The artificial neural network(ANN),as a powerful information processing tool,has made rapid development during the recent decades and been successfully applied into a great deal of practical scientific and engineering problems,such as facial recognition,nature language processing(NLP),etc.However,the efficiency of classical networks is seriously challenged by big data,it is imperative for us to find better information processing methods.At this time,researchers are inspired by quantum computing and expect to solve this problem with the help quantum mechanics.Different from the traditional calculation method based on mathematics,quantum computing is a new mode that follows the laws of quantum mechanics,it performs computational tasks based on the control of quantum units.Considering the efficiency of computation,due to the existence of quantum superposition,some known quantum algorithms are exponentially faster than traditional general-purpose computers.Therefore,through the combination of quantum theory and ANN,we hope that the classical algorithms can be optimized and the limitations of ANN can be overcome.The contributions and creative points of this study lie in:1.A feedforward neural network model based on Levenberg-Marquardt(LM)algorithm is proposed.This network is composed of quantum neurons and their connections,each node is linear combination of basis states.For the training of network,we give the application of LM algorithm on quantum environment.The simulations of nonlinear function fitting and Iris data classification are taken to investigate the feasibility of proposed QNN model,very few hidden neurons are required to reach a good precision and the structure of QNN is simple.2.We propose a novel quantum neural network,to improve the ability of processing information,the connection between input layer and output layer is added.Learning rate is an important parameter of neural network,based on the searching ability of quantum particle swarm optimization(QPSO),the learning rates can be tuned dynamically.For parameters initialization,we use the low discrepancy sequence to optimize the learning algorithm.To analyze the system performance,it has been applied to function fitting and data classification,it is shown by classical simulation and experimentation that the proposed model is more efficient and powerful than ANN for a lot of experimental tasks.3.In practice,stock market behavior is difficult to predict accurately because of its high volatility,to improve market forecasts,a method inspired by Elman neural network and quantum mechanics is presented.In this scheme,the internal self-connection signal that is extremely useful for system modeling is introduced to the proposed technique.To accelerate the learning,double chains quantum genetic algorithm(DCQGA)is employed to tune the learning rates.This model is validated by forecasting closing prices of seven stock markets,in the experimental part,the rule of data selection is introduced firstly,and then the method of data normalization is presented,due to the high volatility of stock market data,we need to normalize all of the data before experiment,simulation results indicate that this algorithm is feasible and effective,it reflects the variations in the stock market.Accordingly,generalizing the method is deemed advantageous.4.Based on the research of feedback neural network,a quantum Hopfield network for image recognition is designed.The structure of this model is similar to that of the classical network,but the samples need to be represented by quantum states,and then the quantum neuron can be evolved into a steady state or fixed point based on the conversion formula,when we input the samples,the output of network ideally corresponds to one of the patterns in training set.The feasibility of this scheme is proved by image recognition. |