SELF-ORGANIZING NEURAL NETWORKS FOR ADAPTIVE-CONTROL

Self-organizing neural networks have been implemented in a wide range of application areas such as speech processing, image processing, opti mization and robotics. Recent variations to the basic model proposed b y the authors enable it to order state space using a subset of the inp ut vector and to apply a local adaptation procedure that does not rely on a predefined test duration limit. Both these variations have been incorporated into a new feature map architecture that forms an integra l part of an Hybrid Learning System (HLS) based on a genetic-based cla ssifier system. Problems are represented within HLS as objects charact erized by environmental features. Objects controlled by the system hav e preset targets set against a subset of their features. The system's objective is to achieve these targets by evolving a behavioural repert oire that efficiently explores and exploits the problem environment. F eature maps encode two types of knowledge within HLS-long-term memory traces of useful regularities within the environment and the classifie r performance data calibrated against an object's feature states and t argets. Self-organization of these networks constitutes non-genetic-ba sed (experience-driven) learning within HLS. This paper presents a des cription of the HLS architecture and an analysis of the modified featu re map implementing associative memory. Initial results are presented that demonstrate the behaviour of the system on a simple control task.