Quantum Advantage in Learning from Experiments

Abstraction: Provable exponential quantum advantage in learning quantum systems

Key points:

• Published in Science (2022); collaboration between Google Quantum AI, Caltech, Harvard, Berkeley, and Microsoft
• A quantum ML (QML) agent achieves provable exponential advantage over the best classical ML algorithm on certain learning tasks about quantum systems — not merely over known classical algorithms
• Demonstrated on Google's Sycamore processor: QML needed 10,000× fewer measurements than classical ML to reach 70% prediction accuracy on a 20-qubit system
• Unlike computational advantage claims, this learning advantage cannot be overcome by unlimited classical computing resources when samples from the quantum state are limited
• QML works by storing quantum states and entangling multiple copies rather than measuring (collapsing) them immediately, extracting exponentially more information per copy
• A second experiment showed QML could unsupervised classify time-reversal symmetry of quantum operators where classical ML failed outright

Connections: Google · Caltech · Quantum Machine Learning · Quantum Computing

Source: https://ai.googleblog.com/2022/06/quantum-advantage-in-learning-from.html