Home // ICAS 2012, The Eighth International Conference on Autonomic and Autonomous Systems // View article

Augmenting Reinforcement Learning Feedback with Prediction for Autonomic Management

Authors:
Khandakar Rashed Ahmed
Raphael Bahati
Michael Bauer

Keywords: autonomic management, prediction, policies, reinforcement learning

Abstract:
Autonomic management depends on a feedback loop between the managed system and the autonomic manager. Adding a learning component to the autonomic manager introduces a second feedback loop – between the manager and the learning agent. In this paper, we describe a policy-based autonomic manager that makes use of a reinforcement learning agent. The reinforcement learning model is based on a state-transition model formed from an active set of policies and the actions of the manager. Based upon this model, this paper describes approaches for prediction of potential policy violations and examines the accuracy of the prediction approaches. Experimental results show that a prediction approach based on the likelihood of a violation performs better than a non-prediction approach and has a positive impact on avoiding policy violations.

Pages: 74 to 79

Copyright: Copyright (c) IARIA, 2012

Publication date: March 25, 2012

Published in: conference

ISSN: 2308-3913

ISBN: 978-1-61208-187-8

Location: St. Maarten, The Netherlands Antilles

Dates: from March 25, 2012 to March 30, 2012