Tailoring Mixup to Data for Calibration (2311.01434v2)

Abstract: Among all data augmentation techniques proposed so far, linear interpolation of training samples, also called Mixup, has found to be effective for a large panel of applications. Along with improved performance, Mixup is also a good technique for improving calibration and predictive uncertainty. However, mixing data carelessly can lead to manifold intrusion, i.e., conflicts between the synthetic labels assigned and the true label distributions, which can deteriorate calibration. In this work, we argue that the likelihood of manifold intrusion increases with the distance between data to mix. To this end, we propose to dynamically change the underlying distributions of interpolation coefficients depending on the similarity between samples to mix, and define a flexible framework to do so without losing in diversity. We provide extensive experiments for classification and regression tasks, showing that our proposed method improves performance and calibration of models, while being much more efficient. The code for our work is available at https://github.com/qbouniot/sim_kernel_mixup.