Evolving Computational Intelligence Systems

A new paradigm of the evolving computational intelligence systems (ECIS) is introduced in a generic framework of the knowledge and data integration (KDI). This 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 datacentred approach marks the novelty of the present paper. ECIS constitutes a suitable paradigm for adaptive modeling of continuous dynamic processes and tracing the evolution of knowledge. The elements of evolution, such as inheritance and structure development are related to the knowledge and data pattern dynamics and are considered in the context of an individual system/model. Another novelty of this paper consists of the comparison at a conceptual level between the concept of models and knowledge captured by these models evolution and the well known paradigm of evolutionary computation. Although ECIS differs from the concept of evolutionary (genetic) computing, both paradigms heavily borrow from the same source – nature and human evolution. As the origin of knowledge, humans are the best model of an evolving intelligent system. Instead of considering the evolution of population of spices or genes as the evolutionary computation algorithms does the ECIS concentrate on the evolution of a single intelligent system. The aim is to develop the intelligence/knowledge of this system through an evolution using inheritance and modification, upgrade and reduction. This approach is also suitable for the integration of new data and existing models into new models that can be incrementally adapted to future incoming data. 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. Another specific of this paper includes bringing together the two working examples of ECIS, namely ECOS and EFS. The ideas are supported by illustrative examples (a synthetic non-linear function for the ECOS case and a benchmark problem of house price modelling from UCI repository for the case of EFS).