Adaptive regularization of the representation rank of value networks in deep reinforcement learning based on the implicit constraints imposed by the Bellman equation.
