Can competitive learning compete? Comparing a connectionist clustering technique to symbolic approaches
J. Jeffrey Mahoney, Raymond J. Mooney · 2002
A comparison of competitive learning (a neural-network-based approach to data clustering) with established symbolic approaches is presented. Some of the shortcomings of competitive learning are discussed along with attempts at correcting them. The algorithm is extended to handle the performance task of missing feature prediction. Experimental results are compared with similar results of symbolic systems, such as Cluster/2 and Cobweb. In these experiments, competitive learning does not perform as well as its symbolic counterparts.>