Robust Feature Learning Against Noisy Labels

TL;DR

This paper introduces a robust feature learning method that effectively handles noisy labels in supervised deep learning by combining unsupervised augmentation, cluster regularization, and self-bootstrapping, improving model robustness.

Contribution

The paper presents a novel, flexible approach that enhances deep neural network robustness against noisy labels through unsupervised and self-supervised techniques with minimal additional overhead.

Findings

Improves model robustness under severe label noise

Enhances generalization in noisy label scenarios

Compatible with existing architectures

Abstract

Supervised learning of deep neural networks heavily relies on large-scale datasets annotated by high-quality labels. In contrast, mislabeled samples can significantly degrade the generalization of models and result in memorizing samples, further learning erroneous associations of data contents to incorrect annotations. To this end, this paper proposes an efficient approach to tackle noisy labels by learning robust feature representation based on unsupervised augmentation restoration and cluster regularization. In addition, progressive self-bootstrapping is introduced to minimize the negative impact of supervision from noisy labels. Our proposed design is generic and flexible in applying to existing classification architectures with minimal overheads. Experimental results show that our proposed method can efficiently and effectively enhance model robustness under severely noisy labels.