Notebook
Concepts I am working to understand, written out.
Folders
FldConceptLnkUpd.
relates · requiresdrag · scroll · dbl-click
Concepts I am working to understand, written out.
Concepts I am working to understand, written out.
updated 2d ago · 1 links · 2,388 words
Differentiate the expected return with respect to the policy parameters without differentiating the state distribution. Two proofs, then REINFORCE and why baselines are free.
links: TD learning
requires: MDPs · Bellman
path: MDPs → Bellman → Policy gradient