Growing Neural Gas
The Growing Neural Gas (GNG) is an artificial neural network in which neurons can be inserted and deleted during the adaptation process. With other artificial neural networks such as Self-Organizing Maps (SOM) or Neural Gas (NG), the network size must be specified in advance. The Growing Neural Gas developed by Bernd Fritzke in 1995 can be seen as a further development of the NG.
Working method
At the beginning there are two start neurons. In the course of conditioning the network, new neurons are inserted so that the network can be adequately adapted to the input data. Connections between the neurons are inserted and deleted similar to the NG.
New neurons are inserted after a number of iterations specified in advance by the user.
The aim of the insertion process is to minimize the quantization error. The quantization error arises from the assignment of the input vectors to the reference vectors, whereby the data record is compressed.
See also
literature
- Bernd Fritzke: Growing Cell Structures - A Self-Organizing Neural Network Model . In: M. Dal Chin et al. (Ed.): Working reports of the Institute for Mathematical Machines and Data Processing (Computer Science) , Vol. 25, No. 9, Institute for Mathematical Machines and Data Processing, Friedrich Alexander University Erlangen, Nuremberg, 1992.
- Bernd Fritzke, A growing neural gas network learns topologies (PDF; 780 kB). In: Tesauro, G .; Touretzky, DS & Leen, TK (Eds.), Advances in Neural Information Processing Systems 7, MIT Press, 1995, 625-632
- Bernd Fritzke: Vector-based Neural Networks , electronically published habilitation thesis, 1998 (PDF; 17.0 MB)