A Quanvolution Architecture for Image Classification

Abstract

This paper proposes a hybrid quantum classic neural network with a fixed quantum circuit structure on image classification, which uses Quanvolution composed of quantum circuit with the super cube structure and randomized single qubit gates. Incorporating quantum circuits into convolutional neural net-works can introduce randomness to training through perturbing images with quantum circuits, acting as data augmentation to increase the model's accuracy. The experiment shows that the designed convolution neural network with quantum circuits has a 2% performance gain against the model without the Quanvolution.