Robust Top-k Multiclass SVM for Visual Category Recognition

Classification problems with a large number of classes inevitably involve overlapping or similar classes. In such cases it seems reasonable to allow the learning algorithm to make mistakes on similar