Improving GA search reliability using maximal hyper-rectangle analysis

In Genetic algorithms it is not easy to evaluate the confidence level in whether a GA run may have missed a complete area of good points, and whether the global optimum was found. We accept this but hope to add some degree of confidence in our results by showing that no large gaps were left unvisited in the search space. This can be achieved to some extent by inserting new individuals in big empty spaces. However it is not easy to find the biggest empty spaces, particularly in multi-dimensional problems. For a GA problem, however, it is not necessary to find the exact biggest empty spaces; a sufficiently large empty space is good enough to insert new individuals. In this paper, we present a method to find a sufficiently large empty Hyper-Rectangle for new individual insertion in a GA while keeping the computational complexity as a polynomial function. Its merit is demonstrated in several domains. Categories and Subject Descriptors I.2.8 [Computing Methodologies]: Artificial Intelligen...

• Type: Conference
• Year: 2005
• Where: GECCO
• Authors: Chongshan Zhang, Khaled Rasheed
• Keywords: Empty Space, GECCO 2005, Genetic Algorithms, Sufficiently Large