Convex Kernelized Sorting

Kernelized sorting is a method for aligning objects across two domains by considering within-domain similarity, without need to specify cross-domain similarity measure. In this paper we present the Convex Sorting where, unlike in previous approaches, object matching formulated as convex optimization problem, leading simpler and global optimum solution. Our outputs soft alignments between objects, which can be used rank best matches each object, or visualize verify correct choice of kernel. It also allows computing hard one-to-one solving resulting Linear Assignment Problem. Experiments on number tasks show strength proposed method, consistently achieves higher accuracy than existing methods.