Sample Prior Guided Robust Model Learning to Suppress Noisy Labels

TL;DR

This paper introduces PGDF, a novel framework that leverages prior knowledge to better identify clean and noisy samples, thereby improving model robustness against noisy labels in both synthetic and real-world datasets.

Contribution

The proposed PGDF framework integrates prior knowledge into sample division and pseudo-label generation, effectively preserving hard clean samples and enhancing pseudo-label quality.

Findings

Significant performance improvements over state-of-the-art methods.

Effective noise suppression in both synthetic and real-world datasets.

Better preservation of informative hard samples.

Abstract

Imperfect labels are ubiquitous in real-world datasets and seriously harm the model performance. Several recent effective methods for handling noisy labels have two key steps: 1) dividing samples into cleanly labeled and wrongly labeled sets by training loss, 2) using semi-supervised methods to generate pseudo-labels for samples in the wrongly labeled set. However, current methods always hurt the informative hard samples due to the similar loss distribution between the hard samples and the noisy ones. In this paper, we proposed PGDF (Prior Guided Denoising Framework), a novel framework to learn a deep model to suppress noise by generating the samples' prior knowledge, which is integrated into both dividing samples step and semi-supervised step. Our framework can save more informative hard clean samples into the cleanly labeled set. Besides, our framework also promotes the quality of pseudo-labels during the semi-supervised step by suppressing the noise in the current pseudo-labels generating scheme. To further enhance the hard samples, we reweight the samples in the cleanly labeled set during training. We evaluated our method using synthetic datasets based on CIFAR-10 and CIFAR-100, as well as on the real-world datasets WebVision and Clothing1M. The results demonstrate substantial improvements over state-of-the-art methods.