在大数据时代，机器学习尤其是深度学习对社会的重要性不言而喻。但随着深度学习算法复杂度的不断提高，在愈加海量数据的条件下运行时间愈发难以接受。为了解决这一困境，本项目将量子计算与神经网络相结合，同时受随机计算重要成果的启发，研究量子深度神经网络的结构设计和训练方法，并对其在QFPGA结构上的逻辑映射进行优化。主要研究内容包括：.1. 经典数据的量子态表示和存储算法研究，消除经典和量子数据转换的瓶颈。.2. 基于已发表的QPNN模型，进行深度量子神经网络设计与优化。拟采用量子随机过程模块来实现特定的量子激活函数，构建经典-量子混合的优化深度网络结构。.3. 将深度量子神经网络，在基于MBQC模型的QFPGA架构上进行逻辑映射，并且优化其实现所需的量子门数量。

As the cornerstone in the big data era, machine learning especially deep learning is playing very important role in our society. However, current deep learning algorithms are not efficient enough for an ever-growing amount of data because of the high complexity. To overcome this dilemma, inspired by the important results of stochastic computation, we will combine the advantages of quantum computation with the deep learning to study quantum deep neural network (QDNN) structures, the training methods, and the optimized logical synthesis method to map QDNN on QFPGA architectures we have proposed..Our research includes the following steps..1. Quantum state representation and storage/retrieve algorithm study, to remove the bottleneck of the conversion between quantum data and classical data. .2. Study of the deep quantum neural network structures based on QPNN. We are planning to use the quantum stochastic process model to realize quantum activation and then construct classical-quantum hybrid deep network structures..3. Finally, the verified deep quantum neural networks will be mapped to QFPGA architecture based on the MBQC model and the quantum costs are optimized as well.