On Generalized Bellman Equations and Temporal-Difference Learning
This work addresses a specific bottleneck in reinforcement learning for policy evaluation, offering a more flexible method that is incremental in nature.
The paper tackles the high variance issue in off-policy temporal-difference learning by proposing a new scheme for setting λ-parameters based on generalized Bellman equations, which allows for larger λ values with bounded traces and is proven to ensure ergodicity under nonrestrictive conditions.
We consider off-policy temporal-difference (TD) learning in discounted Markov decision processes, where the goal is to evaluate a policy in a model-free way by using observations of a state process generated without executing the policy. To curb the high variance issue in off-policy TD learning, we propose a new scheme of setting the $λ$-parameters of TD, based on generalized Bellman equations. Our scheme is to set $λ$ according to the eligibility trace iterates calculated in TD, thereby easily keeping these traces in a desired bounded range. Compared with prior work, this scheme is more direct and flexible, and allows much larger $λ$ values for off-policy TD learning with bounded traces. As to its soundness, using Markov chain theory, we prove the ergodicity of the joint state-trace process under nonrestrictive conditions, and we show that associated with our scheme is a generalized Bellman equation (for the policy to be evaluated) that depends on both the evolution of $λ$ and the unique invariant probability measure of the state-trace process. These results not only lead immediately to a characterization of the convergence behavior of least-squares based implementation of our scheme, but also prepare the ground for further analysis of gradient-based implementations.