Tweeted By @mblondel_ml

Our paper, in which we use entropy-regularized optimal transport to derive a cost-sensitive generalization of the logistic loss, and corresponding softmax. Our loss recovers the usual logistic loss when infinitely penalizing misclassification errors. https://t.co/EdVHIdIvtS

— Mathieu Blondel (@mblondel_ml) May 16, 2019