Posted in | News | Optics and Photonics | Optical Communications

Photonic Quantum CNN Achieves High Accuracy With Adaptive Light Control

Download PDF Copy

Reviewed

Reviewed by Frances BriggsNov 27 2025

A new adaptive technique helps quantum photonic processors make real-time decisions, a key step on the path to practical quantum machine learning.

Study: Photonic quantum convolutional neural networks with adaptive state injection. Image Credit: saicle/Shutterstock.com

In a study published in Advanced Photonics, researchers at SPIE have introduced a method to enhance photonic circuit adaptability without compromising compatibility with current technologies.

The study incorporates adaptive state injection, a controlled step that enables the circuit to adjust its behavior based on measurements taken during processing. This additional control advances photonic QCNNs closer to practical implementation.

Machine learning models, specifically convolutional neural networks (CNNs), drive technologies such as image recognition and language translation. A quantum counterpart, the quantum convolutional neural network, could process information more efficiently by using quantum states instead of classical bits.

Photonic systems are a promising platform for quantum CNNs, given the speed, stability, and ease of manipulation of photons on chips.

However, photonic circuits typically exhibit linear behavior, which limits the flexible operations required by neural networks.

The research team constructed a modular QCNN using single photons from a quantum-dot source and two integrated quantum photonic processors. Similar to classical CNNs, this network processes information in stages. After the initial stage, a portion of the light signal is measured.

Depending on the result, the system either injects a new photon or forwards the existing light, thereby guiding the computation.

Since current photonic hardware cannot switch light in real time without information loss, the researchers simulated this step in the lab using a controlled technique to reproduce the same effect.

The scientists encoded simple 4 × 4 images, specifically patterns of horizontal or vertical bars. Measurements at each stage aligned with theoretical predictions. In the full experimental setup, the quantum CNN achieved a classification accuracy exceeding 92 %, consistent with numerical simulations.

The researchers also explored scalability, noting future photonic devices with fast switching capabilities could enable larger, more powerful quantum CNNs that outperform some classical methods.

This work provides both a theoretical framework and a proof-of-concept implementation of a photonic QCNN. We expect these results to serve as a starting point for developing new quantum machine learning methods.

Fabio Sciarrino, Study Senior Author and Associate Professor, Sapienza Univ di Roma

By integrating a simple adaptive step compatible with existing technology, this study outlines a realistic path toward developing more capable photonic quantum processors.