Tesis "Strategies for Improving the Efficiency of Evolutionary Algorithms in High Dimensional Optimization"
Alumno: Pedro Octavio Reta Aguilar
Asesor: Ricardo Landa Becerra
Sinodales: Dr. César Torres Huitzil, Dr. Gregorio Toscano Pulido
There are real world optimization problems that are particularly hard to solve. One of the main reasons of the increasing in difficulty is because of the high number of decision variables involved. It is well known that the size of the search space of an optimization problem increases exponentially as the number of variables rises. This effect is known in the literature as the curse of dimensionality.
Evolutionary Computing is a popular tool for solving optimization problems. However, it has not been but until recent years that alternatives for improving its efficiency on high dimensional problems have been explored. There are several existing proposals for improving the general performance of Evolutionary Algorithms; among them, it is the use of Local Search mechanisms, or the exploitation of the structure of the problem itself.
In this thesis work we propose to incorporate separability and Local Search concepts into an Evolutionary Algorithm, following the Cooperative Coevolution model. This model is based on finding a solution to a complex problem by decomposing it into subproblems of lower complexity, and then solving each subproblem independently.
Experimentation using a well-known benchmark in the literature show that our proposal achieves quasi-optimal estimation of the separability between variables of the problem for an efficient decomposition. Moreover, a statistical significance analysis favors our approach (with respect to quality solution) when compared against three of the most outstanding proposals of the state of the art up to the date.