# Achieving Generalizable Robustness of Deep Neural Networks by Stability Training

## TL;DR

This paper investigates stability training as a versatile method to enhance deep neural network robustness, demonstrating its effectiveness across various distortions and adversarial examples with fewer hyperparameters and side effects.

## Contribution

It introduces stability training as an alternative to data augmentation, showing it improves robustness against multiple unseen distortions and adversarial attacks.

## Key findings

- Stability training performs comparably or better than data augmentation on specific transformations.
- It offers improved robustness against a wider range of distortions and adversarial examples.
- Requires fewer hyperparameters and has fewer negative side effects.

## Abstract

We study the recently introduced stability training as a general-purpose method to increase the robustness of deep neural networks against input perturbations. In particular, we explore its use as an alternative to data augmentation and validate its performance against a number of distortion types and transformations including adversarial examples. In our image classification experiments using ImageNet data stability training performs on a par or even outperforms data augmentation for specific transformations, while consistently offering improved robustness against a broader range of distortion strengths and types unseen during training, a considerably smaller hyperparameter dependence and less potentially negative side effects compared to data augmentation.