Predictive Advantage of Quantum Machine Learning on Practically Relevant Problems

Determine whether quantum machine learning models—specifically variational quantum classifiers, quantum neural networks, and quantum kernel methods executed on near-term gate-based quantum computers—achieve a predictive accuracy advantage over classical machine learning algorithms on practically relevant datasets when using data encodings that are difficult to simulate classically.

Background

The paper surveys quantum machine learning architectures suitable for near-term quantum devices, including parameterized quantum circuits (variational quantum classifiers and broader quantum neural networks) and quantum kernel methods. These approaches rely on embedding classical data into quantum states and optimizing model parameters through hybrid quantum–classical loops.

While some evidence suggests quantum models may generalize from fewer data points than classical models, the authors explicitly note that establishing a predictive advantage on real-world, practically relevant tasks remains unresolved and likely depends on employing data encodings that are hard to simulate classically. This question is central to evaluating the utility of quantum methods in healthcare domains where datasets are often small or complex.