English

Jo-SRC: A Contrastive Approach for Combating Noisy Labels

Computer Vision and Pattern Recognition 2021-03-25 v1

Abstract

Due to the memorization effect in Deep Neural Networks (DNNs), training with noisy labels usually results in inferior model performance. Existing state-of-the-art methods primarily adopt a sample selection strategy, which selects small-loss samples for subsequent training. However, prior literature tends to perform sample selection within each mini-batch, neglecting the imbalance of noise ratios in different mini-batches. Moreover, valuable knowledge within high-loss samples is wasted. To this end, we propose a noise-robust approach named Jo-SRC (Joint Sample Selection and Model Regularization based on Consistency). Specifically, we train the network in a contrastive learning manner. Predictions from two different views of each sample are used to estimate its "likelihood" of being clean or out-of-distribution. Furthermore, we propose a joint loss to advance the model generalization performance by introducing consistency regularization. Extensive experiments have validated the superiority of our approach over existing state-of-the-art methods.

Keywords

Cite

@article{arxiv.2103.13029,
  title  = {Jo-SRC: A Contrastive Approach for Combating Noisy Labels},
  author = {Yazhou Yao and Zeren Sun and Chuanyi Zhang and Fumin Shen and Qi Wu and Jian Zhang and Zhenmin Tang},
  journal= {arXiv preprint arXiv:2103.13029},
  year   = {2021}
}

Comments

accepted by IEEE Conference on Computer Vision and Pattern Recognition, 2021

R2 v1 2026-06-24T00:30:18.925Z