Neural networks – advantages and applications

摘要: Abstract The main-stream neural network models can be divided in three categories: the signal transfer or feedforward networks, competitive and dynamic state networks. There is presently a strong emphasis to implement these various networks efficient hardware software, find more applications for them. Especially first two of categories above have been widely used practical pattern recognition tasks. Multi-Layer Perceptron (MLP) finding nonlinear associations between input vectors output vectors, based on training sets. Classification, diagnosis, prediction, control are typical problems this nature. Self-Organizing Map (SOM) able clusters by an unsupervised learning algorithm such that preserve some topological relations space. It applied categorization visualization complex data. In following, will reviewed, which developed up production quality. Conclusions derived from experiments given. applications, with theoretical developments left references.