Talk de Pedro Rodrigues

Abstract :

Simulation-based inference (SBI) has the potential to revolutionize experimental science as it opens the door to the inversion of arbitrary complex non-linear computer models, such as those found in physics, biology, or neuroscience (Cranmer et al., 2020). The only requirement is to have access to a mathematical model implemented as a simulator. For instance, when applied to biophysical models and simulators in neuroscience, it could estimate properties of the brain closer to the cellular level, thus closing the gap between the neuroimaging and computational neuroscience communities. Grounded in Bayesian statistics, recent SBI techniques profit from recent advances in deep generative modeling to approximate the posterior distributions over the full simulator parameters. Their intrinsic quantification of uncertainties reveals whether certain parameters are worth (or not) scientific interpretation given some experimental observation. In this talk, I will present a short overview on SBI, showing the fundamental ideas behind its current use by experimentalists and pointing towards some important open research questions.