Use machine learning to organize a collection of objects (e.g. faces of actors) so as to make them interactively searchable via similarity queries. The approach combines concepts from representation learning, reinforcement learning, discrete-choice theory and bayesian inference.