Neural Network Toolbox

The Neural Network Toolbox provides a complete neural network engineering environment within MATLAB. It provides comprehensive support for the design, training, and simulation of many proven network paradigms, from basic perceptron models to advanced associative and self-organizing networks. It can be used for exploring and applying neural networks to applications such as signal processing, nonlinear control, and financial modeling.

Features

Supervised network paradigms: perceptron, linear, backpropagation, Levenberg-Marquardt, radial basis, Elman, Hopfield, and learning vector quantization
Unsupervised networks: Hebb, Kohonen, competitive, feature maps, and self-organizing maps
Competitive, limit, linear, and sigmoid transfer functions
Unlimited number of elements and interconnections
Customizable architectures and transfer functions
Dynamic simulation with SIMULINK
Portable C code generation through Real-Time Workshop

Highlights

Modular Organization. The Toolbox uses a consistent, modular implementation that facilitates research and simplifies customization. The Toolbox imposes no artificial limits on network size or connectivity – so there are no restrictions on the number of neurons in a layer or on the type of transfer function.

Architectures and Learning Rules. The Toolbox provides more than 15 proven network architectures and learning rules, allowing you to choose the paradigm that best fits your application or research problem. For each type of architecture and learning rule, the Toolbox provides functions to initialize, train, adapt, design, simulate, demonstrate, and show an application example of the network.

Supervised and Unsupervised Networks. For supervised networks, you can choose from feed-forward and recurrent architectures, using a variety of learning rules and design methods, such as perceptron, backpropagation, Levenberg-Marquardt backpropagation, radial basis networks, and recurrent networks.

For unsupervised networks, you can choose from associative networks or self-organizing networks, such as competitive, feature maps, and self-organizing maps. Associative networks can be used as building blocks for more complex networks using the Hebbian, Kohonen, instar, or outstar associative learning rules.

Neural Network Engineering Environment. The Neural Network Toolbox gives you access to a complete set of tools for neural network research, design, and simulation. The analysis and simulation tools in MATLAB and SIMULINK let you rapidly evaluate network behavior and performance in the context of complete system designs. With Real-Time Workshop, you can generate C code for use in standalone applications and embedded systems. Flexible data import and transformation functions simplify the preprocessing of input data.

Customizable. You can easily change any architecture, learning rule, or transfer function – or add new ones – without writing a single line of C or Fortran.

Demos and Tutorial. An acclaimed user's guide introduces neural network concepts and reinforces them with numerous examples and a complete reference section.

Stefan Steinhaus, webmaster@steinhaus-net.de