Author: Sofiene Jerbi, Lukas J. Fiderer, Hendrik Poulsen Nautrup, Jonas M. Kübler, Hans J. Briegel, Vedran Dunjko

Journal: Nature Communications

Original Paper

This article presents a study on quantum machine learning models and their comparison to classical models. The authors propose a framework that captures all standard models based on parametrized quantum circuits, focusing on linear quantum models. They analyze the resource requirements and learning performance guarantees of these models, particularly comparing explicit and implicit models. They show that implicit models can achieve a lower training loss but may suffer from poor generalization performance. They also show that data re-uploading models, a type of explicit model, can be more general than both explicit and implicit models. The authors further investigate the advantages of explicit models by testing their performance on a learning task involving quantum-generated data. They find that explicit models can outperform both implicit models and classical models on this task, highlighting the potential learning advantage of explicit quantum models. The study provides insights into the capabilities and limitations of different quantum machine learning models, and it contributes to understanding the possible advantages of quantum models in practical applications.