Finite Time Analysis of Temporal Difference Learning for Mean-Variance in a Discounted MDP

Motivated by risk-sensitive reinforcement learning scenarios, we consider the problem of policy evaluation for variance in a discounted reward Markov decision process (MDP). For this problem, a temporal difference (TD) type learning algorithm with linear function approximation (LFA) exists in the literature, though only asymptotic guarantees are available for this algorithm. We derive finite sample bounds that hold (i) in the mean-squared sense; and (ii) with high probability, when tail iterate averaging is employed with/without regularization. Our bounds exhibit exponential decay for the initial error, while the overall bound is O (1/t), where t is the number of update iterations of the TD algorithm. Further, the bound for the regularized TD variant is for a universal step size. Our bounds open avenues for analysis of actor-critic algorithms for mean-variance optimization in a discounted MDP.