Quantum neural networks based on angle-amplitude mixed encoding and its applications

Models combining traditional quantum neural networks with the self-attention mechanism require a high consumption of qubit resources. In response to the issues of low operational efficiency and high design complexity on current NISQ (noisy intermediate-scale quantum) devices, this paper proposes a mixed encoding method that embeds dataset features into quantum states in a specific manner, achieving an effective mix of angle encoding and amplitude encoding. Based on this encoding method, a uniquely structured double-ring Ansatz is designed. Drawing on the divide-and-conquer strategy of the self-attention mechanism, a quantum neural network with higher expressiveness is constructed. In the Iris classification task, the training loss converges to 0, proving that the model effectively captures the intrinsic relationships between the Iris features. In the text classification task, the classification accuracy improves by an average of 8.9% compared with existing methods, while also significantly reducing the number of training parameters without compromising performance. The lightweight and low-complexity characteristics of the quantum neural network based on angle-amplitude hybrid encoding make it more suitable for current NISQ devices.