Learning to Compare using Operator-Valued Large-Margin Classifiers

The proposed method uses homonymous and heteronymous examplepairs to train a linear preprocessor on a kernel-induced Hilbert space. The algorithm seeks to optimize the expected performance of elementa