ENHANCEMENTS TO THE DYNAMIC SELF-ORGANIZING NEURAL-NETWORK

Class Directed Unsupervised Learning (CDUL) is a dynamic self-organisi ng network which has been shown to overcome many of the problems assoc iated with unsupervised learning, thereby yielding performance charact eristics superior to similar networks such as counter-propagation and LVQ. In this paper, the CDUL algorithm is developed further, to a poin t where the original two-phase learning process is combined into a sin gle system of dynamic parameter variation; a training cycle that can t hen be terminated automatically at a point of zero error over the trai ning set. The ability to improve training times using a FastCDUL algor ithm is also explored. The new algorithm, CDUL2, is subsequently appli ed to the benchmark problem of mine detection given sonar data, and sh own to outperform both backpropagation and LVQ in terms of training sp eed and recall performance. Finally, a measure of computational cost i s estimated for both CDUL2 and LVQ training periods, reinforcing the s uggested efficiency of CDUL2 over its counterparts.