Welcome to my homepage. Since September 2015, I am an Assistant Professor at ENSAI, Bruz, France and an academic member of the Crest lab. I am working in the field of statistical learning/non-parametric statistics. I focus in particular on the problem of hyper-parameter tuning for learning algorithms. I try to obtain methodological improvements from an in-depth theoretical understanding of model selection issues. So far, I studied the so-called Slope Heuristics as well as resampling (penalization) methods. A central theoretical aspect of this line of research, based on Empirical Process Theory, is to obtain precise concentration bounds for the excess risk of M-estimators.

I also work on the related subjects of learning from Markov chains, robust learning (i.e. learning with heavy tailed and/or corrupted data), sketching for massive data, Bregman clustering, Small-ball approach in Learning Theory and non-parametric statistics, as well a some probabilistic inequalities linked to log-concave measures.

Since 2016, I am an elected member of the "groupe stat math" which is a group of researcher aiming at promoting mathematical statistics in France and on behalf of the French statistical society (Société Française de Statistique, SfDS).

Since sept. 2015, I co-organize the Statistical seminar in Rennes.

Email: asaumardatgmaildotcom

Phone: +33 (0)2 99 05 32 10

Curriculum Vitae

Google Scholar

I am a member of the scientific program comittee for CMStatistics 2019, 14-16 december, London.

Research topics

  • statistical learning theory
  • model selection
  • slope heuristics
  • resampling methods
  • small-ball approach in learning theory
  • learning from Markov chains
  • robust learning (MOM principle)
  • sketching for learning with massive data
  • Bregman clustering
  • empirical process theory
  • concentration inequalities
  • functional inequalities
  • Stein's method