Quantum Image Compression With Autoencoders Based On Parameterized Circuits

In the last decade,the performance of computers has been greatly improved.As a result,the machine learning technology has been rapidly developed as a powerful tool,which is widely used in pattern recognition,statistical analysis and optimization.The optimization and improvement of technology,followed by the explosion of information,resulting in the growth of traditional computer performance is becoming more and more difficult.Quantum computing is proposed to solve the bottleneck of classical computers.For the reason of recent quantum devices,the capacity of quantum resources is limited.Therefore,in order to find a common tool to reduce the use of valuable quantum resources,in the application of quantum simulation,quantum communication,and distributed computing in quantum networks.In this paper,a hybrid quantum-classical autoencoder for image compression is proposed.The specific research scheme is as follows:This work proposes quantum autoencoder,which combines deep learning with quantum computing.It uses hybrid quantum-classical algorithm and optimal control optimization algorithm to learn effective low-dimensional representation of quantum data in high-dimensional space,and applies it to compressed picture.In the process of compression,quantum autoencoder needs to discard certain qubit number,so the compression process of quantum autoencoder is a lossy process.Quantum autoencoder uses an improved hybrid quantum-classical algorithm.Computing objective functions and gradients in quantum devices and updating parameters in classical computers.Until the optimal control scheme is found,the optimal control search process is implemented.Quantum encoder and decoder adopt programmable parameterized quantum circuit,which is composed of fixed quantum logic gate combination,and the fixed quantum logic gate combination can be used repeatedly to increase the flexibility of quantum circuit.When the quantum autoencoder compresses 16*16 MNIST handwritten data set,the mean fidelity of the data set can reach 95%.Therefore,the experiments show that the quantum autoencoder has good learning ability and generalization ability.Compared with the classical autoencoder,it is found that the quantum autoencoder has better generalization ability than the classical autoencoder.