Prediction of Compilation Options for Quantum Circuits via Adaptive Deep Residual Neural Network

Quantum circuit compilation is an indispensable part of executing quantum circuits on quantum devices. It translates logical quantum circuits into physical quantum circuits that meet the device constraints. However, nonspecialist users require extensive expertise to effectively select and evaluate quantum circuit compilation processes. Therefore, this paper proposes a quantum circuit compilation option prediction method based on Deep Residual Attention Neural Network. The method aims to automate the selection of the best compilation scheme and enhance the efficiency and accuracy of quantum circuit compilation. Firstly, a quantum circuit feature extraction algorithm based on Time-Weighted Interaction Graph is proposed. This algorithm can effectively represent quantum circuits using fixed-length vectors, making the model input scalable. Secondly, for existing one-dimensional features, a quantum circuit compilation option prediction model based on Deep Residual Attention Neural Network is designed and trained. This model is capable of quickly and accurately predicting the optimal compilation option combination. Extensive experimental evaluations and comparisons are carried out on the MQT (Munich Quantum Toolkit) quantum circuit compilation dataset. These evaluations verify the effectiveness and advancement of the proposed method. Compared with the current state-of-the-art method (Random Forest), the proposed method improves the accuracy by 5.44%, the Top-3 accuracy by 2.6%, and the F1 by 2.05%.