Raj Ghugare

We take a probabilistic view of decision making tasks, and show that normalizing flows offer many properties which make them highly performant models in imitation learning, offline, goal conditioned and unsupervised RL.