Section: New Results
PAC-Bayesian Online Clustering
Participants : Benjamin Guedj, Le Li.
We have extended the PAC-Bayesian framework to online learning. Our algorithm (called PACBO) performs online clustering of random sequences, and is supported by strong theoretical (regret bounds) and algorithmic (ergodicity of an MCMC implementation) results. This work is now submitted to an international journal [46].
Joint work with Sébastien Loustau.