Noise-Robust Bidirectional Learning with Dynamic Sample Reweighting

Deep neural networks trained with standard crossentropy loss are more prone to memorize noisy labels, which degrades their performance. Negative learning using complementary labels is more robust when noisy labels intervene but with an extremely slow model convergence speed. In this paper, we first introduce a bidirectional learning scheme, where positive learning ensures convergence speed while negative learning robustly copes with label noise. Further, a dynamic sample reweighting strategy is proposed to globally weaken the effect of noise-labeled samples by exploiting the excellent discriminatory ability of negative learning on the sample probability distribution. In ad-dition, we combine self-distillation to further improve the model performance. The code is available at https://github.com/chenchenzong/BLDR.

• Publication year

  2022

• Venue

  arXiv.org

• Publication date

  2022-09-03

• Fields of study

  Computer Science

• Identifiers
  DOI 10.48550/arXiv.2209.01334arXiv 2209.01334
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Robust Training under Label Noise by Over-parameterization

  2022cited by this paper
• Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations

  2021cited by this paper
• DualGraph: A graph-based method for reasoning about label noise

  2021cited by this paper
• Learning from Noisy Labels with Complementary Loss Functions

  2021cited by this paper
• Can Cross Entropy Loss Be Robust to Label Noise?

  2020cited by this paper
• Be Your Own Teacher: Improve the Performance of Convolutional Neural Networks via Self Distillation

  2019cited by this paper
• How does Disagreement Help Generalization against Label Corruption?

  2019cited by this paper
• Curriculum Loss: Robust Learning and Generalization against Label Corruption

  2019cited by this paper
• Symmetric Cross Entropy for Robust Learning With Noisy Labels

  2019cited by this paper
• Co-teaching: Robust training of deep neural networks with extremely noisy labels

  2018cited by this paper
• Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels

  2018cited by this paper
• MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels

  2017cited by this paper
• Robust Loss Functions under Label Noise for Deep Neural Networks

  2017cited by this paper
• Active Bias: Training More Accurate Neural Networks by Emphasizing High Variance Samples

  2017cited by this paper
• Training deep neural-networks based on unreliable labels

  2016cited by this paper
• Webly Supervised Learning of Convolutional Networks

  2015cited by this paper
• Learning from massive noisy labeled data for image classification

  2015cited by this paper
• Ieee Transactions on Pattern Analysis and Machine Intelligence Classification with Noisy Labels by Importance Reweighting

  year unknowncited by this paper

• Dirichlet-Based Prediction Calibration for Learning with Noisy Labels

  2024cites this paper
• Bidirectional Uncertainty-Based Active Learning for Open Set Annotation

  2024cites this paper