Deep Learning is Robust to Massive Label Noise

TL;DR

Deep neural networks can effectively learn from large datasets with massive label noise, maintaining high test accuracy despite extensive incorrect labels, which challenges traditional views on data quality requirements.

Contribution

This paper demonstrates that deep learning models can generalize well even with heavily noisy labels and analyzes how noise impacts training dynamics and dataset size requirements.

Findings

High test accuracy (>90%) on MNIST with 100x label noise

Robustness across different noise patterns, including biased errors

Increased dataset size mitigates effects of label corruption

Abstract

Deep neural networks trained on large supervised datasets have led to impressive results in image classification and other tasks. However, well-annotated datasets can be time-consuming and expensive to collect, lending increased interest to larger but noisy datasets that are more easily obtained. In this paper, we show that deep neural networks are capable of generalizing from training data for which true labels are massively outnumbered by incorrect labels. We demonstrate remarkably high test performance after training on corrupted data from MNIST, CIFAR, and ImageNet. For example, on MNIST we obtain test accuracy above 90 percent even after each clean training example has been diluted with 100 randomly-labeled examples. Such behavior holds across multiple patterns of label noise, even when erroneous labels are biased towards confusing classes. We show that training in this regime requires a significant but manageable increase in dataset size that is related to the factor by which correct labels have been diluted. Finally, we provide an analysis of our results that shows how increasing noise decreases the effective batch size.