Simulations of a quantum perceptron on IBM-Qiskit

The principal aim of this thesis is to try to simulate the functioning of a quantum perceptron, that is an artificial neural network which task is to catalogue different input patterns. The model implemented has been tested through the realization of a learning procedure. This consists in teaching to the network to correctly recognize a pattern, in order to finally obtain a machine that is able to recognize an image. This procedure is fundamental in neural networks since it permits to the system, once it has been performed, to correctly recognize also patterns not encountered during learning. To this aim, this thesis will at first go through the explanation of some perceptron models in the first two chapters, and gives also in the Appendix some fundamental notions of quantum mechanics and quantum computing necessary for this treatment. In fact, the first chapter tries to explain how classical artificial neural networks have been thought and developed until now, in order to arrive to the theory of implementation of a perceptron and give some examples. Then, in the second chapter one of this models is taken and a perceptron of the same logic is implemented on a quantum circuit. This model of a quantum perceptron is then used in the third chapter and realized through a software development kit called Qiskit, provided by IBM Research. Here the steps followed for the implementation of the learning procedure will be explained and the results obtained will be analyzed. The process in which the machine is trained in this discussion is realized with a classical algorithm, and the quantum circuits have been run on a classical simulator, which behaves as an ideal quantum computer.