Research On Optimization And Application Of Variational Quantum Algorithms

Quantum computer has excellent performance in analyzing high-dimensional data and performing complex computations far beyond classical computer,which not only poses a threat to the existing cyber security system,but also provides a new development opportunity for the field.However,a fault-tolerant quantum computer requires millions of quantum bits with low error rates,which is still a long way from current techniques and may take decades.Variational quantum algorithms(VQAs)have emerged as a promising near-term technique to explore practical quantum advantage on noisy intermediate-scale quantum(NISQ)devices.Therefore,it is of great significance to carry out our research on the optimization and application of VQAs.In this paper,we focus on the optimization method and cloud security application of the VQAs.The main results are as follows:(a)Parameter Initialization Method of Variational Quantum Algorithms with Near Clifford Circuits.Parameter initialization of quantum circuits is an important factor affecting the convergence of VQAs.Warm Start Algorithms with Near Clifford Circuits(WS-NCC)is proposed to obtain better initialization parameters and accelerate the optimization process of conventional VQAs.WS-VQAs provides good initialization parameters for the optimization of VQAs by efficiently pre-training the Near-Clifford circuit and performing quantum gate decomposition on a classical computer,which speeds up the convergence of VQAs while saving a large amount of quantum resources.The relationship between the expressibility of Near-Clifford circuits and performance of the algorithm is further studied and it is concluded that the better the expressibility is,the better the performance is.The acceleration effect of WS-NCC is verified by performing a classification task,and the achieved results show that warm start with Near-Clifford circuit which has optimal expressibility can improve the training efficiency of VQAs by at least 90%.(b)Parameter-Parallel Distributed Variational Quantum Algorithms.Distributed machine learning utilizes multiple computing nodes to execute target tasks simultaneously,which greatly improves the execution efficiency of algorithms.Parameter-Parallel Distributed Variational Quantum Algorithms(PPD-VQA)is proposed to accelerate the training process of VQAs,which utilizes multiple quantum processors for parameter-parallel training with an approximate linear acceleration effect regarding the number of quantum processors.The convergence of PPD-VQA in noise scenarios is theoretically proved to be the same as that of conventional VQAs.Further,an alternate training strategy is proposed to alleviate the acceleration attenuation caused by excessive noise difference among multiple quantum processors.The achieved results show that the speedup ratio is only 39% of theoretical value when noise difference among quantum processors is large,and the speed-up ratio using this strategy reaches at least 91.8% of the theoretical value.In addition,the gradient compression algorithm is also employed to overcome the potential communication bottlenecks,and the achieved results show that the algorithm can save 60% to 87% communication cost while guaranteeing the acceleration of PPD-VQA.The algorithm effectively improves the training efficiency of VQAs,which also provides a practical solution for coordinating multiple quantum processors to handle large-scale real-word applications.(c)Cloud-Security Variational Quantum Algorithms.Considering the high R&D costs and technical threshold of quantum computers,cloud computing will be the main deployment format of quantum computing in the future,and the increasingly prominent information security issue between client and cloud will be a major challenge for that.A cloud-security variational quantum algorithm is proposed and proved to be secure and correct,which combines the VQAs based on measurement quantum computing model with universal blind quantum computing protocol.The algorithm serves as a universal cloud quantum machine learning framework that can incorporate most of the current blind quantum computing protocols and the design method of VQAs,further extends in terms of verifiability of the algorithm.This work implements a secure and reliable cloud VQA training framework and provides a theoretical basis for the application and further development of cloud quantum machine learning.