For each iteration of the standard competitive learning rule, a single winning neuron is determined. The weights are adapted so that this neuron will respond more strongly to this input.
NeuroSolutions also implements competitive learning "with a conscience". This variant keeps track of how often the outputs win the competition with the goal of equilibrating the winnings (i.e., each unit in a set of N will win the competition 1/N on average). This implements a second level of competition among the elements to determine the winning neuron. It avoids the common occurrence in competitive learning that one element (or a subset) may always win the competition.
