Saved in:
Bibliographic Details
Main Authors: Leblanc, Benjamin, Bazinet, Mathieu, D'Amours, Nathaniel, Drouin, Alexandre, Germain, Pascal
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.13577
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916779840765952
author Leblanc, Benjamin
Bazinet, Mathieu
D'Amours, Nathaniel
Drouin, Alexandre
Germain, Pascal
author_facet Leblanc, Benjamin
Bazinet, Mathieu
D'Amours, Nathaniel
Drouin, Alexandre
Germain, Pascal
contents Both PAC-Bayesian and Sample Compress learning frameworks are instrumental for deriving tight (non-vacuous) generalization bounds for neural networks. We leverage these results in a meta-learning scheme, relying on a hypernetwork that outputs the parameters of a downstream predictor from a dataset input. The originality of our approach lies in the investigated hypernetwork architectures that encode the dataset before decoding the parameters: (1) a PAC-Bayesian encoder that expresses a posterior distribution over a latent space, (2) a Sample Compress encoder that selects a small sample of the dataset input along with a message from a discrete set, and (3) a hybrid between both approaches motivated by a new Sample Compress theorem handling continuous messages. The latter theorem exploits the pivotal information transiting at the encoder-decoder junction in order to compute generalization guarantees for each downstream predictor obtained by our meta-learning scheme.
format Preprint
id arxiv_https___arxiv_org_abs_2410_13577
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Generalization Bounds via Meta-Learned Model Representations: PAC-Bayes and Sample Compression Hypernetworks
Leblanc, Benjamin
Bazinet, Mathieu
D'Amours, Nathaniel
Drouin, Alexandre
Germain, Pascal
Machine Learning
Both PAC-Bayesian and Sample Compress learning frameworks are instrumental for deriving tight (non-vacuous) generalization bounds for neural networks. We leverage these results in a meta-learning scheme, relying on a hypernetwork that outputs the parameters of a downstream predictor from a dataset input. The originality of our approach lies in the investigated hypernetwork architectures that encode the dataset before decoding the parameters: (1) a PAC-Bayesian encoder that expresses a posterior distribution over a latent space, (2) a Sample Compress encoder that selects a small sample of the dataset input along with a message from a discrete set, and (3) a hybrid between both approaches motivated by a new Sample Compress theorem handling continuous messages. The latter theorem exploits the pivotal information transiting at the encoder-decoder junction in order to compute generalization guarantees for each downstream predictor obtained by our meta-learning scheme.
title Generalization Bounds via Meta-Learned Model Representations: PAC-Bayes and Sample Compression Hypernetworks
topic Machine Learning
url https://arxiv.org/abs/2410.13577