Biggs, Felix;
Guedj, Benjamin;
(2022)
On Margins and Derandomisation in PAC-Bayes.
In: Camps-Valls, Gustau and Ruiz, Francisco JR and Valera, Isabel, (eds.)
Proceedings of The 25th International Conference on Artificial Intelligence and Statistics.
(pp. pp. 3709-3731).
Proceedings of Machine Learning Research (PMLR): Virtual conference.
Preview |
Text
biggs22a.pdf - Published Version Download (855kB) | Preview |
Abstract
We give a general recipe for derandomising PAC-Bayesian bounds using margins, with the critical ingredient being that our randomised predictions concentrate around some value. The tools we develop straightforwardly lead to margin bounds for various classifiers, including linear prediction—a class that includes boosting and the support vector machine—single-hidden-layer neural networks with an unusual erf activation function, and deep ReLU networks. Further we extend to partially-derandomised predictors where only some of the randomness of our estimators is removed, letting us extend bounds to cases where the concentration properties of our estimators are otherwise poor.
| Type: | Proceedings paper |
|---|---|
| Title: | On Margins and Derandomisation in PAC-Bayes |
| Event: | 25th International Conference on Artificial Intelligence and Statistics (AISTATS) 2022 |
| Open access status: | An open access version is available from UCL Discovery |
| Publisher version: | https://proceedings.mlr.press/v151/biggs22a.html |
| Language: | English |
| Additional information: | This is an Open Access paper published under a Creative Commons Attribution 4.0 International (CC BY 4.0) Licence (https://creativecommons.org/licenses/by/4.0/). |
| UCL classification: | UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Computer Science UCL > Provost and Vice Provost Offices > UCL BEAMS UCL |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10152711 |
Archive Staff Only
 |
View Item |
