Consistency regularization is a commonly-used technique for semi-supervised and self-supervised learning. It is an auxiliary objective function that encourages the prediction of the network to be similar in the vicinity of the observed training samples. Hendrycks et al. (2020) have recently shown such regularization naturally brings test-time robustness to corrupted data and helps with calibration. This paper empirically studies the relevance of consistency regularization for training-time robustness to noisy labels. First, we make two interesting and useful observations regarding the consistency of networks trained with the standard cross entropy loss on noisy datasets which are: (i) networks trained on noisy data have lower consistency than those trained on clean data, and(ii) the consistency reduces more significantly around noisy-labelled training data points than correctly-labelled ones. Then, we show that a simple loss function that encourages consistency improves the robustness of the models to label noise on both synthetic (CIFAR-10, CIFAR-100) and real-world (WebVision) noise as well as different noise rates and types and achieves state-of-the-art results.
翻译:一致性正规化是半监督和自我监督学习的常用技术,是一种辅助性目标功能,鼓励预测网络在观测到的培训样本附近类似。Hendrycks等人(2020年)最近表明,这种正规化自然地为腐败数据带来了测试-时间的稳健性,并有助于校准。本文以经验方式研究了培训-时间稳健性在培训-时间稳健性与吵闹标签之间的一致性的相关性。首先,我们就经过培训的网络与噪音数据集的标准交叉孔虫损失的一致性问题提出了两项有趣和有益的意见,即:(一) 热度数据培训的网络比清洁数据培训的网络的一致性低,以及(二) 噪音标签化培训数据点的一致性比正确标签点要明显减少。然后,我们表明,鼓励一致性的简单损失功能能够提高模型在合成(CIFAR-10、CIFAR-100)和真实世界(WebVisionion)上标注噪音的稳健性。噪音以及不同噪音率和类型以及实现状态的结果。