AutoML-driven optimization of variational quantum circuit

Variational Quantum Circuits (VQCs) offer a powerful framework for quantum machine learning models, where circuit parameters are optimized to learn specific tasks. Quantum architecture search refines VQCs by automating the design of circuit structures. Automated Machine Learning (AutoML) automates model selection, hyperparameter tuning, and optimization, enhancing accessibility for nonexperts and improving efficiency. In this work, we propose an AutoML-driven approach that automates not only the optimization of VQC structures and parameters but also the tuning of training settings, an aspect overlooked in previous studies. We pretrain a graph neural network on a large, unlabeled dataset to learn quantum circuit embeddings. The pretrained model is then fine-tuned on a small set of labeled data from a downstream task to develop a performance predictor that estimates the performance of quantum circuits based on their structures and training settings. This enables us to rank abundant circuit structures and training settings, effectively identifying the optimal configurations for a given task. Numerical experiments demonstrate a strong correlation between the true and predicted performance, as well as its efficiency in VQC optimization. These results highlight the potential of AutoML to improve both the performance and efficiency of VQCs in quantum machine learning applications.