Quantum machine learning beyond kernel methods

Area of Science:

• Quantum Computing
• Machine Learning
• Quantum Information Theory

Background:

• Parametrized quantum circuits are key for near-term quantum machine learning.
• Current understanding of quantum machine learning model comparisons is limited.
• Various quantum machine learning models exist, but their relationships are unclear.

Purpose of the Study:

• To establish a unifying framework for quantum machine learning models.
• To compare resource requirements of different quantum machine learning models.
• To provide insights into model compatibility with Noisy Intermediate-Scale Quantum (NISQ) constraints.

Main Methods:

• Identification of a constructive framework: linear quantum models.
• Application of quantum information theory tools.
• Analysis of qubit number and data requirements for learning tasks.

Main Results:

• All standard parametrized quantum circuit models fit within the linear quantum model framework.
• Data re-uploading circuits can be efficiently mapped to linear models.
• Linear quantum models require exponentially more qubits than data re-uploading models for specific learning tasks.

Conclusions:

• A comprehensive view of quantum machine learning models is provided.
• Insights into resource efficiency for NISQ devices are offered.
• The study clarifies relationships between different quantum machine learning approaches.