NFDRL

Normalizing flows enable parameter-efficient distributional RL

Researchers have developed NFDRL, a novel architecture for distributional reinforcement learning that utilizes continuous normalizing flows to model return distributions. This approach offers a more parameter-efficient …