Modifikace metody Q-učení z diskrétní na spojitou

Q-Learning: From Discrete to Continuous Representation

článek v časopise - ostatní, Jost

en

Q-learning standard algorithm is restricted by using discrete states and actions. In this case Q-function is usually represented as a discrete table of Q-values. Conversion of continuous variables to adequate discrete variables evokes some problems. Problems can be avoided if the continuous algorithm of Q-learning is used. In this paper we discus method, which is used to convert discrete to continuous algorithm. The method used suitable approximator to replace the discrete table. We choose local approximator called Locally Weighted Regression (LWR) (Atketson &Moore & Shaal, 1996) from the group of memory based approximators.

Tento článek pojednává o způsobu převedení standardní metody Q-učení, která je pouze diskrétní, na spojitou. K tomuto účelu je použit jednoduchý lokální aproximátor Lokálně vážená regrese (LVR). Tento aproximátor slouží k převedení diskrétní tabulky Q-hodnot na spojitou Q-funkci.

Q-learning standard algorithm is restricted by using discrete states and actions. In this case Q-function is usually represented as a discrete table of Q-values. Conversion of continuous variables to adequate discrete variables evokes some problems. Problems can be avoided if the continuous algorithm of Q-learning is used. In this paper we discus method, which is used to convert discrete to continuous algorithm. The method used suitable approximator to replace the discrete table. We choose local approximator called Locally Weighted Regression (LWR) (Atketson &Moore & Shaal, 1996) from the group of memory based approximators.

Q-learning, Machine learning, Locally Weighted Regression

2004

23.08.2004

Warsaw, Poland

0033-2089

Elektronika

XVL

8

3