PatchUp : A Feature-Space Block-Level Regularization Technique for Convolutional Neural Networks

Mojtaba Faramarzi, Mohammad Amini, Akilesh Badrinaaraayanan, Vikas Verma, Sarath Chandar

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

7 Citations (Scopus)

Abstract

Large capacity deep learning models are often prone to a high generalization gap when trained with a limited amount of labeled training data. A recent class of methods to address this problem uses various ways to construct a new training sample by mixing a pair (or more) of training samples. We propose PatchUp, a hidden state block-level regularization technique for Convolutional Neural Networks (CNNs), that is applied on selected contiguous blocks of feature maps from a random pair of samples. Our approach improves the robustness of CNN models against the manifold intrusion problem that may occur in other state-of-the-art mixing approaches. Moreover, since we are mixing the contiguous block of features in the hidden space, which has more dimensions than the input space, we obtain more diverse samples for training towards different dimensions. Our experiments on CIFAR10/100, SVHN, Tiny-ImageNet, and ImageNet using ResNet architectures including PreActResnet18/34, WRN-28-10, ResNet101/152 models show that PatchUp improves upon, or equals, the performance of current state-of-the-art regularizers for CNNs. We also show that PatchUp can provide a better generalization to deformed samples and is more robust against adversarial attacks.

Original languageEnglish
Title of host publicationAAAI-22 Technical Tracks 1
PublisherAAAI Press
Pages589-597
Number of pages9
ISBN (Electronic)978-1-57735-876-3
Publication statusPublished - 30 Jun 2022
MoE publication typeA4 Conference publication
EventAAAI Conference on Artificial Intelligence - virtual conference, Virtual, Online
Duration: 22 Feb 20221 Mar 2022
Conference number: 36
https://aaai.org/Conferences/AAAI-22/

Publication series

NameProceedings of the AAAI Conference on Artificial Intelligence
Volume36

Conference

ConferenceAAAI Conference on Artificial Intelligence
Abbreviated titleAAAI
CityVirtual, Online
Period22/02/202201/03/2022
Internet address

Fingerprint

Dive into the research topics of 'PatchUp : A Feature-Space Block-Level Regularization Technique for Convolutional Neural Networks'. Together they form a unique fingerprint.

Cite this