Neuro-Fuzzy Computing: Structure, Performance Measure and Applications
P. A. Stadterl, N.K. Bose · Studies in fuzziness and soft computing · 2000
Nonrecurrent and recurrent neural network structures are used, respectively, for pattern classification and image processing. An approach for automated pattern classification is developed based on a Fuzzy Voronoi Neural Network (FVNet) architecture. The FVNet training, based on spatial tessellation, results in a flexible, generic, modular structure that is subsequently refined by locally tuning the initial class decision surfaces, which are generated from the Voronoi diagram. The FVNet has subnetworks that are capable of learning fuzzy logic constructs from empirical data. A new performance measure, called the modified fuzzy divergence, which is particularly suited to evaluating neuro-fuzzy classifi-ers, has been developed. Processing of images by a Hopfield network and a Boltzmann machine is discussed with particular attention to the consequences of applying deterministic and stochastic learning rules and the different needs for neurocomputing in contrast to classical computing. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.