Computer Science

Shadows of quantum machine learning

S. Jerbi, C. Gyurik, et al.

Discover groundbreaking insights from the research conducted by the authors, who unveil a new class of quantum machine learning models that require quantum computing resources only during training. This innovative approach allows for classical deployment and shows a remarkable learning advantage over traditional methods, expanding the potential applications of quantum machine learning even with limited access to quantum resources.

00:00

Playback language: English

Index

Abstract

Quantum machine learning (QML) models, while promising, necessitate quantum computer access for both training and evaluation. This paper introduces a class of QML models where quantum resources are only needed during training, enabling classical deployment. These models generate a 'shadow model' allowing classical evaluation. The authors prove this model class is universal for classically-deployed QML, possesses restricted learning capacity compared to fully quantum models, yet achieves a provable learning advantage over classical learners under widely-believed complexity theory assumptions. This demonstrates QML's potential for broader applicability, even with limited quantum computing access.

Publisher

Nature Communications

Published On

Jul 06, 2024

Authors

S. Jerbi, Casper Gyurik, Simon Marshall, Riccardo Molteni, V. Dunjko