Section: New Results

An Oracle Inequality for Quasi-Bayesian Non-Negative Matrix Factorization

Participant : Benjamin Guedj.

The quasi-Bayesian perspective has been extended to the popular setting of non-negative matrix factorization. This is a pivotal problem in machine learning (image segmentation, recommendation systems, audio source separation, ...) and an original estimator of the unobserved matrix has been proposed. An oracle inequality is derived, along with several possible implementations. This work is published in Mathematical Methods of Statistics [12].

It a joint work with Pierre Alquier from ENSAE - Université Paris-Saclay.