Finite-Sample Convergence Rates for Q-Learning and Indirect Algorithms finite-sample convergence rates for q-learning and indirect algorithms
The Asymptotic Convergence-Rate of Q-learning the-asymptotic-convergence-rate-of-q-learning The asymptotic rate of convergence of Q-learning is Ο( 1/tR(1-γ) ), if R(1-γ)<0.5, where R=Pmin/Pmax, P is state-action occupation frequency. |Qt (x,a) − Q*(x,a)| < B/tR(1-γ) Convergence-rate is the difference between True value and Optimum value, i.e., the smaller it is, the faster convergence Q-learning is. We hope the Ο( 1/tR(1-γ) ) should… read more »
There is no excerpt because this is a protected post.
There is no excerpt because this is a protected post.