Information Technology Reference
In-Depth Information
4
Algorithm Selection:
From Meta-Learning
to Hyper-Heuristics
Laura Cruz-Reyes
1
, Claudia Gómez-Santillán
1
,
Joaquín Pérez-Ortega
2
, Vanesa Landero
3
,
Marcela Quiroz
1
and Alberto Ochoa
4
1
Instituto Tecnológico de Cd. Madero
2
Centro Nacional de Investigación y Desarrollo Tecnológico
3
Universidad Politécnica de Nuevo León
4
Universidad de Ciudad Juárez
México
1. Introduction
In order for a company to be competitive, an indispensable requirement is the efficient
management of its resources. As a result derives a lot of complex optimization problems
that need to be solved with high-performance computing tools. In addition, due to the
complexity of these problems, it is considered that the most promising approach is the
solution with approximate algorithms; highlighting the heuristic optimizers. Within this
category are the basic heuristics that are experience-based techniques and the metaheuristic
algorithms that are inspired by natural or artificial optimization processes.
A variety of approximate algorithms, which had shown satisfactory performance in
optimization problems, had been proposed in the literature. However, there is not an
algorithm that performs better for all possible situations, given the amount of available
strategies, is necessary to select the one who adapts better to the problem. An important
point is to know which strategy is the best for the problem and why it is better.
The chapter begins with the formal definition of the Algorithm Selection Problem (ASP),
since its initial formulation. The following section describes examples of "Intelligent
Systems" that use a strategy of algorithm selection. After that, we present a review of
the literature related to the ASP solution. Section four presents the proposals of our
research group for the ASP solution; they are based on machine learning, neural network
and hyper-heuristics. Besides, the section presents experimental results in order to
conclude about the advantages and disadvantages of each approach. Due to a fully
automated solution to ASP is an undecidable problem, Section Five reviews other less
rigid approach which combines intelligently different strategies: The Hybrid Systems of
Metaheuristics.
