Pascal Delange

Local data augmentation, not just class rebalancing, is the key to improving imbalanced binary classification recall, with anchor points near the decision boundary and diffusion kernels yielding state-of-the-art results.