Prof. Dino Sejdinovic

(University of Adelaide)
hosted by Machine Learning Group of Prof. Dr. Marius Kloft

"Generalised Variational Inference Meets Bayesian Deep Learning"

Dino Sejdinovic will describe the framework of generalised variational inference and two ways it can be connected to Bayesian deep learning. When considering generalised variational inference in infinite-dimensional function spaces, we leverage the Wasserstein distance between Gaussian measures on the Hilbert space of square-integrable functions. The resulting method avoids pathologies arising in standard variational function space inference and uses deep neural networks in the variational parameterisation, combining their superior predictive performance with the principled uncertainty quantification analogous to that of Gaussian processes. Additionally, we can also study generalised variational objectives through the lens of Wasserstein Gradient Flows. The result is a unified theory of various seemingly disconnected approaches that are commonly used for uncertainty quantification in deep learning – including deep ensemble. This offers a fresh perspective which also allows the derivation of new ensembling schemes with convergence guarantees.

Bio: Dino Sejdinovic is a Professor of Statistical Machine Learning in the School of Computer and Mathematical Sciences at The University of Adelaide. He is affiliated with the Australian Institute for Machine Learning (AIML) and serves on the steering committee of the Adelaide Data Science Centre (ADSC). Before moving to Adelaide in October 2022, he was an Associate Professor at the Department of Statistics, University of Oxford, a Fellow of Mansfield College, and a Faculty Fellow of the Alan Turing Institute. His research spans a wide variety of topics at the interface between machine learning and statistical methodology, including large-scale nonparametric and kernel methods, robust and trustworthy machine learning, causal inference, and uncertainty quantification.

Time:	Friday, 15.03.2024, 14:00
Place:	room 48/453

Termin als Datei downloaden und in den Kalender importieren.