A0271
Title: An extended empirical saddlepoint approximation for intractable likelihoods
Authors: Matteo Fasiolo - University of Bristol (United Kingdom) [presenting]
Abstract: The use of simulation-based inferential approaches is widespread in computational biology and ecology. We will focus on one such approach: Synthetic Likelihood (SL). This method reduces the observed and simulated data into a set of features or summary statistics, and quantifies the discrepancy between them through a synthetic likelihood function. While requiring less tuning than some alternative approaches (such as approximate Bayesian computation), SL has the drawback of relying on the summary statistics being approximately normally distributed. We will describe how this shortcoming can be addressed by adopting a more flexible density estimator: the Extended Empirical Saddlepoint Approximation (ESA). This new density estimator is able to capture large departures from normality, while being scalable to high dimensions. This can lead to more accurate parameter estimates, relative to the Gaussian alternative.
