Séminaire de Probabilités et Statistique
lundi 18 février 2019 à 13:45 - UM - Bât 09 - Salle de conférence (1er étage)
Benjamin Guedj (INRIA Lille)
A primer on PAC-Bayesian learning
Generalized Bayesian learning algorithms are increasingly popular in machine learning, due to their PAC generalization properties and flexibility. I will present a self-contained introduction on generalized Bayesian learning and the PAC-Bayes theory, and discuss their theoretical and algorithmic ins and outs. I will then focus on the recent paper Alquier and Guedj (2018), and present how PAC-Bayesian ideas may be used to efficiently learn with dependent and/or heavy-tailed (aka hostile) data. References: * Alquier and Guedj (2018), Simpler PAC-Bayesian Bounds for Hostile Data, Machine Learning. https://link.springer.com/article/10.1007%2Fs10994-017-5690-0 * Guedj (2019), A primer on PAC-Bayesian learning, arXiv preprint. https://arxiv.org/abs/1901.05353