On the rate of convergence in topology preserving neural networks (Q808996): Difference between revisions

Kohonen learning (and his feature maps) attracts still more attention of neural network people. Perhaps everybody who worked with it faced the question of proper choice of time-dependent gain and shrinking coefficients as well as the good choice of the ``mexican hut''. The paper uses Gaussian type of the latter and shows that this ``neighborhood interaction function'' leads to a faster convergence.