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