Back to Search Start Over

Learning PAC-Bayes Priors for Probabilistic Neural Networks

Authors :
Pérez-Ortiz, María
Rivasplata, Omar
Guedj, Benjamin
Gleeson, Matthew
Zhang, Jingyu
Shawe-Taylor, John
Bober, Miroslaw
Kittler, Josef
Department of Computer science [University College of London] (UCL-CS)
University College of London [London] (UCL)
The Inria London Programme (Inria-London)
University College of London [London] (UCL)-University College of London [London] (UCL)-Institut National de Recherche en Informatique et en Automatique (Inria)
Inria-CWI (Inria-CWI)
Centrum Wiskunde & Informatica (CWI)-Institut National de Recherche en Informatique et en Automatique (Inria)
MOdel for Data Analysis and Learning (MODAL)
Laboratoire Paul Painlevé (LPP)
Université de Lille-Centre National de la Recherche Scientifique (CNRS)-Université de Lille-Centre National de la Recherche Scientifique (CNRS)-Université de Lille, Sciences et Technologies-Inria Lille - Nord Europe
Institut National de Recherche en Informatique et en Automatique (Inria)-Institut National de Recherche en Informatique et en Automatique (Inria)-Evaluation des technologies de santé et des pratiques médicales - ULR 2694 (METRICS)
Université de Lille-Centre Hospitalier Régional Universitaire [Lille] (CHRU Lille)-Université de Lille-Centre Hospitalier Régional Universitaire [Lille] (CHRU Lille)-École polytechnique universitaire de Lille (Polytech Lille)
Centre for Vision, Speech and Signal Processing (CVSSP)
University of Surrey (UNIS)
Support and funding from the U.S. Army Research Laboratory and the U. S. Army Research Office, and by the U.K. Ministry of Defence and the U.K. Engineering and Physical Sciences Research Council (EPSRC) under grant number EP/R013616/1.
Computer science department [University College London] (UCL-CS)
Inria Lille - Nord Europe
Institut National de Recherche en Informatique et en Automatique (Inria)-Institut National de Recherche en Informatique et en Automatique (Inria)-Laboratoire Paul Painlevé - UMR 8524 (LPP)
Centre National de la Recherche Scientifique (CNRS)-Université de Lille-Centre National de la Recherche Scientifique (CNRS)-Université de Lille-Evaluation des technologies de santé et des pratiques médicales - ULR 2694 (METRICS)
Université de Lille-Centre Hospitalier Régional Universitaire [Lille] (CHRU Lille)-Université de Lille-Centre Hospitalier Régional Universitaire [Lille] (CHRU Lille)-École polytechnique universitaire de Lille (Polytech Lille)-Université de Lille, Sciences et Technologies
Publication Year :
2021
Publisher :
HAL CCSD, 2021.

Abstract

Recent works have investigated deep learning models trained by optimising PAC-Bayes bounds, with priors that are learnt on subsets of the data. This combination has been shown to lead not only to accurate classifiers, but also to remarkably tight risk certificates, bearing promise towards self-certified learning (i.e. use all the data to learn a predictor and certify its quality). In this work, we empirically investigate the role of the prior. We experiment on 6 datasets with different strategies and amounts of data to learn data-dependent PAC-Bayes priors, and we compare them in terms of their effect on test performance of the learnt predictors and tightness of their risk certificate. We ask what is the optimal amount of data which should be allocated for building the prior and show that the optimum may be dataset dependent. We demonstrate that using a small percentage of the prior-building data for validation of the prior leads to promising results. We include a comparison of underparameterised and overparameterised models, along with an empirical study of different training objectives and regularisation strategies to learn the prior distribution.

Details

Language :
English
Database :
OpenAIRE
Accession number :
edsair.doi.dedup.....83b31abd1ece13d412ee74fbe1c4f10e