Reduction of Visual Information
Growing Neural Gas
Description: The maximum number of nodes (neurons) in the network was set to 10,15 for Task1,Task2 respectively. Lambda - the period to insert new neuron (given in number of presentations of the training set). Beta - the decay constant for the local variables E and U describing the error and utility values of a neuron. All other variables present in the GNG algorithm have their values set according to [1]. The Utility extension present in GNG-U algorithm [2] was not used as a static task was studied with not changing distribution of input patterns.
Training patterns (Task1):
Training patterns (Task2):
Bibliography
[1]. Fritzke, B.: A growing neural gas network learns topologies. In Tesauro, G., Touretzky, D.S., Leen, T.K., eds.: Advances in Neural Information Processing Systems 7. MIT Press, Cambridge MA (1995) 625-632.
[2]. Fritzke, B.: A self-organizing network that can follow non-stationary distributions. In: ICANN'97: International Conference on Artificial Neural Networks, Springer (1997) 613-618.