Chapter 7. Evolving Connectionist and Fuzzy -
Connectionist Systems: Theory and Applications for
Adaptive, On-line Intelligent Systems
Nikola Kasabov
Department
of Information Science
University of Otago, P.O Box 56, Dunedin, New Zealand
Phone: +64 3 479 8319, fax: +64 3 479 8311
nkasabov@otago.ac.nz
Abstract. The paper introduces one paradigm of neuro-fuzzy techniques and an
approach
to building on-line, adaptive intelligent systems. This approach is called
evolving connectionist systems (ECOS). ECOS evolve through incremental, on-
line learning, both supervised and unsupervised.
They can accommodate new
input data, including new features, new classes, etc. New connections and new
neurons are created during the operation of the system.
The ECOS framework is
presented and illustrated on a particular type of evolving neural networks -
evolving fuzzy neural networks. ECOS are three to six orders of magnitude faster
than multilayer perceptrons, or fuzzy neural networks,
both trained either with the
backpropagation algorithm, or with a genetic programming technique. ECOS
belong to the new generation of adaptive intelligent systems. This is illustrated on
several real world problems for adaptive,
on-line classification, prediction,
decision making and control: phoneme-based speech recognition; moving person
identification; wastewater flow time-series
prediction and control; intelligent
agents; financial time series prediction and control. The principles of recurrent
ECOS and reinforcement learning are outlined.
Key words: evolving neuro-fuzzy systems;
fuzzy neural networks; on-line
adaptive control; on-line decision making; intelligent agents
1. Introduction: Adaptive, on-line, incremental learning -
problems with the conventional
