| This thesis begins by describing current methods of automatic program creation based on swarm behavior, modeled after particle swarm optimization (PSO). It then describes and compares two implementations of a PSO approach to a symbolic regression problem (a subset of genetic programming). In these techniques, particle encoding is based on multi-expression programming and Cartesian genetic programming chromosome encoding. It allows for the representation of both real and discrete values. The first PSO approach called pseudo swarm crossover uses a crossover-based particle update function, while the second approach, Cartesian swarm programming, uses the standard PSO distance-based update function modified for discrete variables. These proposed techniques are then compared with established genetic programming techniques. |
