TrustNet: Learning from Trusted Data Against (A)symmetric Label Noise

TL;DR

TrustNet is a novel method that learns noise patterns from trusted data to improve classifier robustness against both symmetric and asymmetric label noise in large datasets.

Contribution

It introduces a noise pattern learning approach combined with a robust loss function, enhancing label noise robustness in weakly-supervised classifiers.

Findings

TrustNet outperforms state-of-the-art methods on CIFAR-10 and CIFAR-100 with synthetic noise.

TrustNet demonstrates strong robustness on real-world Clothing1M dataset.

The method effectively models both symmetric and asymmetric label noise patterns.

Abstract

Robustness to label noise is a critical property for weakly-supervised classifiers trained on massive datasets. Robustness to label noise is a critical property for weakly-supervised classifiers trained on massive datasets. In this paper, we first derive analytical bound for any given noise patterns. Based on the insights, we design TrustNet that first adversely learns the pattern of noise corruption, being it both symmetric or asymmetric, from a small set of trusted data. Then, TrustNet is trained via a robust loss function, which weights the given labels against the inferred labels from the learned noise pattern. The weight is adjusted based on model uncertainty across training epochs. We evaluate TrustNet on synthetic label noise for CIFAR-10 and CIFAR-100, and real-world data with label noise, i.e., Clothing1M. We compare against state-of-the-art methods demonstrating the strong robustness of TrustNet under a diverse set of noise patterns.