Features

Small number of functions constructed so that only the minimum amount of information has to be specified by the user
Well-organized palettes with command templates, options, and links to online documentation
Intelligent initialization algorithms to begin the training with good performance and speed
Extensive documentation including an introduction to neural network theory as well as highly illustrative application examples

Support for most of the commonly used neural network structures including radial basis function, feedforward, dynamic, Hopfield, perceptron, vector quantization, unsupervised, and Kohonen networks
Support for advanced training algorithms including Levenberg-Marquardt, Gauss-Newton, and steepest descent as well as for traditional algorithms including backpropagation with and without momentum
Support for typical neural network applications including function approximation, classification, dynamic systems modeling, time series, auto-associative memory, clustering, and self-organizing maps

Visualization tools for viewing network models, the training process, and network performance
Special network object to identify the type of network and list its parameters and properties
Special training record to keep intermediate information from the learning process
Functions equipped with a large number of advanced options to modify and control the training algorithms
Support for neural networks with any number of hidden layers and any number of neurons (hidden neurons) in each layer
Access to all of the capabilities of Mathematica to prototype new algorithms or to perform further manipulations on neural network structures

Optimization of expressions before numerical evaluation to minimize the number of operations and to reduce computational load
Compile command to send compiled code directly to Mathematica to increase computational speed
Special performance-evaluation functions included to validate and illustrate the quality of a mapping

More Learning