Learning similarity measures from pairwise constraints with neural networks

This paper presents a novel neural network model, called Similarity Neural Network (SNN), designed to learn similarity measures for pairs of patterns exploiting binary supervision. The model guarantees to compute a non negative and symmetric measure, and shows good generalization capabilities even if a small set of supervised examples is used for training. The approximation capabilities of the proposed model are also investigated. Moreover, the experiments carried out on some benchmark datasets show that SNNs almost always outperform other similarity learning methods proposed in the literature.