The power of quantum neural networks

Area of Science:

• Quantum computing
• Machine learning
• Statistical modeling

Background:

• The potential advantage of near-term quantum computers for machine learning remains an open question.
• Assessing the power and trainability of quantum machine learning models against classical counterparts is crucial.

Purpose of the Study:

• To investigate if quantum machine learning models offer advantages over classical neural networks.
• To introduce and validate a new metric, the effective dimension, for evaluating model generalization and trainability.

Main Methods:

• Proposed the 'effective dimension' as a data-dependent measure utilizing Fisher information.
• Compared the effective dimension of quantum neural networks with classical feedforward networks.
• Validated findings numerically and on real quantum hardware.

Main Results:

• Quantum neural networks achieved a significantly better effective dimension than comparable classical feedforward networks.
• Quantum machine learning models demonstrated faster training times.
• The effective dimension proved to be a reliable metric for assessing generalization ability.

Conclusions:

• Near-term quantum computers show a potential advantage for machine learning tasks.
• The proposed effective dimension metric effectively quantifies model performance and trainability.
• Quantum machine learning is a promising area for future research and application.