Robust and Efficient Transfer Learning with Hidden Parameter Markov Decision Processes

We introduce a new formulation of the Hidden Parameter Markov Decision Process (HiP-MDP), a framework for modeling families of related tasks using low-dimensional latent embeddings.