Robust training with ensemble consensus

Jisoo Lee, Sae-Young Chung

Keywords: Annotation noise, Noisy label, Robustness, Ensemble, Perturbation

TL;DR: This work presents a method to robustly train neural networks by using ensemble in the presence of label noise.

Abstract: Since deep neural networks are over-parameterized, they can memorize noisy examples. We address such a memorization issue in the presence of label noise. From the fact that deep neural networks cannot generalize to neighborhoods of memorized features, we hypothesize that noisy examples do not consistently incur small losses on the network under a certain perturbation. Based on this, we propose a novel training method called Learning with Ensemble Consensus (LEC) that prevents overfitting to noisy examples by removing them based on the consensus of an ensemble of perturbed networks. One of the proposed LECs, LTEC outperforms the current state-of-the-art methods on noisy MNIST, CIFAR-10, and CIFAR-100 in an efficient manner.

Code: https://github.com/jisoolee0123/Robust-training-with-ensemble-consensus

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:1910.09792/code)