Research And Application Of Quantum Boltzmann Machine Model In Machine Learning

In the past few decades,with the development of computer and artificial intelligence,machine learning algorithms have attracted more and more attention from scientists.In the era of big data,it is particularly important to screen out useful information.Machine learning is the core of machine intelligence,and some machine learning models have also brought unprecedented results in practical tests.As an important model of machine learning,Boltzmann machine has many applications in speech and image recognition,medical health and quantum multi-body problems.Boltzmann machine is a probability model.The depth model can be obtained through its stack.After combining Gibbs sampling and CD method,Boltzmann machine can quickly and effectively realize the training of various data types and complete the tasks of classification and regression.With the increasing demand for data processing,when dealing with large-scale tasks such as computer vision,image recognition and semantic transformation,the classical Boltzmann machine model has some problems,such as slow training speed,complex model and ineffective training.Some quantum algorithms can achieve exponential acceleration in computational complexity compared with classical algorithms,so quantum Boltzmann machine has the potential to solve these problems by introducing the characteristics of quantum.In the current era of big data,the research on quantum Boltzmann machine is of great significance.Firstly,this paper introduces the supervised model,unsupervised model,as well as some important algorithms needed in the training of these models,such as random gradient descent,back propagation.Then the classical Boltzmann machines such as fully connected Boltzmann machine,restricted Boltzmann machine,deep belief network,deep Boltzmann machine and their algorithms are introduced.This paper also introduces some quantum Boltzmann machine models,and combined with these models,the quantum semi restricted Boltzmann machine constructed in this paper is proposed.We use quantum semi restricted Boltzmann machinimagee to study the training and application of image recognition.The model constructs quantum visible and hidden units through Pauli operator,and trains the parameters based on the method of maximizing the lower bound of negative log likelihood function.Finally,the model is tested with handwritten digital image set and Bernoulli distribution set.We compare the training results with those on the classical restricted Boltzmann machine model,and the results show that our model is slightly higher in fidelity than the classical model.This shows the feasibility and application prospect of quantum algorithm in Boltzmann machine training.The research content of this paper is helpful for the understanding of machine learning model and Boltzmann machine training and optimization process.At the same time,it also provides a reference for the training of high-dimensional complex data with quantum Boltzmann machine in the future.