Faster Meta Update Strategy for Noise-Robust Deep Learning

30 Apr 2021  ยท  Youjiang Xu, Linchao Zhu, Lu Jiang, Yi Yang ยท

It has been shown that deep neural networks are prone to overfitting on biased training data. Towards addressing this issue, meta-learning employs a meta model for correcting the training bias. Despite the promising performances, super slow training is currently the bottleneck in the meta learning approaches. In this paper, we introduce a novel Faster Meta Update Strategy (FaMUS) to replace the most expensive step in the meta gradient computation with a faster layer-wise approximation. We empirically find that FaMUS yields not only a reasonably accurate but also a low-variance approximation of the meta gradient. We conduct extensive experiments to verify the proposed method on two tasks. We show our method is able to save two-thirds of the training time while still maintaining the comparable or achieving even better generalization performance. In particular, our method achieves the state-of-the-art performance on both synthetic and realistic noisy labels, and obtains promising performance on long-tailed recognition on standard benchmarks.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Classification CIFAR-100, 40% Symmetric Noise FaMUS Percentage correct 75.91 # 1
Image Classification CIFAR-100, 40% Symmetric Noise MentorMix Percentage correct 71.3 # 2
Image Classification CIFAR-100, 60% Symmetric Noise MentorMix Percentage correct 64.6 # 1
Image Classification CIFAR-10, 40% Symmetric Noise FaMUS Percentage correct 95.37 # 1
Image Classification CIFAR-10, 40% Symmetric Noise MentorMix Percentage correct 94.2 # 2
Image Classification CIFAR-10, 60% Symmetric Noise MentorMix Percentage correct 91.3 # 1
Image Classification CIFAR-10, 60% Symmetric Noise FaMUS Percentage correct 26.42 # 2
Image Classification mini WebVision 1.0 FaMUS Top-1 Accuracy 79.4 # 14
Top-5 Accuracy 92.80 # 9
ImageNet Top-1 Accuracy 77 # 6
ImageNet Top-5 Accuracy 92.76 # 10
Image Classification Red MiniImageNet 20% label noise FaMUS Accuracy 51.42 # 5
Image Classification Red MiniImageNet 40% label noise FaMUS Accuracy 48.06 # 5
Image Classification Red MiniImageNet 60% label noise FaMUS Accuracy 45.1 # 4
Image Classification Red MiniImageNet 80% label noise FaMUS Accuracy 35.5 # 5

Methods


No methods listed for this paper. Add relevant methods here