Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Jo-SRC: A Contrastive Approach for Combating Noisy Labels

About

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.

Yazhou Yao, Zeren Sun, Chuanyi Zhang, Fumin Shen, Qi Wu, Jian Zhang, Zhenmin Tang• 2021

Related benchmarks

TaskDatasetResultRank
Image ClassificationCompCars Web (test)
Top-1 Acc88.13
33
Image ClassificationCIFAR-100 Sym-20% (test)
Accuracy58.15
33
Image ClassificationCIFAR-100 Sym-50% (test)
Accuracy51.26
32
Image ClassificationWeb-Bird (test)
Accuracy81.22
26
Image ClassificationWeb-Aircraft (test)
Test Accuracy82.73
26
Image ClassificationCIFAR80N-O Sym-50% (test)
Accuracy58.51
17
Image ClassificationCIFAR80N-O Sym-20% (test)
Accuracy65.83
17
Image ClassificationCIFAR80N-O Sym-80% (test)
Test Acc29.76
17
Image ClassificationCIFAR80N-O Asym-40% (test)
Accuracy53.03
17
ClassificationCIFAR-100 Symmetry-80% noise (test)
Test Accuracy23.8
17
Showing 10 of 14 rows

Other info

Follow for update