K_Learning: A Meta-Control for a Satisficing Model of a Dynamic Environment

Madeleine Girard-Faugere

How shall learning deal with dynamical environments? This paper presents a new reinforcement learning scheme which allows to build and update a satisficing model of the environment in controlling the learning process: control what to learn and where through an explicit control of the quality of the model. We have applied this scheme to the Q_Learning algorithm, and tested the new obtained scheme (K_Learning) on a simulated superdistribution network. It provides better results in learning and adapting than the Q_Learning in static and dynamic environments.


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.