Understanding and Utilizing Deep Neural Networks Trained with Noisy Labels

Abstract

Noisy labels are ubiquitous in real-world datasets, which poses a challenge for robustly training deep neural networks (DNNs) as DNNs usually have the high capacity to memorize the noisy labels. In this paper, we find that the test accuracy can be quantitatively characterized in terms of the noise ratio in datasets. In particular, the test accuracy is a quadratic function of the noise ratio in the case of symmetric noise, which explains the experimental findings previously published. Based on our analysis, we apply cross-validation to randomly split noisy datasets, which identifies most samples that have correct labels. Then we adopt the Co-teaching strategy which takes full advantage of the identified samples to train DNNs robustly against noisy labels. Compared with extensive state-of-the-art methods, our strategy consistently improves the generalization performance of DNNs under both synthetic and real-world training noise.

arXiv commentCorrespondence to: Guangyong Chen <[email protected]>

Cited by in corpus (41)

• DivideMix: Learning with Noisy Labels as Semi-supervised Learning
• Image Classification with Deep Learning in the Presence of Noisy Labels: A Survey
• Annotation-efficient deep learning for automatic medical image segmentation
• Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks
• A Survey of Label-noise Representation Learning: Past, Present and Future
• Contrast to Divide: Self-Supervised Pre-Training for Learning with Noisy Labels
• Identifying Mislabeled Data using the Area Under the Margin Ranking
• MoPro: Webly Supervised Learning with Momentum Prototypes
• Heteroskedastic and Imbalanced Deep Learning with Adaptive Regularization
• MetaLabelNet: Learning to Generate Soft-Labels from Noisy-Labels
• How does Early Stopping Help Generalization against Label Noise?
• Dirichlet-Based Prediction Calibration for Learning with Noisy Labels
• Towards Noise-resistant Object Detection with Noisy Annotations
• Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise
• Towards Understanding Deep Learning from Noisy Labels with Small-Loss Criterion
• Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels
• Self-Enhanced GNN: Improving Graph Neural Networks Using Model Outputs
• Coresets for Robust Training of Neural Networks against Noisy Labels
• Learning Fast Sample Re-weighting Without Reward Data
• Faster Meta Update Strategy for Noise-Robust Deep Learning
• Instance Correction for Learning with Open-set Noisy Labels
• Confidence Adaptive Regularization for Deep Learning with Noisy Labels
• Deep Learning with Label Differential Privacy
• Which Strategies Matter for Noisy Label Classification? Insight into Loss and Uncertainty
• A Meta Approach to Defend Noisy Labels by the Manifold Regularizer PSDR
• Learning with Noisy Labels by Efficient Transition Matrix Estimation to Combat Label Miscorrection
• NGC: A Unified Framework for Learning with Open-World Noisy Data
• Learning to Combat Noisy Labels via Classification Margins
• Co-matching: Combating Noisy Labels by Augmentation Anchoring
• Evaluating AI systems under uncertain ground truth: a case study in dermatology
• P-DIFF: Learning Classifier with Noisy Labels based on Probability Difference Distributions
• Combating Label Noise With A General Surrogate Model For Sample Selection
• Rethinking Noisy Label Models: Labeler-Dependent Noise with Adversarial Awareness
• Multi-Label Gold Asymmetric Loss Correction with Single-Label Regulators
• CoDiM: Learning with Noisy Labels via Contrastive Semi-Supervised Learning
• Improving Generalization of Deep Fault Detection Models in the Presence of Mislabeled Data
• INN: A Method Identifying Clean-annotated Samples via Consistency Effect in Deep Neural Networks
• Learning to Selectively Learn for Weakly-supervised Paraphrase Generation
• In-Loop Meta-Learning with Gradient-Alignment Reward
• Mitigating Memorization in Sample Selection for Learning with Noisy Labels
• One-shot Weakly-Supervised Segmentation in Medical Images