Integrating Machine Learning into Meta-RaPS via Q Learning

DOI: 10.5176/2251-2179_ATAI33

Authors: Arif Arin and Ghaith Rabadi

Abstract:

A new metaheuristic, Meta-RaPS (Metaheuristic for Randomized Priority Search) creates high quality solutions for discrete optimization problems, and is classified as a memoryless metaheuristic. However, in many cases, it has been observed that memory and learning mechanisms can increase the effectiveness of the solution search process, and as a result the solution quality. Thus, it is proposed that incorporating memory/learning mechanisms into Meta-RaPS can help the algorithm produce better results. To incorporate learning mechanisms, Q learning, a type of machine learning, is introduced into Meta-RaPS. The 0-1 multidimensional knapsack problem will be used to evaluate the proposed algorithms.

Keywords: component; Metaheuristics, Meta-RaPS, Machine Learning, Q Learning