论文部分内容阅读
为提高过程神经网络的逼近和泛化能力,从研究过程神经元信息处理的量子计算实现机理入手,提出基于量子旋转门及多位受控非门的物理意义构造量子过程神经元的新思想.将离散化后的过程式输入信息作为受控非门的控制位,经过量子旋转门作用后控制目标量子位的状态,以目标量子位处于状态|1〉的概率幅作为量子过程神经元的输出.以量子过程神经元为隐层,普通神经元为输出层,可构成量子过程神经网络.基于量子计算机理推导了该模型的学习算法.将该模型用于太阳黑子数年均值预测,应用结果表明,所提方法与普通过程神经网络相比,预测精度有所提高,对于复杂预测问题具有一定理论意义和实用价值.
In order to improve the approximation and generalization ability of process neural networks, a new idea of constructing quantum process neurons based on quantum revolving gates and multi-bit controlled non-gates is put forward based on the realization of quantum computation of neuron information processing in the process of research. The discretized process input information is taken as the control bit of the controlled NOT gate, and the state of the target qubit is controlled after the function of the quantum rotation gate. The probability amplitude of the target qubit at state 1 is used as the output of the quantum process neuron Based on the quantum computing mechanism, the learning algorithm of this model is deduced by using quantum neurons as hidden layer and ordinary neuron as output layer, which can be used to predict annual average sunspot number. The application results The results show that compared with ordinary process neural networks, the proposed method improves the prediction accuracy and has certain theoretical and practical values for complex prediction problems.