Evolution of Virtual Prey using a Virtual Predator

Evolution of Virtual Prey using a Virtual Predator

Can evolution happen on a computer is it possible for Images to evolve? This paper tries to show that Mother Nature has got it right and evolution is the future for programs. How many times will a new release come to solve my previous problem and in the process create new problems! Evolution can happen on a computer is very clear in this work titled Evolution of a Virtual Prey using a Virtual Predator. The Virtual Prey can evolve its skin color based on an Artificial Intelligent Virtual Predator. Monochrome patterns have evolved very successfully. Some repeatable patterns in the Brodatz Texture Database have also evolved. Charles Darwin in his famous book Origin of Species suggests that evolution takes place using a method he termed Natural Selection. Survival of fittest has to be considered in context of the environment in which the species are living. One of the criteria of survival depends on whether the specie was able to propagate before it was hunted down by a predator. For example the skin color of a prey like an insect evolves to match the background it is living on such that the predators like birds and frogs will not detect them. Assume there are thousand insects on a surface and if nine hundred and ninety are eaten over a period of time are eaten. Then the last 10 will be those which have got skin color, pattern, and texture closest to the original. Some findings have been in the comparison of different genetic operators to evolve specie. Use of Mutation and Crossover operator is critical. Some level of continuation is required in the next generation to maintain and not lose the quality which has been already attained. The number of generations required depend on DNA length and on the correct use of these three parameters.. The runtime from thirty minutes has been reduced to ten minutes to evolve specie of 100 plus variables in the DNA of the images. If we are ready to run the program for a day on multiple machine in parallel then the possibilities of great diversity is very much possible. The predator will be an algorithm which tries to do a basic computer vision operation of correlating two images and ranking them. Then it eliminates a certain percentage of the population. The dissertation is broken down in chapters. Chapter 1 gives the Introduction and Chapter 2 gives an idea of what all research is happening in this area. Chapter 3 is the core content of my work and it shows the different approaches I have used to evolve textures, the use of different operators. This chapter is followed by the testing I had to do due to the complexity of the application and high chance of error creeping into the results. An idea of the application can be got from the Appendix where I have given a walk through of the Zing World

Jos Jibrael

Computer Science