Michael Kagan: Generative Model Based Design Optimization and Unfolding

A host of scientific disciplines encapsulate acquired knowledge into high fidelity simulators that subsequently allow the generation of plausible experimental outcomes. The evaluation of a sample’s likelihood, however, is often infeasible due to the complexity of the simulated processes. As such, asking inferential questions with simulators can be a resource intensive and arduous task, especially in high dimensional settings. This talk will discuss new methods in simulation-based inference which combine the power of deep generative models with simulators to aid in approaching key scientific challenges such as design optimization and unfolding / measurement de-corruption. The application of these methods within the domain of High Energy Physics will be shown.