Quantum Squeeze-and-Excitation Networks

In this paper, we introduce Quantum Squeeze-and-Excitation (QSE) Networks, a pioneering approach within the domain of quantum computing designed to enhance the excitation module of classical Squeeze-and-Excitation (SE) networks. Our method significantly enhances performance by leveraging quantum computing techniques while simplifying the model's complexity. Neural network data encoding is performed through quantum amplitude coding, substantially reducing the parameter count of the classical SE network's fully connected layers. Experimental results show that, after 100 training rounds, the accuracy of our proposed QSE ResNet-18 on the CIFAR-10 data set reached 82.70%, while the classical SE ResNet-18 was only 82.20%. At the same time, on the CIFAR-100 data set, the top-5 error of QSE ResNet-50 is only 18.14%, while the classic SE ResNet-18 is 20.28%. In addition, our parameters are reduced by 0.4% compared to classic SE ResNet-18 and 4.8% compared to classic SE ResNet-50, respectively. In the analysis of quantum noise, the CIFAR-10 accuracy of QSE ResNet-18 under different noise models fluctuates around 0.4%.