====== Recent activities ======
  * Reviewing committee Member [[http://www.ecmlpkdd2009.net/|ECML 2009]] [[http://www.ecmlpkdd2010.org/| ECML 2010]], [[http://www.ecmlpkdd2011| ECML 2011]], [[http://www.ecmlpkdd2012 | ECML 2012]][[http://www.ecmlpkdd2013| ECML 2013]], [[http://www.ecmlpkdd2014 | ECML 2014]], [[http://www.ecmlpkdd2015| ECML 2015]].
  * Program committee Member [[http://oregonstate.edu/conferences/icml2007/|ICML 2007]], [[http://www.cs.mcgill.ca/~icml2009/|ICML 2009]], [[http://www.icml2010.org/|ICML 2010]], [[http://www.icml-2011.org/|ICML 2011]], [[http://icml.cc/2012/|ICML 2012]], [[http://icml.cc/2013/|ICML 2013]], [[http://icml.cc/2014/|ICML 2014]], [[http://icml.cc/2015/|ICML 2015]].
  * Reviewing committee Member [[http://nips.cc/|NIPS 2007 -- today]]
  * PC Member CAp 2005-  ([[http://afia2007.imag.fr/cap/|2007 edition]])

====== Recent interests =======


===== Ranking and interdependent data =====
I'm particularly interested in the problems that go beyong the classical assumption of IID training data. In particular, I'm interested in
  * theoretical aspects (e.g. generalization bounds, concentration inequalities) of learning with interdependent data;
  * bipartite ranking.

===== Dealing with noisy/incomplete labels =====
Another side of my research is devoted to learning with noisy data. I've worked on questions like
  * learning with classification noise, with a special emphasis on (kernel) hyperplane learning;
  * learning with weakly supervised data (which has strong connections with learning with noisy data)

===== (Stochastic) convex optimization =====
The need for convex optimization procedures is of the utmost importance for machine learning.
I've been working on various problems requiring convex optimization. In particular, I've worked on
  * learning with indefinite kernels and structured sparsity with mixed-norm regularization;
  * cutting plane algorithms for noisy data;
  * stochastic optimization (in progress).

===== Kernels for Chemoinformatics =====

Over the past years, I've been fortunate to collaborate with [[http://www.ics.uci.edu/~pfbaldi|Pierre Baldi]] and his group at IGB, on the one hand, and [[http://cg.ensmp.fr/~mahe/index_eng.html|Pierre Mahé]] and [[http://cg.ensmp.fr/~vert|Jean-Philippe Vert]], on the other hand, on the design of kernels for chemoinformatics problems. The question addressed is to be able to derive (kernel) similarities both efficient (wrt computation time) and "accurate" (i.e. that make sense given a predefined problem). These similarities are to be leveraged in classifiers such as support vector machines and boosting methods.