Performance Evaluation of TD-Learning Methods for Bandwidth Provisioning

Abstract

Q-learning and SARSA are two methods of TDlearning. Researchers interested in this field proposed the Eligibility concept in order to speed up Q-learning and SARSA. They proved their claim by running the algorithms in a static environment. Authors of this paper have used Q-learning, SARSA and also their eligibility versions for bandwidth provisioning in DiffServ networks that is an absolutely dynamic environment. Performance of these methods in this absolutely dynamic environment is evaluated.