approach is to learn the posterior from adaptively proposed simulations using neural network-based conditional density estimators. However, existing methods are limited to a narrow range of proposal distributions or require importance weighting that can limit performance in practice. Here we present automatic posterior transformation (APT), a new sequential neural posterior estimation method for simulation-based inference. APT can modify the posterior estimate using arbitrary, dynamically updated proposals, and is compatible with powerful flow-based density estimators. It is more flexible, scalable and efficient than previous simulation-based inference techniques. APT can operate directly on high-dimensional time series and image data, opening up new applications for likelihood-free inference.