An end-to-end approach for the verification problem: learning the right distance

In this contribution, we augment the metric learning setting by introducing a\nparametric pseudo-distance, trained jointly with the encoder. Several\ninterpretations are thus drawn for the learned distance-like model's output. We\nfirst show it approximates a likelihood ratio which can be used for hypothesis\ntests, and that it further induces a large divergence across the joint\ndistributions of pairs of examples from the same and from different classes.\nEvaluation is performed under the verification setting consisting of\ndetermining whether sets of examples belong to the same class, even if such\nclasses are novel and were never presented to the model during training.\nEmpirical evaluation shows such method defines an end-to-end approach for the\nverification problem, able to attain better performance than simple scorers\nsuch as those based on cosine similarity and further outperforming widely used\ndownstream classifiers. We further observe training is much simplified under\nthe proposed approach compared to metric learning with actual distances,\nrequiring no complex scheme to harvest pairs of examples.\n