Principled Amortized Bayesian Inference with Deep Learning

This workshop will provide an introduction to deep learning methods and architectures for efficient Bayesian inference with complex models. It will include a self-contained theoretical part and a practical part, focusing on the topics of posterior estimation, model comparison, likelihood estimation, and model misspecification. In the theoretical part, participants will learn about neural density estimation with normalizing flows, simulation-based optimization, embedding networks, sequential and amortized inference, as well as the rationale of principled Bayesian workflows. In the practical part, participants will apply existing software packages for neural Bayesian inference (e.g., BayesFlow, SBI) to build their own amortized Bayesian workflows. Participants are highly encouraged to "bring" their own models and ideas to the workshop.