DisturbLabel: Regularizing CNN on the Loss Layer

TL;DR

DisturbLabel is a simple regularization technique that introduces label noise during training to prevent overfitting in CNNs, working effectively alongside Dropout and improving recognition accuracy.

Contribution

This paper introduces DisturbLabel, the first method to add noise at the loss layer, providing a novel regularization approach that enhances CNN generalization.

Findings

DisturbLabel prevents overfitting effectively.

It complements Dropout for better regularization.

Achieves competitive results on image datasets.

Abstract

During a long period of time we are combating over-fitting in the CNN training process with model regularization, including weight decay, model averaging, data augmentation, etc. In this paper, we present DisturbLabel, an extremely simple algorithm which randomly replaces a part of labels as incorrect values in each iteration. Although it seems weird to intentionally generate incorrect training labels, we show that DisturbLabel prevents the network training from over-fitting by implicitly averaging over exponentially many networks which are trained with different label sets. To the best of our knowledge, DisturbLabel serves as the first work which adds noises on the loss layer. Meanwhile, DisturbLabel cooperates well with Dropout to provide complementary regularization functions. Experiments demonstrate competitive recognition results on several popular image recognition datasets.