Pseudo-Feature Generation for Imbalanced Data Analysis in Deep Learning

All the real data are imbalanced, which makes the learning difficult. We generate pseudo-feature by multivariate probability distributions obtained from feature maps in the layer of trained deep neural networks. Then, we augment the data of minor classes by the pseudo-feature in order to overcome imbalanced data problems. Because all the wild data are imbalanced, the proposed method may improve the capability of deep learning in broad range of problems. We synthesis imbalanced data from ImageNet, and our proposed method improves accuracy from 42.96% to 46.39%.