Temporal-difference learning with nonlinear function approximation: lazy training and mean field regimes
This provides theoretical guarantees for TD learning in neural networks, addressing stability issues in reinforcement learning, but it is incremental as it builds on existing scaling regimes.
The paper tackles the convergence of Temporal-Difference learning with nonlinear function approximation by analyzing lazy training and mean-field regimes, proving exponential convergence to minimizers in both under- and over-parametrized settings and showing that all fixed points are global minimizers in the mean-field case.
We discuss the approximation of the value function for infinite-horizon discounted Markov Reward Processes (MRP) with nonlinear functions trained with the Temporal-Difference (TD) learning algorithm. We first consider this problem under a certain scaling of the approximating function, leading to a regime called lazy training. In this regime, the parameters of the model vary only slightly during the learning process, a feature that has recently been observed in the training of neural networks, where the scaling we study arises naturally, implicit in the initialization of their parameters. Both in the under- and over-parametrized frameworks, we prove exponential convergence to local, respectively global minimizers of the above algorithm in the lazy training regime. We then compare this scaling of the parameters to the mean-field regime, where the approximately linear behavior of the model is lost. Under this alternative scaling we prove that all fixed points of the dynamics in parameter space are global minimizers. We finally give examples of our convergence results in the case of models that diverge if trained with non-lazy TD learning, and in the case of neural networks.