On the Statistical Benefits of Temporal Difference Learning
This provides a theoretical foundation for understanding when TD learning offers statistical benefits in reinforcement learning, which is incremental but clarifies long-standing intuitions.
The paper tackles the problem of quantifying the statistical advantages of temporal difference (TD) learning over direct estimation for value function approximation in finite state Markov chains, showing that TD can reduce mean-squared error by a percentage characterized by an inverse trajectory pooling coefficient and improve error bounds for value differences based on trajectory crossing time.
Given a dataset on actions and resulting long-term rewards, a direct estimation approach fits value functions that minimize prediction error on the training data. Temporal difference learning (TD) methods instead fit value functions by minimizing the degree of temporal inconsistency between estimates made at successive time-steps. Focusing on finite state Markov chains, we provide a crisp asymptotic theory of the statistical advantages of this approach. First, we show that an intuitive inverse trajectory pooling coefficient completely characterizes the percent reduction in mean-squared error of value estimates. Depending on problem structure, the reduction could be enormous or nonexistent. Next, we prove that there can be dramatic improvements in estimates of the difference in value-to-go for two states: TD's errors are bounded in terms of a novel measure - the problem's trajectory crossing time - which can be much smaller than the problem's time horizon.