AIDRC Seminar Series - Jakub Tomczak

Title:

Deep Generative Modeling is a key to unlocking AI potential

Abstract:

Deep learning achieves state-of-the-art results in tasks like image or audio classification. However, deep-learning-based predictors could be easily fooled by out-of-distribution data or noisy examples. During this talk, we will discuss a possible solution, namely, deep generative modeling that is a combination of deep learning and probabilistic modeling. We will start with a motivation following from information theory that naturally places learning a joint distribution as a crucial problem of learning in AI systems. Next, we will outline various approaches to model a distribution over objects (e.g., images). More specifically, we will mainly focus on latent-variable models (e.g., Variational Auto-Encoders, Diffusion-based Deep Generative Models, Flow-based models). In the conclusion, we will indicate possible future research directions.

Bio:

Jakub M. Tomczak is an assistant professor of Artificial Intelligence in the Computational Intelligence group (led by Prof. A.E. Eiben) at Vrije Universiteit Amsterdam. Before joining Vrije Universiteit Amsterdam, he was a deep learning researcher (Engineer, Staff) in Qualcomm AI Research in Amsterdam, a Marie Sklodowska-Curie individual fellow in Max Welling's group at the University of Amsterdam, and an assistant professor and postdoc at the Wroclaw University of Technology. His main research interests include deep generative modeling, deep learning, and Bayesian inference, with applications to image processing, robotics, and life sciences. He is the author of the book entitled "Deep Generative Modeling".