R-CCS Cafe 第287回（2026年3月2日）

Training a Quantum Machine Learning model on real quantum hardware

Quantum Machine Learning (QML) aims to study the possible ways to combine Machine Learning methods with Quantum Computing. We focus here on using the higher expressivity of quantum circuits to improve the quality of the final model. Although promising for some applications, training a QML model requires too many quantum circuit evaluations to be feasible on real hardware, even for very small datasets. As such, these methods remain confined to simulators, which limits their ability to generalize. Post-Variational Quantum Neural Networks (QNN) are a recent type of algorithm which require significantly fewer circuit evaluations compared to other methods. We implemented this algorithm on a RIKEN's HPC-QC system using the Fugaku supercomputer linked to two different quantum machines (superconducting and ion-trap), and showed that Post-Variational QNNs can be trained in reasonable time on real hardware.