First Photon Machine Learning

Quantum techniques are expected to revolutionize how information is acquired, exchanged, and processed. Yet it has been a challenge to realize and measure their value in practical settings. We present first photon machine learning as a new paradigm of neural networks and establish the first unambiguous advantage of quantum effects for artificial intelligence. By extending the physics behind the double-slit experiment for quantum particles to a many-slit version, our experiment finds that a single photon can perform image recognition at around $30\%$ fidelity, which beats by a large margin the theoretical limit of what a similar classical system can possibly achieve (about 24\%). In this experiment, without count the energy consumed for drive spatial light modulators, the entire neural network is implemented in sub-attojoule optics and the equivalent per-calculation energy cost is below $10^{-24}$ joule, highlighting the prospects of quantum optical machine learning for unparalleled advantages in speed, capacity, and energy efficiency.