Bayesian Statistics: Modeling & Variational Autoencoders

Talk 1

Title: Bayesian Modeling applied in Industrial Testers

Speaker: Maksim Greiner

Abstract: Bayesian statistics combined with mathematical modeling is a powerful tool for drawing conclusions from limited data or for estimating otherwise unobtainable quantities. In this talk, I will present the concepts of Bayesian modeling using easily understandable examples and introduce Python libraries used in Bayesian modeling. After these basics, I will walk you through a real-life example in which Bayesian modeling is applied to enhance an industrial end-of-line tester.

Bio: Maksim Greiner studied physics at the Ludwig Maximilian University in Munich and at the Université Paris Diderot. He earned his PhD at the Max Planck Institute for Astrophysics, where his research focused on information theory and Bayesian statistics. Since 2016 he develops Machine Learning solutions for industrial applications. Maksim is also managing director and CTO of Erium GmbH.

Talk 2

Title: Probabilistic Variational Autoencoders, a Bayesian perspective

Speaker: Philipp Frank

Abstract: Variational Autoencoders (VAEs) are derived by combining Bayesian reasoning with variational inference. In this talk, I will briefly outline this probabilistic perspective and show how its reconstruction capacity as well as its sampling quality can be improved using the Fisher Information Metric and Normalizing Flows, respectively. I will also display an application of a VAE variant to an astrophysical component separation problem.

Bio: Philipp Frank is an Astrophysicist with an interest in probability theory and machine learning. He studied physics at the Ludwig Maximilian University in Munich and earned his PhD at the Max Planck Institute for Astrophysics where he worked on solving and approximating Bayesian inverse problems in the context of astronomy.