Research On The Construction Of Quantum Neural Network Based On Quantum Superposition Coding

Quantum Computing has shown more remarkable computing ability than classical computing in large number decomposition and data search,which has attracted wide attention of researchers and is also popular research directions in quantum mechanics.Artificial intelligence algorithm is mainly built on the basis of multilayer neural networks.In order to cope with increasing networks complexity and memory requirements,it is hoped that the operation mode and structure of these powerful intelligence algorithms can be improved.The powerful parallel computing abilities of quantum computer and the huge storage capacity of Hilbert space are excepted to solve these problems more efficiently then any classical computer.Currently,quantum machine learning and quantum deep learning technology have become an important research direction in the field of quantum computing.By simulating the structure of classic deep learning neuron,quantum neuron can realize the function similar to classic neuron.Quantum can store2~Ndimension classical information in N qubits attribute to the huge storage capacity of Hilbert space.With the development of quantum computing,building a more powerful quantum neural network structure,and transplanting classic training tasks or neural network structure to the quantum domain is a common idea to improve neural network performance.Based on the research of quantum neural network,this thesis has done the following work in combination with the principle of quantum computing:First,this thesis analyzed the construct method of several quantum neurons,and discussed their quantum circuits structure and calculation process,including binary weight value neuron and neuron based on quantum rotation gates.On the basis of above neurons,this thesis designed a quantum neural network structure based on quantum rotation gate neuron and quantum superposition.According to the theory of quantum superposition state,the input training data and labels are encoded into a quantum superposition state,and all the input data are calculated at the same time through a quantum neural network.Utilize the nature of quantum parallel computing to reduce the number of executions of quantum circuit,and the feasibility of the structure bas verified through simple training tasks.Second,according to the problems of quantum neurons in this thesis,especially when using the neuron in multi-layer network structure,this thesis introduces a structure for multi-layer quantum neural networks using the neuron this thesis proposed.Since the{0,1}quantum neuron in this thesis has the problem of weak learning ability when training for complex tasks,the quantum neurons in this thesis can be extended to{-1,+1}encoding by introducing qubit flip gate X.Through the actual training tasks,the expended neuron has stronger expression ability than the original neuron.Finally,a comparison simulation experiment between the binary weight neuron and the neuron in this thesis was carried out,and the advantages and disadvantages of the two neuron structures were analyzed and summarized.