Policy gradient method: Difference between revisions

Revision as of 02:45, 13 April 2025

Revision as of 10:07, 15 May 2025

Line 124:

The algorithm uses the modified gradient estimator<math display="block">g_t \leftarrow

\frac 1N \sum_{k=1}^N \left[\sum_{j\in 0:T} \nabla_{\theta_t}\ln\pi_\theta(A_{j,k}| S_{ij,k})\left(\sum_{i \in j:T} (\gamma^i R_{i,k}) - b_t(S_{j,k})\right) \right]</math> and the original REINFORCE algorithm is the special case where <math>b_t = 0</math>.

=== Actor-critic methods ===