Real World Robustness from Systematic Noise

TL;DR

This paper investigates how systematic errors, such as implementation differences in image decoding, cause adversarial examples that reduce neural network accuracy in real-world scenarios, and introduces a dataset for benchmarking this robustness.

Contribution

It identifies systematic errors as a source of adversarial examples, and proposes the ImageNet-S dataset to evaluate classifier robustness against such errors.

Findings

ResNet-50 accuracy drops 1%-5% due to systematic errors

Systematic errors significantly impact real-world model performance

The ImageNet-S dataset enables benchmarking of systematic error robustness

Abstract

Systematic error, which is not determined by chance, often refers to the inaccuracy (involving either the observation or measurement process) inherent to a system. In this paper, we exhibit some long-neglected but frequent-happening adversarial examples caused by systematic error. More specifically, we find the trained neural network classifier can be fooled by inconsistent implementations of image decoding and resize. This tiny difference between these implementations often causes an accuracy drop from training to deployment. To benchmark these real-world adversarial examples, we propose ImageNet-S dataset, which enables researchers to measure a classifier's robustness to systematic error. For example, we find a normal ResNet-50 trained on ImageNet can have 1%-5% accuracy difference due to the systematic error. Together our evaluation and dataset may aid future work toward real-world robustness and practical generalization.