Hi T SHAO,

This is the policy gradient loss calculation. It is related to cross-entropy in the sense that we want to increase the log-likelihood of rewarding actions and decrease it for un-rewarding actions.