I successfully defended my PhD yesterday, in front of a jury composed of:

  • Arthur Gretton (reviewer)
  • Romain Couillet (reviewer)
  • Francis Bach (president)
  • Mike Davies
  • Karin Schnass
  • François Taiani
  • Rémi Gribonval (advisor)

Thanks to all who attended. The manuscript should be online soon, in the meantime the slides can be found here.

Preprint available

Our preprint is (finally !) available:


Rémi Gribonval, Gilles Blanchard, Nicolas Keriven, Yann Traonmilin. Compressive Statistical Learning and Random Features Moments


It contains a lot of sketch-related theoretical stuff. Any form feedback is welcome !

SPARS 2017 Best Student Paper Award !

Our two-pages abstract:

N. Keriven, R. Gribonval, G. Blanchard, Y. Traonmilin. Random Moments for Sketched Mixture Learning

has been awarded the Best Student Paper Award at SPARS 2017 !

Lots of thanks to the committee and my co-authors !



Lisbon ! (SPARS 2017)

We’ll give two oral presentations at SPARS2017 in Lisbon (June 5-8), associated to one-page abstracts:

See you there !

New Orleans ! (ICASSP 2017)

Our paper Compressive K-means has been accepted at ICASSP 2017 in New Orleans, in the Special Session « Random Embeddings and Geometry-Preserving Dimensionality Reduction ». I’ll be at the venue to present it, see you there !

The list of accepted papers.

Compressive K-means

Our paper on Compressive K-means is online ! Read it here. Download the code here.

The SketchMLbox is available !

The SketchMLbox, a Matlab toolbox for sketched mixture learning, is available ! It allows you to learn mixture models on large databases, as described in this paper.

Estimations of mixtures of Diracs (« K-means ») and Gaussian mixture models are available, and the toolbox is structured so that new mixture models can be easily implemented.

The toolbox can be downloaded on this page.

Peyresq 2016

I presented our work at the Peyresq 2016 summer school, a yearly event organized by GRETSI.

See the presentation here.

New preprint available

Our latest preprint is available !

Sketching for Large-Scale Learning of Mixture Models (arXiv:1606.02838)

This is the « long » version of the paper presented at ICASSP 2016 in Shanghai.

Any comments are welcome !


I presented our work on sketching methods to the GdR ISIS, at a meeting entitled « Algorithmes gloutons pour l’optimisation sous contrainte de parcimonie ».

For members of the GdR a summary of the meeting can be found here, and you can find my presentation here.