Photons learn to classify with a quantum edge

by Nicolas Harris & Darius Bunandar

The quantum nature of light has been harnessed in a photonic chip to perform machine-learning tasks. For specifically designed problems, the approach outperforms established classical methods.

Machine-learning algorithms have become indispensable tools, driving advances from scientific discovery to everyday application, although the insatiable appetite of these algorithms for computational power is a growing concern. Quantum machine-learning, however, offers a tantalizing prospect: leveraging the principles of quantum mechanics to tackle these demanding machine-learning tasks more efficiently or effectively than classical computers alone1,2. The promise of a broad quantum advantage over classical computation remains under intense investigation, but specific applications are beginning to show quantum-derived benefits. One such benefit is the demonstration, now presented in Nature Photonics by Yin and colleagues, of an experimental quantum kernel-based machine-learning protocol on an integrated photonic processor1, which showcases how quantum properties of single photons can provide a tangible performance boost.

At the heart of modern machine-learning are kernel methods. Imagine trying to separate two intertwined spirals of data points on a flat sheet of paper using only a straight line — it’s impossible. Kernel methods elegantly solve...

Check out the full article here.