Feature-Learning from Pairs of Examples in Collections of Supervised Learning Tasks

We present an algorithm which uses example pairs of equal or unequal class labels to select a projection on a kernel-induced Hilbert space. A representation of .nite dimensional projections as bounded