Welcome to my homepage. I am an assistant professor in the statistics group at Leiden University in the Netherlands, working on machine learning and statistics, and teaching in the Statistical Science master program.
News
- Co-organizing the Young European Statisticians workshop on the Theoretical Foundations of Deep Learning
- We recently had a very interesting Netherlands N(eur)IPS debriefing
- Co-organizing the thematic seminar with current theme machine learning
- Started a mailing list for machine learning in the Netherlands
Research Interests
- Adaptive methods in online convex optimization and prediction with expert advice
- Faster-than-minimax rates for ‘easy data’ in statistical and online learning
- Statistical learning theory, frequentist analysis of Bayesian methods
- PAC-Bayesian concentration inequalities
- Information theory in statistics, minimum description length
- Model selection
See also my publications.
PhD Students
- Dirk van der Hoeven, started October 2016
Recent Talks
- Invited talk with some highlights in online convex optimization in the machine learning session at the Netherlands Mathematical Congress, 2018
- Invited talks in Toulouse, Delft, Rennes, Leuven, Poznań and Amsterdam, 2017-2018, about using multiple learning rates in online learning with MetaGrad to automatically adapt to the best possible learning rate. A related tutorial introduction to adaptive online learning at ABN AMRO, Amsterdam, 2018.
- Invited talk at Inria Lille, 2017, about fast rates in statistical and online learning. Also gave this talk in the Thematic Seminar 2018.
- Talked about MetaGrad, a new adaptive online optimization method, at the Theoretical Foundations for Learning from Easy Data workshop in Leiden, 2016
- Invited talk in the WIC Midwintermeeting 2016 on Big Data and Data Analytics
- Presented my work with Wouter Koolen on automatically learning the learning rate parameter in online learning at Benelearn 2015 and for the Leiden Centre of Data Science
- Introduction to Online Learning for a general mathematical audience in the Leiden Mathematics Colloquium, 2014
- Presented an open problem in the Online Algorithms and Learning international workshop in Leiden, 2014
- Talk about Follow the Leader with Dropout Perturbations at COLT 2014 in Barcelona
- Invited talk on convergence of the MDL estimator in the session on Model Selection in Statistical Learning at the ERCIM conference in Londen, December 14, 2013
- Invited talk about Follow the Leader If You Can, Hedge If You Must in the Learning Faster from Easy Data workshop at NIPS in Lake Tahoe, USA, December 10, 2013
- Making Regional Forecasts Add Up at the WIPFOR workshop in Paris, June 6, 2013
- The Catch-up Phenomenon in Bayesian and MDL Model Selection in the “Bayes in Paris” seminar, May 23, 2013
- Invited lecture giving An Introduction to Online Learning for Bayesians at the University of Amsterdam, February 15, 2013
- Mixability in Statistical Learning in the SMILE seminar in Paris, September 24, 2012
Contact Information
Dr. Tim van Erven
Mathematisch instituut
Universiteit Leiden
P.O. Box 9512
2300 RA Leiden
The Netherlands
Office: Mathematisch instituut, Snellius building, Niels Bohrweg 1, Leiden, room 226
E-mail: tim@ No spam, please timvanerven. No really, no spam nl
Phone: +31 71 527 7126