On the Role of Label Noise in the Feature Learning Process

TL;DR

This paper provides a theoretical analysis of how label noise affects feature learning in deep neural networks, revealing two key training stages and supporting techniques like early stopping and sample selection.

Contribution

It offers a novel theoretical framework for understanding the impact of label noise on feature learning dynamics in neural networks.

Findings

Stage I involves fitting clean samples and learning the signal.

Stage II leads to overfitting noisy samples and memorization of label noise.

Early stopping and sample selection are validated as effective noise mitigation techniques.

Abstract

Deep learning with noisy labels presents significant challenges. In this work, we theoretically characterize the role of label noise from a feature learning perspective. Specifically, we consider a signal-noise data distribution, where each sample comprises a label-dependent signal and label-independent noise, and rigorously analyze the training dynamics of a two-layer convolutional neural network under this data setup, along with the presence of label noise. Our analysis identifies two key stages. In Stage I, the model perfectly fits all the clean samples (i.e., samples without label noise) while ignoring the noisy ones (i.e., samples with noisy labels). During this stage, the model learns the signal from the clean samples, which generalizes well on unseen data. In Stage II, as the training loss converges, the gradient in the direction of noise surpasses that of the signal, leading to overfitting on noisy samples. Eventually, the model memorizes the noise present in the noisy samples and degrades its generalization ability. Furthermore, our analysis provides a theoretical basis for two widely used techniques for tackling label noise: early stopping and sample selection. Experiments on both synthetic and real-world setups validate our theory.