Currently, we use the SparesNetworks implemented in LBjava directly for the multi-class classification. To have more flexibility in the weight updates for the multi-class setting, specially for the joint training we better to do implement multi-class setting based on the binary classifiers and constraints.
Currently, we use the SparesNetworks implemented in LBjava directly for the multi-class classification. To have more flexibility in the weight updates for the multi-class setting, specially for the joint training we better to do implement multi-class setting based on the binary classifiers and constraints.