Reduction of Visual Information
Hybrid Growing Neural Gas
Description: A network of given number of neurons is generated with each neuron connected to more than Eta randomly chosen neurons. The Lambda parameter specifies the neuron relocation period. In our implementation the neuron with minimal error variable is relocated near the neuron with maximal error variable after Lambda presentations of the training set. The umax parameter specifies the threshold for neuron relocation based on the neuron usage parameter. The threshold is equal to one presentation of the training set in our implementation.
Training patterns (Task1):
Training patterns (Task2):
Bibliography
[1]. Graham, J., Starzyk, J. A.: A hybrid self-organizing neural gas based network. IEEE World Congress on Computational Intelligence (2008).