Multimodal Deep Representation Learning for Quantum Cross-Platform Verification.

Area of Science:

• Quantum Computing
• Machine Learning
• Quantum Information Science

Background:

• Cross-platform verification is crucial for early-stage quantum computing, comparing imperfect quantum devices with minimal measurements.
• The random measurement approach faces scalability challenges due to quasi-exponential computational costs with increasing qubit count.
• Existing methods struggle with large-qubit systems, necessitating advanced techniques.

Purpose of the Study:

• To introduce an innovative multimodal learning approach for cross-platform verification in quantum computing.
• To address the limitations of existing methods in handling large-qubit quantum systems.
• To enhance the characterization of similarity between quantum devices.

Main Methods:

• Developed a multimodal neural network to process two distinct data modalities: measurement outcomes and classical circuit descriptions.
• Independently extracted knowledge from each modality.
• Fused the extracted knowledge to create a comprehensive data representation for device comparison.

Main Results:

• Achieved a 3-orders-of-magnitude improvement in prediction accuracy compared to random measurements.
• Demonstrated effective characterization of quantum device similarity for new quantum algorithms.
• Evaluated the approach on systems up to 50 qubits with diverse noise models.

Conclusions:

• Multimodal learning offers a powerful solution for cross-platform verification in quantum computing.
• The complementary roles of measurement outcomes and circuit descriptions are vital for accurate device comparison.
• This approach paves the way for broader applications of machine learning in quantum system analysis.