Stochastic Methods: Implementation of Improved Genetic Algorithm for Shape Recognition Purposes

Considering parameter optimization tasks, a fundamental advantage of the genetic algorithm lies in its robustness, i.e. that it is capable of providing a solution of good quality even when the error surface is unknown or of extreme shape. Its disadvantage is its demand for large system memory and computational capacity, as well as the fact that because of its stochastic basis it is not to be used in most real-time applications. However, if the adjustment or training is rarely to be made, or the comparison of different parameter vectors can be performed rapidly, it is without doubt advantageous to apply. The author of this paper has examined and modified the standard methods to improve their performance. One of these was the implementation of the Special Linear Combination crossover method. Another one was the Maximum Fitness Guided Migration method. Both methods will be discussed in detail in this paper. The author has implemented the advanced methods in a computer program, which was then tested on a shape recognition task