A Framework for Memetic Algorithms

Many optimization problems are fundamentally hard. Essentially, a ‘hard’ problem is one for which we cannot guarantee to find the best solution in a reasonable amount of time. In practice, however, the quest to solve hard problems is not quite so hopeless as this definition suggests. This is due to the use of approximate methods. An approximate method is an algorithm that we use to try to find solutions to hard optimization problems, and which runs quickly, but which gives no guarantee that the solution it will find is the best one. The existing, successful methods in approximate optimization fall into two broad classes: local search, and population-based search. There are many population-based optimization algorithms and various ways to handle the optimization issues. In this thesis, a special emphasis has been given to Memetic Algorithms introduced by P. Moscato, which represents one of the most successful emerging ideas in the ongoing research effort to understand population based and local search algorithms. Based on a general template forMemetic Algorithms, an object-oriented framework was developed in C++ to experiment with this approximate method. When the user wants to use this framework for a special NP-Hard optimization problem, he needs only to define suitable derived classes, which implement the virtual functions of the abstract classes, and supply the problem specific details. Using and instantiating this Memetic Algorithms framework, two applications were encoded for two well-known combinatorial optimization problems, Vertex Cover and Independent Set. Comparisons of effectiveness were made between several approximate methods.