Relative Comparison Kernel Learning with Auxiliary Kernels

In this work we consider the problem of learning a positive semidefinite kernel matrix from relative comparisons of the form: "object A is more similar to object B than it is to C", where comparisons are given by humans. Existing solutions to this problem assume many comparisons are provided to learn a meaningful kernel. However, this can be considered unrealistic for many real-world tasks since a large amount of human input is often costly or difficult to obtain. Because of this, only a limited number of these comparisons may be provided. We propose a new kernel learning approach that supplements the few relative comparisons with "auxiliary" kernels built from more easily extractable features in order to learn a kernel that more completely models the notion of similarity gained from human feedback. Our proposed formulation is a convex optimization problem that adds only minor overhead to methods that use no auxiliary information. Empirical results show that in the presence of few training relative comparisons, our method can learn kernels that generalize to more out-of-sample comparisons than methods that do not utilize auxiliary information, as well as similar metric learning methods.