The unsupervised training model (see Fig. 8.1) is similar to the supervised model, but differs in that no teacher is employed in the training process. It is analogous to students learning the lesson on their own. Two of the most popular unsupervised learning techniques used in the neural-network community are the self-organizing map (SOM), developed by Teuvo Kohonen, and the adaptive resonance theory (ART) network, developed by Stephen Grossberg and Gail Carpenter. The unsupervised training model consists of the environment, represented by a measurement vector. The measurement vector is fed to the learning system and the system response is obtained. Based upon the system response and the adaptation rule employed, the learning-system weights are adjusted to obtain the desired performance. The learning process is an open loop with a set of adaptation rules that govern general behavior such as a neighborhood and learning rate in the case of the Kohonen SOM [Kohonen, 1982] and the vigilance parameter in the case of the ART network [Carpenter, 1987].
Online access to SPIE eBooks is limited to subscribing institutions.