DeLTA seminar by Yevgeny Seldin
Follow this link to participate on Zoom
Speaker
Yevgeny Seldin, DIKU
Title
PAC-Bayes Tutorial
Abstract
The tutorial will take you on a journey from basic reflection on the meaning of generalization in learning to modern tools for incorporation of prior knowledge within frequentist learning frameworks and sequential updates of prior knowledge within frequentist learning frameworks. I will also relate and contrast frequentist and Bayesian interpretation of priors and posteriors.
The tutorial is a selection and compression of topics that I teach in MLA and MLB and assumes no prior knowledge beyond what is needed to enrol for MLA. It includes both high-level reflections and some technical derivations. Concrete topics that will be covered include:
- Occam’s razor (“the little brother of PAC-Bayes”)
- PAC-Bayesian analysis
- Recursive PAC-Bayes (sequential updates of prior knowledge within frequentist learning frameworks)
- PAC-Bayesian analysis of weighted majority votes (if we get time for it)
Further reading: Yevgeny Seldin. Machine Learning. The science of selection under uncertainty. https://arxiv.org/pdf/2509.21547, 2025.
Join the DeLTA community
You can subscribe to the DeLTA Seminar mailing list by sending an empty email to delta-seminar-join@list.ku.dk.
DeLTA online calendar
DeLTA Lab page