[UCLA RL-LLM] Chapter 1.3: Deep policy gradient methods (A3C)

Ernest Ryu · 76:58

Policy-gradient methods maximize expected discounted return of a neural policy \( \pi\theta \) by stochastic gradient ascent; the naive score-function estimator is unbiased but far too high-variance, so practical algo...

Read the full summary on tuber

Redirecting...