Easy Batch Normalization

TL;DR

This paper explores the use of easy examples in neural network training, proposing an auxiliary batch normalization technique to improve both standard and robust accuracy.

Contribution

It introduces a novel auxiliary batch normalization method specifically for easy examples, a first step in leveraging easy samples to enhance training.

Findings

Auxiliary batch normalization improves standard accuracy.

It enhances robust accuracy against adversarial attacks.

The method shows promising results in initial experiments.

Abstract

It was shown that adversarial examples improve object recognition. But what about their opposite side, easy examples? Easy examples are samples that the machine learning model classifies correctly with high confidence. In our paper, we are making the first step toward exploring the potential benefits of using easy examples in the training procedure of neural networks. We propose to use an auxiliary batch normalization for easy examples for the standard and robust accuracy improvement.