Research On Quantum Feature Selection And Quantum Perceptron Algorithm

Machine learning is a multi-disciplinary,data-based discipline that aims to extract a "pattern" from the data and use it to understand previously unknown input data.Machine learning algorithms have important applications in many fields,such as economics,military,and medical.Because of some important features of quantum mechanics,such as parallelism,superposition and coherence,it may bring exponential acceleration to computing and a significant increase in capacity.So researchers try to combine machine learning with quantum mechanics,and produce a new research direction,namely quantum machine learning.The combination of machine learning and quantum information processing achieves a win-win situation.On the one hand,machine learning and quantum computing system control have a good effect;on the other hand,quantum mechanics provides an attractive prospect for improving the learning efficiency of machine learning,firstly to reduce computational complexity,and secondly to strengthen machine learning.This thesis mainly studies two directions of quantum machine learning: quantum feature selection algorithm and quantum perceptron algorithm.Researchers pay more attention to the exploration and research of learning algorithms,such as the research of quantum perceptron algorithms.There are still some room for improvement in the existing results.In addition,the preprocessing of data features,such as feature selection,plays an important role in the training process of machine learning algorithms.However,there are few related researches on existing quantum feature selection algorithms.Selecting the most relevant features efficiently and accurately is also a direction worthy of attention.Therefore,in order to improve the computational efficiency,ensure the accuracy of the algorithm,and reduce the resource consumption,the thesis analyzes and discusses these two parts.The main research contents are as follows:(1)A quantum corresponding version of the classic Relief algorithm based on two classifications is proposed: QRelief algorithm.The algorithm mainly includes four steps: quantum state preparation,similarity calculation,weight vector update and feature selection.After evaluation,the algorithm is superior to the corresponding classical algorithm in efficiency and resource consumption.Not only that,quantum experiments on the IBM Q platform are implemented to verify the correctness of our algorithm.(2)According to the classic ReliefF algorithm,the QRelief algorithm is extended,and a feature selection algorithm based on multi-classification is proposed: QReliefF algorithm.Different from the QRelief algorithm,this algorithm uses the oracle mechanism to find the nearest k neighbor samples,and has a speedup than the previous one.(3)An efficient one-time iterative perceptron algorithm based on unity weight is proposed.Compared with the previously proposed algorithm,our algorithm not only has better availability,higher accuracy and wider applicability,but also can be applied to non-ideal training sets.