The capacity of the Omega rule

The author presents results on the performance of a neural network using Omega learning. Unlearning is performed on single neurons. The performance was computed experimentally. Each memory consisted of 36 neurons, and an unlearning rate of 20 was used. Seven runs were performed starting with 10 memories, each run increased by five memories. The results indicate a performance intermediate between that of Hebbian and Delta learning. The memories used for each run were generated randomly. It was noted on the larger memory runs that the learning/unlearning limit was reached on the later memories. Different versions of the learning scheme were analyzed. Time complexity in learning/unlearning, and sensitivity to the learning/unlearning rate are discussed