A Genetic Learning Algorithm for the Analysis of Complex Data
Norman H. Packard
Center for Complex Systems Research, Beckman Institute,
and
Physics Department, University of Illinois,
405 North Mathews Avenue, Urbana, IL 61801 USA
Abstract
A genetic learning algorithm modeled after biological evolution is presented to discern patterns relating one observable that is taken to be dependent on many others. The problem is reduced to an optimization procedure over a space of conditions on the independent variables. The optimization is performed by a genetic learning algorithm, using an information theoretic fitness function on conditional probability distributions, all derived from data that has a very sparse distribution over a very high-dimensional space. We will discuss applications in forecasting, management, weather, neuroanalysis, large-scale modeling, and other areas.
