TrustNet: Learning from Trusted Data Against (A)symmetric Label Noise
This addresses the critical issue of label noise for weakly-supervised classifiers, particularly in large datasets, with incremental improvements over existing methods.
The paper tackles the problem of robust classification under symmetric and asymmetric label noise by proposing TrustNet, which learns noise patterns from a small trusted dataset and uses a robust loss function with uncertainty-based weighting. It demonstrates strong robustness on synthetic CIFAR-10/100 and real-world Clothing1M datasets, outperforming state-of-the-art methods.
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.