| Summary: | According to Hadjar's experimental results, ANN with Boolean function algorithm offered many advantages than complicated ANN such as Neocognitron. Although it was successful because of its simplicity and efficiency. Some of weights generators generate binary weights using Boolean function and the other using non-binary weights such as Perceptron.
Because of lack of information was contained in Hadjar's experimental results, lead me to compare some binary weights generators using AND, OR, and XOR (common operators in Boolean function) and also non-binary weights generators that were modified from classical ANN, Perceptron.
After some experiments, I can conclude that the best binary weights generator is using OR operator. But the best weights generator of all is 'non-binary weights generator 1' that have already modified from Perceptron. AND maybe the worst one, but it can be considered if images are homogeneous.
Key words : Boolean function algorithm, Weights generator
|
|---|