Limitations of Post-Hoc Feature Alignment for Robustness

03/10/2021
by   Collin Burns, et al.
5

Feature alignment is an approach to improving robustness to distribution shift that matches the distribution of feature activations between the training distribution and test distribution. A particularly simple but effective approach to feature alignment involves aligning the batch normalization statistics between the two distributions in a trained neural network. This technique has received renewed interest lately because of its impressive performance on robustness benchmarks. However, when and why this method works is not well understood. We investigate the approach in more detail and identify several limitations. We show that it only significantly helps with a narrow set of distribution shifts and we identify several settings in which it even degrades performance. We also explain why these limitations arise by pinpointing why this approach can be so effective in the first place. Our findings call into question the utility of this approach and Unsupervised Domain Adaptation more broadly for improving robustness in practice.

READ FULL TEXT

page 5

page 6

page 7

research
11/12/2018

Exploiting Local Feature Patterns for Unsupervised Domain Adaptation

Unsupervised domain adaptation methods aim to alleviate performance degr...
research
06/21/2021

ToAlign: Task-oriented Alignment for Unsupervised Domain Adaptation

Unsupervised domain adaptive classification intends to improve theclassi...
research
04/13/2019

Towards Self-similarity Consistency and Feature Discrimination for Unsupervised Domain Adaptation

Recent advances in unsupervised domain adaptation mainly focus on learni...
research
12/01/2021

The Norm Must Go On: Dynamic Unsupervised Domain Adaptation by Normalization

Domain adaptation is crucial to adapt a learned model to new scenarios, ...
research
03/25/2021

MetaAlign: Coordinating Domain Alignment and Classification for Unsupervised Domain Adaptation

For unsupervised domain adaptation (UDA), to alleviate the effect of dom...
research
10/07/2020

Revisiting Batch Normalization for Improving Corruption Robustness

Modern deep neural networks (DNN) have demonstrated remarkable success i...
research
12/05/2019

AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty

Modern deep neural networks can achieve high accuracy when the training ...

Please sign up or login with your details

Forgot password? Click here to reset