期刊名称：Indian Journal of Computer Science and Engineering
印刷版ISSN：2231-3850
电子版ISSN：0976-5166
出版年度：2017
卷号：8
期号：5
页码：601-612
出版社：Engg Journals Publications
摘要：In this paper, we exhibit new convergent Q learning algorithm that consolidate components ofpolicy iteration and classical Q learning/esteem iteration to effectively learn and control arrangements fora dynamic load adjusting situations utilizing reinforcement learning techniques. The model is preparedwith a variation of memory optimization strategy for dynamic load adjusting recreation on multi-coreprocessors making utilization of a machine learning approach, whose inputs areseveral time consumingcomputational processes and whose yield are time situated wrapper towards adjusting the computationaland correspondence stack individually with an evaluation future rewards. The primary point ofpreference over this Q learning methodology is lower overhead; as most iteration doesn’t require aminimization over all controls, in the context of modified policy iteration.We apply our technique tomulti-core Q-learning way to make an algorithm which is a combination of the results from enhancedload and effective memory utilization on multiple cores. This technique gives a learning situation inhandling computational load with no modification of the architecture resources or learningalgorithm.These executions conquer a portion of the conventional convergence difficulties of offbeatmodified policy iteration particularly in handling circumstances like that of multicore processors, andgive policy iteration-like option Q-learning plans with as dependable convergence as classical Q learning.
