Mechanistic modeling aims to explain observed phenomena in terms of underlying causes. However, determining which model parameters agree with complex and stochastic data presents a significant challenge. We address this challenge with a machine learning tool which uses deep neural density estimators — trained using model simulations — to carry out Bayesian inference and retrieve the full space of parameters compatible with empirical measurements.
I will explain how our approach can be used to perform parameter estimation in general simulation-based models, and demonstrate its power on several challenging neuroscience problems, from the retrieval of complex input-output functions of biophysically-detailed single neurons to the characterisation of mechanisms of compensation for perturbations in neural circuits.