Improvement Of Quantum Neural Network

The Purpose of ResearchThe research purpose of this article is to find a suitable breaking point to enable my research strength, to import Quantum theories into Artificial Neural Network. After detecting present models, some points to improve were founded and changed to improve the constringency speed, reliability and stability. Then different application scopes are tried for the new model. It is lucky enough if some suitable application scope were founded.ResultsThe main efforts of this article include analyzing quantum algorithms and quantum neural network models, choosing quantum associate memory as a research breaking point, based on introducing some basic quantum theories. By adjusting operators, phase defects can be corrected in time by improving operators during the collapsing process, which makes the collapsing speed much faster and increases astringency.In the no noise wholly input experiment, with the only effect of improved I_φ, operator, the condition of getting good result only under certain iterative times in original quantum system with 200 solution states is changed into sustainable convergence into a small range of 99.28% to 99.68%, which is nearly 100%, only after 11 iterative times. In partly input experiment, with the effect of improved I_φoperator, possibility of target solution states is improved from a large range of 0%-72.32% to a small range of 43.75%-72.32%; possibility of target un-solution state's is improved from a range of 0%-72.32% to a range of 8.04%-22.32%; un-target solution state's decreases from 0.89%-14.29% to 0.89%-8.04%. To limit the possibility of target un-solution state more, I_p operator is imported into the model. Under the effect of I_p and I'_φQuAM improves from varying in a large range with a high speed into the result of target un-solution state under 3.2%, un-target solution state under 3%, target solution state above 80.31%, only after the 3^rd iterative time. Because of the uncertainty of the quantum double base states theory, the collapsing possibility of quantum system cannot become 100%, a result of above 80% is good enough to accept in quantum system.Applying the improved quantum associated memory model into pattern storage & recognition, a QuAM model with 12 qubits was created to remember patterns, which has more than 4000 base states. In microcosmic, the base state of |1910> has been selected, the constringency rate can keep more than 99.07% after the 26th iterative time. In macrocosmic, 70.52% successfully recalling rate has been got in 23.88% large area data disappearing experiment. In strip noisy experiment with 11.11% noisy rate, a good result with 88.47% successfully recalling rate and 1.36% noisy rate in the output has been got. In random noisy experiment, input noisy rate changes in 3 levels: 90.55%, 78.64%, 54.06%, the noisy rate of output reached 1.11%, 2.52%and 5.41%. Trying the potential to generate output according to the largest swing in a period, not in the special time, 11.11% noisy rate is put into input pattern, then the successful rate increases into 73.35%, and the noisy rate in output is only 2.96%.For key experiments, multi different patterns have been selected in different noise rates, which can certainly confirm the universality of application and the credit of results. The improved QuAM model can not only tolerance specific noise, including large area data disappearing, stripe noise, but also perform well under random noise in different noise intensity. It can be concluded that the original model has been improved distinctly in constringency speed and reliability by comparing experiment.New FindingsQuantum theory has been founded more than 100 years. Many branches have been growing during this period. As research strength is here and there, Quantum theory is becoming an incomplete and unbalanced large system. As time resource is very limited during my graduate, it is a great decision for me to select Quantum Associate Memory and apply it to image pattern storage & recognition scope. Also, this is a lucky selection.Based on the result of experiments, the structure of original model and the effect of operator's parts have been analyzed in detail, and improvement has been made accordingly. This increment has both been demonstrated in theory and confirmed by experiments. Different scopes were tried, then apply the improved QuAM model into image pattern storage & recognition. It is found that the improved QuAM model enjoys a better stability and noise shielding capability in both specific noise, including large area data disappear, stripe noise, and random noise under different intensity from microcosmic and macrocosmic.Meaning and ValueQuantum Associate Memory is the embryo of storage structure for quantum system in the future. As an important quantum neural network model, the optimization of structure and the improvement of performance can not only present the great power of quantum theory more clearly to academia, but also do some front preparation for the future quantum system. By applying the quantum associate memory model to image pattern storage & recognition and simulating the experiment implement, a new path has been found for QuAM. Also, it makes much easier to find some evidence for the increment of quantum computation performance, and measurable to keep improving of the quantum model for further research.Quantum neural network is proposed by absorbing quantum theory into artificial neural network. It is a new kind of neural network, which has become a fast developing scope directly related to algorithms in quantum informatics and quantum computation. By importing quantum theories into ANN, the capability of neural network can be greatly strengthened and become more close to biologic intelligence. More, some worldwide difficulties, including "the consciousness of human being", "the formation of memory", are expected to be solved in the scope of quickly developing quantum theories.