Towards Automated Selection of Quantum Encoding Circuits via Meta-Learning

TL;DR

This paper introduces an automated meta-learning approach to predict the best quantum encoding circuit for a dataset, reducing costly evaluations on near-term quantum devices.

Contribution

It presents a recommender system that uses classical data complexity metrics and machine learning to select quantum encoding circuits without quantum evaluations.

Findings

Achieves up to 85.7% Top-3 accuracy in circuit selection.

Classical complexity metrics are sufficient for accurate predictions.

Two training approaches and four configurations were evaluated.

Abstract

In recent years, quantum kernel methods have shown promising applications on near-term quantum devices. However, selecting an appropriate encoding circuit for a given dataset requires costly evaluation of multiple candidates, formulated as a meta-learning problem. In this paper, we propose an automated recommender that utilizes the intrinsic characteristics of datasets to predict the optimal circuit without any quantum evaluation. Nine candidates are assessed alongside 24 classical complexity metrics serving as features, evaluated through two training approaches with four configurations, along with 14 machine learning models. Both approaches achieve Top-3 accuracy of up to 85.7% in identifying the best-performing encoding circuit, and demonstrate that classical data complexity metrics provide sufficient predictive signal for circuit selection.