Quantum advantage in learning from experiments

Area of Science:

• Quantum Information Science
• Computational Physics

Background:

• Conventional experiments rely on classical computers to process quantum data, which can be inefficient.
• Quantum technology offers a novel approach to data processing with potential for significant advantages.

Purpose of the Study:

• To demonstrate an exponential advantage of quantum machines in learning from experimental data compared to classical methods.
• To explore the feasibility of achieving quantum advantage with current quantum hardware.

Main Methods:

• Utilizing a quantum computer to process quantum data directly.
• Designing experiments for tasks including property prediction, quantum principal component analysis, and learning physical dynamics.
• Conducting experiments with 40 superconducting qubits and 1300 quantum gates.

Main Results:

• Quantum machines learned from exponentially fewer experiments than classical approaches.
• An exponential advantage was observed in predicting physical system properties, performing quantum principal component analysis, and learning physical dynamics.
• The quantum resources required for this advantage were found to be modest in certain scenarios.

Conclusions:

• Quantum computing provides a powerful new paradigm for scientific discovery, offering exponential speedups in learning from data.
• Substantial quantum advantage is attainable with existing quantum processors, paving the way for near-term applications.
• This work highlights the practical potential of quantum computation for advancing our understanding of the physical world.