APRIORI Project – A PAC-Bayesian Representation Learning Perspective

Keywords of the Project

PAC-Bayesian Theory

A Statistical Machine Learning well-designed for functions that take the form of a weighted combination

APRIORI in few words

In data science, defining a meaningful data representation is often a crucial preliminary step of the data processing pipeline. The Apriori project is rooted in the sub-field of machine learning called representation learning, which overlaps many topics such as deep learning, metric learning and kernel methods. Apriori focuses on understanding and replicating the successes of those methods. While classical machine learning algorithms are supported by sound theoretical analyses and clear heuristics, representation learning is often deemed to be lacking both. The originality of the project stands in bridging the gap between the representation learning algorithms and the theory by using the PAC- Bayesian arsenal: a statistical machine learning theory that appears naturally tailored to study representation learning problems. We believe that a sound analysis of representation learning—that will also lead to new well-founded algorithms—can be a significant contribution to machine learning and representation learning practitioners.

Welcome to our new post doc

emorvant 2021/03/03 2021/03/03News

We are pleased to welcome a new member : Valentina Zantedeschi a new post-doctoral researcher

Accepted paper @ NeurIPS 2019

emorvant 2019/10/02 2019/10/02Paper

Dichotomize and Generalize: PAC-Bayesian Binary Activated Deep Neural Networks Gaël Letarte, Pascal Germain, Benjamin Guedj, François Laviolette Conference on Neural Information Processing Systems (NeurIPS), 2019 [pdf] and also a communication in a workshop: Interpreting Neural Networks as Majority Votes through the PAC-Bayesian Theory Paul Viallard ; Rémi Emonet ; Pascal…

Welcome to two new APRIORI members

emorvant 2019/10/01 2019/10/01News

We are pleased to welcome two new members : Hemant Tyagi who is a researcher at Inria Lille – Nord Europe Paul Viallard has a Ph.D. Student co-supervised by A. Habrard and E. Morvant (Lab. H. Curien) and P. Germain (Inria Lille – Nord Europe, GRAAL Québec)

Benjamin Guedj will give a Tutorial on PAC-Bayes @ ICML

emorvant 2019/05/28 2019/05/28News

You are going to ICML 2019 ? Then don’t miss the tutorial on PAC-Bayes by Benjamin Guedj and John Shawe-Taylor on Monday, June 10th! More informations here

A Priori Members


Rémi Emonet Assistant Professor Lab. Hubert Curien	Pascal Germain Assistant Professor GRAAL & Inria Lille - Nord Europe	Benjamin Guedj Research Scientist UCL & Inria Lille - Nord Europe	Amaury Habrard Professor Lab Hubert Curien	Emilie Morvant Assistant Professor Lab. Hubert Curien

Hemant Tyagi Research Scientist Inria Lille - Nord Europe	Léo Gautheron Post Doc Lab. Hubert Curien	Valentina Zantedeschi Post-Doc UCL & Inria Lille - Nord Europe	Paul Viallard Ph.D. Student Lab. Hubert Curien