Bayesian Meta-Learning for the Few-Shot Setting via Deep Kernels

As a result we can learn a kernel that transfers to new tasks; we call this Deep Kernel Transfer (DKT) .