Evaluation of the Neural Gas Network Vector Quantization and Approximation Components.
Michael T. Cozart May 1996
The University of Tennessee Space Institute
One neural networking design technique for the prediction of nonlinear time-series implements two primary computational components, vector quantization and function approximation. A "Neural Gas" network which uses this design technique has reportedly produced superior prediction results of time- series. This research evaluates the effectiveness or influence of each component of the Neural Gas network by substituting other reputable methods of vector quantization and approximation for those of the Neural Gas network. The substitute quantization component is the Generalized Learning Vector Quantizer. The substitute approximation component is Radial Basis Functions. Results of this research indicate the approximation component of the Neural Gas network contributes most to the excellent prediction results. The prediction results from the substituted components versus the Neural Gas components are discussed.
Neural Network FAQ
Frequently asked questions about Neural Networking.
Neural Innovation's list of useful neural networking sites
Web sites, FTP sites, and Journals
NEuroNet Software
NEuroNet's repository of sites with some commercial and many free Neural Network software packages. Many with source code.
atree-2.0
Atree adaptive logic network simulation software. This package will help you experiment with learning networks that will instruct and perhaps surprise you.
lee-1.1
Latent Energy Environments software package. LEE is an artificial life simulator which can be used to evolve populations of neural networks adapting to environments of increasing complexity.
lvq_pak-2.1
The learning vector quantization package. This package contains all the programs necessary for the correct application of certain LVQ algorithms in an arbitrary statistical classification or pattern recognition task.
neuralnet-2.0
Motif neural net package. NeuralNet is a GUI to the designing, training, and evaluating of neural networks using the backpropagation training algorithm. Simple sample network, pattern, and weights files are included. No explanation of the file formats are included except the pattern files, which are explained in the help file supplied with the program. The point being that the program takes care of the generation of the other two files, network and weights, for the user. The pattern files must be created by the user. All file formats are in text (ASCII).
snns-3.1
Stuttgart Neural Network Simulator. This package consists of three main components. The NESSUS (network definition language of the University of Stuttgart) compiler allows the generation of medium-sized to large structured nets using a programming language developed especially for this task. The output from the compile (an ASCII file) is read in by the kernel which builds its own data structures in memory. The graphical user interface, XGU, built on top of the kernel, gives a graphical representation of the neural networks and controls the kernel during the simulation run.
som_pak-1.2
Package for application of Self-Organising Map algorithms. This package contains the necessary programs for the correct application of SOM algorithms in the visualisation of complex experimental data.
Back to Co's Corner
mcozart@tds.net