![[Search]](/static/images/search.gif){kind=small}
![[A-Z Index]](/static/images/az.gif){kind=small}
![[Contact]](/static/images/contact.gif){kind=small}
![[University of Cambridge]](/static/images/identifier2.gif){kind=small}
![[Talks.cam]](/static/images/talkslogosmall.gif)
Dr. Yanxiao Liu, Imperial College London 
Wednesday 13 May 2026, 14:00-15:00
A Unified Framework for Change of Measure Inequalities: Applications to Generalization, Memorization and Privacy
- đ¤ Speaker: Dr. Yanxiao Liu, Imperial College London đ Website
- đ Date & Time: Wednesday 13 May 2026, 14:00 - 15:00
- đ Venue: MR5, CMS Pavilion A
Abstract
We propose a novel class of change of measure inequalities via a unified framework based on the data processing inequality for f-divergences, which is surprisingly elementary yet powerful enough to yield tighter inequalities. We provide change of measure inequalities in terms of a broad family of information measures, including f-divergences (with Kullback-Leibler divergence and $\chi^2$-divergence as special cases), Renyi divergence, and $\alpha$-mutual information (with maximal leakage as a special case). A key advantage of our framework is its flexibility: it readily adapts to a range of settings, including the generalization error analyses, the conditional mutual information framework, PAC-Bayesian theory, differential privacy mechanisms and data memorization problem, with simplified analyses
Series This talk is part of the Information Theory Seminar series.
Included in Lists
- All CMS events
- All Talks (aka the CURE list)
- bld31
- CMS Events
- DPMMS info aggregator
- DPMMS lists
- DPMMS Lists
- Hanchen DaDaDash
- Information Theory Seminar
- Interested Talks
- School of Physical Sciences
- Statistical Laboratory info aggregator
Note: Ex-directory lists are not shown.