Abstract:Modeling noise transition matrix is a kind of promising method for learning with label noise. Based on the estimated noise transition matrix and the noisy posterior probabilities, the clean posterior probabilities, which are jointly called Label Distribution (LD) in this paper, can be calculated as the supervision. To reliably estimate the noise transition matrix, some methods assume that anchor points are available during training. Nonetheless, if anchor points are invalid, the noise transition matrix might be poorly learned, resulting in poor performance. Consequently, other methods treat reliable data points, extracted from training data, as pseudo anchor points. However, from a statistical point of view, the noise transition matrix can be inferred from data with noisy labels under the clean-label-domination assumption. Therefore, we aim to estimate the noise transition matrix without (pseudo) anchor points. There is evidence showing that samples are more likely to be mislabeled as other similar class labels, which means the mislabeling probability is highly correlated with the inter-class correlation. Inspired by this observation, we propose an instance-specific Label Distribution Regularization (LDR), in which the instance-specific LD is estimated as the supervision, to prevent DCNNs from memorizing noisy labels. Specifically, we estimate the noisy posterior under the supervision of noisy labels, and approximate the batch-level noise transition matrix by estimating the inter-class correlation matrix with neither anchor points nor pseudo anchor points. Experimental results on two synthetic noisy datasets and two real-world noisy datasets demonstrate that our LDR outperforms existing methods.

The Manifold Regularized SVDD for Noisy Label Detection

The fuzzy support vector data description based on tightness for noisy label detection

FGCM: Noisy Label Learning via Fine-Grained Confidence Modeling

SLV: Spatial Likelihood Voting for Weakly Supervised Object Detection

Spatial Likelihood Voting with Self-Knowledge Distillation for Weakly Supervised Object Detection.

Noise-resistant Graph Neural Networks with Manifold Consistency and Label Consistency

Robust Semi-Supervised Classification for Noisy Labels Based on Self-Paced Learning

Learning with Noisy Labels Via Self-supervised Adversarial Noisy Masking

SV-Learner: Support-Vector Contrastive Learning for Robust Learning with Noisy Labels

On Better Detecting and Leveraging Noisy Samples for Learning with Severe Label Noise

Safeguarded Dynamic Label Regression for Noisy Supervision

Learning With Non-Uniform Label Noise: A Cluster-Dependent Weakly Supervised Approach.

A Meta Approach to Defend Noisy Labels by the Manifold Regularizer PSDR

Instance-specific Label Distribution Regularization for Learning with Label Noise

Online Multi-Label Classification under Noisy and Changing Label Distribution

Unleashing the Potential of Regularization Strategies in Learning with Noisy Labels

Safeguarded Dynamic Label Regression for Generalized Noisy Supervision

Robust Image Classification with Noisy Labels by Negative Learning and Feature Space Renormalization

Noisy Label Processing for Classification: A Survey

Label Distribution Learning with Noisy Labels Via Three-Way Decisions

A Bi-level Formulation for Label Noise Learning with Spectral Cluster Discovery