Data Interpolating Prediction: Alternative Interpretation of Mixup
This work addresses a specific issue in data augmentation for classification tasks, offering an incremental improvement over Mixup.
The paper tackles the problem of Mixup's ineffectiveness due to a gap between augmented training samples and original test samples, which can hinder learning optimal decision boundaries and increase generalization error. The proposed Data Interpolating Prediction (DIP) framework encapsulates sample-mixing in the classifier's hypothesis class to treat train and test samples equally, showing reduced Rademacher complexity and empirical outperformance over existing Mixup.
Data augmentation by mixing samples, such as Mixup, has widely been used typically for classification tasks. However, this strategy is not always effective due to the gap between augmented samples for training and original samples for testing. This gap may prevent a classifier from learning the optimal decision boundary and increase the generalization error. To overcome this problem, we propose an alternative framework called Data Interpolating Prediction (DIP). Unlike common data augmentations, we encapsulate the sample-mixing process in the hypothesis class of a classifier so that train and test samples are treated equally. We derive the generalization bound and show that DIP helps to reduce the original Rademacher complexity. Also, we empirically demonstrate that DIP can outperform existing Mixup.