Boson Sampling Achieves First Practical Application in Quantum Artificial Intelligence}

Editor | Baicai Ye

Boson sampling is a restricted non-universal quantum computing model that can theoretically solve problems beyond classical computers, with its quantum advantage already confirmed. However, practical applications have remained elusive.

In recent research, scientists from the Okinawa Institute of Science and Technology (OIST) demonstrated that using a random interferometer to drive boson sampling can generate complex dynamics necessary for quantum reservoir computing. They applied these dynamics to various image recognition tasks, proving the method's practicality even in medium-scale systems.

Their approach uses only three photons and a linear optical network, marking a significant step toward low-energy quantum AI systems.

The related study titled "Quantum optical reservoir computing powered by boson sampling" was recently published in Optica Quantum, 2025, Vol. 3, Issue 3.

Using Quantum Complexity

Bosons—particles similar to photons that follow Bose-Einstein statistics—exhibit complex interference effects when passing through certain optical paths. In boson sampling, researchers inject single photons into such paths and measure their output probability distributions after interference.

Imagine marbles dropping onto a pegboard. Sampling the landing positions forms a bell-shaped distribution. But with single photons, the results are vastly different, showing wave-like interference and highly complex probability distributions that are hard to predict with classical methods.

From Quantum Reservoirs to Image Recognition

Researchers developed a new quantum AI image recognition method called QORC based on boson sampling. In simulations, they generated complex photon quantum states and encoded simplified image data onto them.

They used grayscale images from three datasets, with pixel values compressed via PCA to retain key features. These data were encoded into the quantum system by adjusting photon properties. The photons then pass through a complex optical network, creating high-dimensional interference patterns.

Detectors record photon positions, and repeated sampling constructs the boson sampling probability distribution. This quantum output is combined with the original image data and processed by a simple linear classifier.

This hybrid approach preserves information and outperforms all tested machine learning methods of similar scale, achieving high accuracy in image recognition across datasets.

Figure: QORC schematic diagram. (Source: the paper)

"Although this system sounds complex, it is much simpler to use than most quantum machine learning models," explains Dr. Akitada Sakurai, the first author and member of the quantum information science department.

"Only the final step—a simple linear classifier—requires training. Traditional quantum machine learning models often need optimization across multiple quantum layers."

Professor William John Munro, co-author and head of the quantum engineering department, added: "Most notably, this method works across various image datasets without modifying the quantum reservoir, unlike traditional approaches that require dataset-specific tuning."

Opening a New Era in Image Recognition

Image recognition plays a vital role in applications like forensic handwriting analysis and tumor detection in MRI scans. This research shows that quantum methods can outperform similar-scale classical machine learning in accuracy, opening new avenues for quantum AI.

"This system is not universal—it cannot solve every computational problem we pose," states Professor Kae Nemoto, director of the OIST Quantum Technology Center and co-corresponding author. "But it marks an important step forward for quantum machine learning, and we look forward to exploring its potential on more complex images."

Paper link: https://opg.optica.org/opticaq/fulltext.cfm?uri=opticaq-3-3-238&id=572317

Related report: https://phys.org/news/2025-06-boson-sampling-applications-quantum-ai.html