Section: New Results

Simpler PAC-Bayesian Bounds for Hostile Data

Participant : Benjamin Guedj.

We have introduced an original and much simpler way of deriving PAC-Bayesian bounds, through the use of $f$-divergences (therefore generalizing earlier works on Renyi's divergence and Kullback-Leibler divergence). This work is now submitted to an international conference [39].

Joint work with Pierre Alquier.