CEB Improves Model Robustness
Intuitively, one way to make classifiers more robust to their input is to have them depend less sensitively on their input. The Information Bottleneck (IB) tries to learn compressed representations of input that are still predictive. Scaling up IB approaches to large scale image classification tasks...
Main Authors: | , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2020-09-01
|
Series: | Entropy |
Subjects: | |
Online Access: | https://www.mdpi.com/1099-4300/22/10/1081 |
id |
doaj-58571da5045b4d7e8d4e0d604ddb3f32 |
---|---|
record_format |
Article |
spelling |
doaj-58571da5045b4d7e8d4e0d604ddb3f322020-11-25T03:52:48ZengMDPI AGEntropy1099-43002020-09-01221081108110.3390/e22101081CEB Improves Model RobustnessIan Fischer0Alexander A. Alemi1Google Research, Mountain View, CA 94043, USAGoogle Research, Mountain View, CA 94043, USAIntuitively, one way to make classifiers more robust to their input is to have them depend less sensitively on their input. The Information Bottleneck (IB) tries to learn compressed representations of input that are still predictive. Scaling up IB approaches to large scale image classification tasks has proved difficult. We demonstrate that the Conditional Entropy Bottleneck (CEB) can not only scale up to large scale image classification tasks, but can additionally improve model robustness. CEB is an easy strategy to implement and works in tandem with data augmentation procedures. We report results of a large scale adversarial robustness study on CIFAR-10, as well as the ImageNet-C Common Corruptions Benchmark, ImageNet-A, and PGD attacks.https://www.mdpi.com/1099-4300/22/10/1081information theoryinformation bottleneckmachine learning |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Ian Fischer Alexander A. Alemi |
spellingShingle |
Ian Fischer Alexander A. Alemi CEB Improves Model Robustness Entropy information theory information bottleneck machine learning |
author_facet |
Ian Fischer Alexander A. Alemi |
author_sort |
Ian Fischer |
title |
CEB Improves Model Robustness |
title_short |
CEB Improves Model Robustness |
title_full |
CEB Improves Model Robustness |
title_fullStr |
CEB Improves Model Robustness |
title_full_unstemmed |
CEB Improves Model Robustness |
title_sort |
ceb improves model robustness |
publisher |
MDPI AG |
series |
Entropy |
issn |
1099-4300 |
publishDate |
2020-09-01 |
description |
Intuitively, one way to make classifiers more robust to their input is to have them depend less sensitively on their input. The Information Bottleneck (IB) tries to learn compressed representations of input that are still predictive. Scaling up IB approaches to large scale image classification tasks has proved difficult. We demonstrate that the Conditional Entropy Bottleneck (CEB) can not only scale up to large scale image classification tasks, but can additionally improve model robustness. CEB is an easy strategy to implement and works in tandem with data augmentation procedures. We report results of a large scale adversarial robustness study on CIFAR-10, as well as the ImageNet-C Common Corruptions Benchmark, ImageNet-A, and PGD attacks. |
topic |
information theory information bottleneck machine learning |
url |
https://www.mdpi.com/1099-4300/22/10/1081 |
work_keys_str_mv |
AT ianfischer cebimprovesmodelrobustness AT alexanderaalemi cebimprovesmodelrobustness |
_version_ |
1724480875764121600 |