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


Jun 13, 2019
Pengfei Chen, Benben Liao, Guangyong Chen, Shengyu Zhang


 Click To Get Model/Code


Noisy labels are ubiquitous in real-world datasets, which poses a challenge for robustly training deep neural networks (DNNs) since DNNs can easily overfit to the noisy labels. Most recent efforts have been devoted to defending noisy labels by discarding noisy samples from the training set or assigning weights to training samples, where the weight associated with a noisy sample is expected to be small. Thereby, these previous efforts result in a waste of samples, especially those assigned with small weights. The input $x$ is always useful regardless of whether its observed label $y$ is clean. To make full use of all samples, we introduce a manifold regularizer, named as Paired Softmax Divergence Regularization (PSDR), to penalize the Kullback-Leibler (KL) divergence between softmax outputs of similar inputs. In particular, similar inputs can be effectively generated by data augmentation. PSDR can be easily implemented on any type of DNNs to improve the robustness against noisy labels. As empirically demonstrated on benchmark datasets, our PSDR impressively improve state-of-the-art results by a significant margin.

* Correspondence to: Guangyong Chen <gycchen@tencent.com> 

 Click To Get Model/Code


Share this with someone who'll enjoy it:

   Click for Paper Source