Provable Benefit of Cutout and CutMix for Feature Learning
This provides theoretical insights into why CutMix outperforms other methods in vision tasks, though it is incremental as it builds on existing augmentation techniques.
The paper tackles the lack of theoretical understanding of patch-level data augmentation techniques like Cutout and CutMix by analyzing their effects on two-layer neural networks using a feature-noise data model. The result shows that Cutout learns low-frequency features missed by vanilla training, CutMix learns even rarer features, and CutMix achieves the highest test accuracy.
Patch-level data augmentation techniques such as Cutout and CutMix have demonstrated significant efficacy in enhancing the performance of vision tasks. However, a comprehensive theoretical understanding of these methods remains elusive. In this paper, we study two-layer neural networks trained using three distinct methods: vanilla training without augmentation, Cutout training, and CutMix training. Our analysis focuses on a feature-noise data model, which consists of several label-dependent features of varying rarity and label-independent noises of differing strengths. Our theorems demonstrate that Cutout training can learn low-frequency features that vanilla training cannot, while CutMix training can learn even rarer features that Cutout cannot capture. From this, we establish that CutMix yields the highest test accuracy among the three. Our novel analysis reveals that CutMix training makes the network learn all features and noise vectors "evenly" regardless of the rarity and strength, which provides an interesting insight into understanding patch-level augmentation.