Boson Sampling finds the first practical application with quantum AI

For more than a decade, researchers have considered boson sampling (quantum computing protocols containing optical particles), an important milestone for demonstrating the advantages of quantum methods over classical computing. However, previous experiments have shown that boson sampling is difficult to simulate with classical computers, but practical applications remain out of reach.

now, Optica QuantumResearchers at the Okinawa Institute of Science and Technology (OIST) present the first practical application of boson sampling for image recognition, a critical task in many fields, from forensic medicine to medical diagnosis. Their approach uses only three photons and linear optical networks, marking an important step into low-energy quantum AI systems.

Utilize the complexity of quantum

Bourson-Photon-like particles following the Bourse-Einstein statistics eliminate complex interference effects when passed through a particular optical circuit. In Boson sampling, researchers inject a single photon into one such circuit and measure the output probability distribution after interference.

To understand how such sampling works, consider pegboard marbles. When the marble is dropped, a bell curve is formed when sampled the probability distribution where the marble lands. However, when we perform this same experiment using single photons, the results are completely different.

They display wave-like properties, which allow them to interfere with each other and interact with the environment very differently from large objects. This means displaying a highly complex probability distribution. This means that classical computing methods are difficult to predict.

From quantum reservoirs to image recognition

In this paper, the researchers developed a new quantum AI method for image recognition based on boson sampling. In the simulated experiments, they started by generating complex photonic quantum states, on top of which simplified image data was encoded.

The researchers used grayscale images from three different data sets as input. Because each pixel is grayscale, information can be easily expressed numerically and can be preserved using key component analysis (PCA).

This simplified data was encoded into a quantum system by adjusting the properties of a single photon. The photons then passed through quantum reservoirs (complex optical networks) that created interference-rich and high-dimensional patterns.

The detector recorded the photon positions and repeated sampling to construct a boson sampling probability distribution. This quantum output was combined with the original image data and processed by a simple linear classifier.

This hybrid approach stored information and outperformed all relatively sized machine learning methods tested by researchers, providing highly accurate image recognition across all datasets.

“The system may sound complicated, but in reality it's much easier to use than most quantum machine learning models,” explained Dr. Akitada Sakurai, the first author of the study and a member of the Quantum Information Science and Technology Unit.

“Only the final step (a simple linear classifier) ​​should be trained. In contrast, traditional quantum machine learning models typically require optimization across multiple quantum layers.”

“What's particularly noticeable is that this method works on a variety of image datasets without the need to modify the Quantum Reservoir, which is quite different from most traditional approaches,” said Professor William J Munro, co-author and director of the Quantum Engineering and Design Unit.

Unlock new frontiers with image recognition

Whether analyzing handwriting from a crime scene or identifying tumors with MRI scans, image recognition plays a key role in many real-world applications. The promising results of this study found that this quantum approach identifies images with higher accuracy than similarly sized machine learning methods, paving a new pathway for quantum AI.

“This system is not universal. We cannot solve all the computational problems we give,” said Professor Kae nemoto, director of the Quantum Information Science and Technology Unit, who is the center director of OIST Quantum Technologies and a co-author of the study.

“But it is an important step forward in quantum machine learning and we are excited to explore its possibilities in the future with more complex images.”