Concurrent Credit Assignment for Data-efficient Reinforcement Learning

The capacity to widely sample the state and action spaces is a key ingredient toward building effective reinforcement learning algorithms. The method presented in this paper relies on an occupancy model, that is the empirical distribution of the states encountered by the agent under a given policy, that is its "domain of operation". Then, under a uniform occupancy prior assumption, an evidence lower bound on the parameters of the policy provides a way to express a balance between two concurrent tendencies, namely the widening of the occupancy space and the maximization of the rewards, reminding of the classical exploration/exploitation trade-off. During training, both the policy and the occupancy model are updated as the exploration progresses, and that new states are undisclosed during the course of the training. Implemented on an actor-critic off-policy on classic continuous action benchmarks, this approach is shown to provide significant increase in the sampling efficacy, that is reflected in a reduced training time and higher returns, in both the dense and the sparse rewards cases.