We propose a novel transfer learning framework for regression where outputs are probability distributions residing in the Wasserstein space.