Since Jan 2010, I am a PhD student at Paris-Sud 11 University, in the INRIA TAO Team and with the LRI and LIMSI laboratories and under the supervision of Philippe Tarroux (LIMSI) and Hélène Paugam-Moisy (Lyon University and TAO Team).

The subject of my thesis is “Deep Architectures for Computational Vision”.

Deep Architecture are a recent subject (2006) in Artificial Intelligence, in which a machine learns hierarchical representations of the world, somewhat like humans. Unfortunately, the awesomeness of deep architectures cannot be fathomed quickly as it requires a lot of computer time for the system to learn things. Unlike Humans who usually have a few years before we expect them to be productive (decades anyone ?), machines have weeks or months at most (time between two publications). I'm currently working on speeding things up a bit, we'll see how it turns out.

This year, I will be one of the organizing members of the ESANN 2011 special session dedicated to deep learning. Here is a quick a tutorial on Deep Learning we wrote for the occasion entitled “An Introduction to Deep Learning” publications2011.

Talk for the ECAI 2010 conference: talk-ecai2010.pdf

The paper “Unsupervised Layer-Wise Model Selection in Deep Neural Networks” (publications2010) has just been accepted for the conference ECAI 2010 which will be held in Lisbon at the end of August (August 16-20). The paper will be available on the publication page as soon as it is camera ready.

The source code used to perform experiments on RBMs and DBNs is now available on the Machine Learning project page.

The Paper “Optimisation de la Topologie pour les Reseaux de Neurones Profonds” (publications2010) submited to the French conference RFIA has just been accepted. I will be presenting it in the city of Caen (France) at the end of January 2010.