| The quantron is a new model of artificial neuron. In order to take advantage of its unique attributes, we must develop a learning algorithm.; Initially, we investigate various existing neuron models similar to the quantron. A document review reveals various supervised learning algorithms. However, from the abundance of supervised learning algorithms, none satisfies the unique attributes of the quantron. It is therefore necessary to produce an original learning algorithm.; Secondly, the propose learning algorithm is detailed. The principal difficulty is to manage the conduction or non-conduction of information in the quantron. The learning algorithm that we created does not apply only to the quantron. Before testing it with the quantron, we observe its behavior on a simplified quantron model. The results are excellent, since the learning algorithm makes it possible to minimize error. This part of the research is innovative because no other learning algorithm has been applicable to this type of neuron.; Finally, encouraged by results from the simplified quantron, we established the learning algorithm into a multilayer network of quantrons. We test the learning algorithm by image recognition. The results obtained are very encouraging. Our learning algorithm for the quantron is more accurately reflects recent discoveries in neurobiology. We believe that it will allow the neural network to solve more complex problems of classification and regression. |
