Evolving intelligent systems: methods, learning, & applications

The basic concept, formulation, background, and a panoramic view over the recent research results and open problems in the newly emerging area of Evolving Intelligent Systems are summarized in this short communication. Intelligent systems can be defined as systems that incorporate some form of reasoning that is typical for humans. Fuzzy Systems are well known for being able to formalize human knowledge that still separates humans from machines. Artificial Neural Networks have proven to be a useful form of parallel processing of information that employs principles from the organization of the brain. Finally, the evolution is a phenomenon that was initially used to solve optimization problems inspired by the progress in Genetic Algorithms, Evolutionary Computing, and Genetic Programming. These types of evolutionary algorithms are mimicking the natural selection that takes place in populations of living creatures over generations. More recently, the evolution of individual systems within their life-span (self-organization, learning through experience, and self-developing) has attracted attention. These systems called 'evolving' came as a result of the research on practical intelligent systems and on-line learning algorithms that are capable of extracting knowledge from data and performing a higher level adaptation of model structure as well as model parameters. Evolving systems can also be considered an extension of the multi-model concept known from the control theory, and of the on-line identification of fuzzy rule-based models. They can also be regarded as an extension of the methods for on-line learning neural networks with flexible structure that can grow and shrink This new concept of evolving intelligent systems can also be treated in the framework of knowledge and data integration. Evolutionary, population / generation based computation, can be applied to optimize parameters and features of an individual system, that learns incrementally from incoming data. The specifics of this paper lays in the generalization of the recent advances in the development of evolving fuzzy and neuro-fuzzy models and the more analytical angle of consideration through the prism of knowledge evolution as opposed to the usually used data-centered approach. This powerful new concept has been recently introduced by the authors in a series of parallel works and is still under intensive development. It forms the conceptual basis for the development of the truly intelligent systems. A number of applications of this technique to a range of industrial and benchmark processes have been recently reported. Due to the lack of space only some of them will be mentioned primarily with illustrative purpose. ©2006 IEEE