Abstract: | Most machine learning techniques, such as the Genetic Algorithm or the Neural Network, are relatively unsuccessful at achieving knowledge discovery goals on relatively unknown relationships or classes in complex data sets. The goal in developing the superant system was to use a population-based algorithm to evolve multi-distributed agents with "collective knowledge" that may shed light on complex relations between classes in a set. The core component of the system, an evolution of heterogeneous agents, was quite successful in generating agents capable of identifying classes. Testing has established certain techniques of evolving populations to be better than others, given the particular nature of the data set.
|