2017 (modified: 11 Nov 2022)ICML 2017Readers: Everyone
Abstract:We create a new online reduction of multiclass classification to binary classification for which training and prediction time scale logarithmically with the number of classes. We show that several ...