December 9, 2002–The new Neural Networks application package from Wolfram Research provides a robust environment for modeling and predicting data. Artificial neural networks have revolutionized the way researchers solve complex, real-world problems in engineering, science, economics, and finance. Now users can test and explore neural network models faster and easier than ever before, using the computational power and flexibility of Mathematica.

Neural Networks is designed to give professionals and students the tools to train, visualize, and validate neural network models. It supports a comprehensive set of neural network structures and implements state-of-the-art training algorithms while taking full advantage of Mathematica‘s number-crunching and visualization capabilities. It also includes special functions to address typical problems in data analysis, such as function approximation, classification and detection, clustering, nonlinear time series, and nonlinear system identification problems.

Building on a user-friendly command structure, Neural Networks offers:

• Support for proven neural network paradigms, structures, and applications
• Support for traditional and advanced training algorithms
• Intelligent initialization algorithms to begin training with superior performance and speed
• Fast, reliable optimization of expressions and compiling of code with a single command
• A powerful modeling environment with support for hidden layers and neurons

Built-in palettes facilitate the input of any parameter for the analysis, evaluation, and training of your data, making the product easy to learn and use. This was a high priority during the development cycle, as was using Mathematica‘s rich computation and programming environment to incorporate additional functionality. “Unlike other solutions that consist of restrictive black box implementations, Neural Networks allows you to see and evaluate how your data is being trained from start to finish, at every step along the way,” says Yezabel Dooley, Applications Product Manager at Wolfram Research.

With Neural Networks you can fit the network to your data, visualize the fitted network, and view the distribution of errors with only a few commands. Professionals and students with little or no background in neural networks will benefit from the comprehensive online documentation, the large number of examples, and the interactive online tutorial. At the same time, advanced users will appreciate the flexible options for each function and numerous possibilities for modifying the included algorithms, as well as the ability to develop new training algorithms of their own to further extend the capabilities of the package.

Neural Networks is designed for use with Mathematica 4 or later and is available for Windows, Mac OS, Mac OS X, Linux (PC, Alpha, PowerPC), Solaris, HP-UX, IRIX, AIX, Compaq Tru64 Unix, and compatible systems. Further information about Neural Networks is available.