The conscience algorithm is a modification of the SOM algorithm that forces the density matching to be exact (DeSieno, 1988). In the conscience algorithm, summarized in Table P9.7, each neuron keeps track of how many times it has won the competition (i.e., how many times its synaptic-weight vector has been the neuron closest to the input vector in Euclidean distance). The notion used here is that if a neuron wins too often, it "feels guilty" and therefore pulls itself out of the competition.
To investigate the improvement produced in density matching by the use of the conscience algorithm, consider a one-dimensional lattice (i.e., linear array) made up of 20 neurons that is trained with the linear input density plotted in Fig. P9.7
(a) Using computer simulations, compare the density matching produced by the conscience algorithm with that produced by the SOM algorithm. For the SOM algorithm use η = 0.05, and for the conscience algorithm use B = 0.0001, C = 1.0, and η = 0.05.
(b) As frames of reference for this comparison, include the "exact" match to the input density. Discuss the results of your computer simulations.
