Abstract

The constantly increasing amount of resources available to engineers and scientists have allowed them to target larger problems whose size and complexity introduce new challenges: the time required to find the solution is longer, the solutions are more error prone, and the tests and repairs are more expensive. Self-organizing methods have recently been the promising pioneers in dependable robust design. Distributed self-organizing patterns can emerge to demonstrate the desired characteristics, either in form or functionality. At the same time, being inspired by the natural development of multicellular organisms, researchers have started using artificial development to improve features such as scalability or fault tolerance of the solution. However, the current solutions resulting from artificial development are either very small in size or very simple in architecture.

The first part of this thesis introduces a method to emerge patterns that demonstrate given functionality whose architecture is not known in advance. The notable achievement is the innovative fitness function in the evolutionary algorithm used there, which increases the density of the solutions in the search space and more importantly, makes the often-extremely-rough search space smoother.

The second and major part of this thesis studies formation of given large patterns from simpler initial patterns. This problem is solved in the framework of Cellular Automata. We push our methods to their limits by targeting large non-periodic patterns that have not been originally created by developmental methods. We use patterns for which the similar existing methods take a long time to find the solution, and their solutions are often large and seldom scalable. We suggest improvements to the existing methods to allow them find more efficient solutions, and also present two new methods to improve the results even further. In the end, we show that our suggested method also contributes to scalability. More specifically, our second suggested method decreases the growth rate of the solution to be slower than the growth rate of the problem size.