A collaborative research effort between teams in Italy, France, and the U.K. has developed a method to make photonic circuits more adaptable, while retaining their current technological compatibility. The researchers added a controlled step, called adaptive state injection, which allows a circuit to adjust its behavior based on a measurement taken during processing. According to the team, the extra control moves photonic quantum convolutional neural networks (QCNNs) closer to practical use. Convolution neural networks (CNNs) are used to power technologies such as image recognition and language translation. QCNNs may provide advantages in processing efficiency by using quantum states instead of classical bits. Photons are fast, stable, and easy to manipulate on chip, making them a promising platform for QCNNs. However, the linear behavior of photonic circuits limits the flexible operations that neural networks need. A new approach to photonic neural networks incorporates adaptive photon injection during the pooling stage. Courtesy of L. Monbroussou et al., doi 10.1117/1.AP.7.6.066012. The researchers built a modular QCNN using single photons from a quantum-dot source and two integrated quantum photonic processors that process information in stages. After the first stage, part of the light signal is measured. Depending on the result, the system either injects a new photon or sends the existing light forward, gently steering the computation. Since current photonic hardware can't switch light in real time without losing information, the researchers emulated this step in the lab using a controlled technique that reproduces the same effect. The system was tested using 4 × 4 images of horizontal or vertical bar patterns. Measurements at each stage matched theoretical predictions. In the full experimental setup, the QCNN achieved a classification accuracy above 92%, consistent with numerical simulations. The researchers also explored scalability, noting that future photonic devices with fast switching could enable larger, more powerful QCNNs that outperform some classical methods. “This work provides both a theoretical framework and a proof-of-concept implementation of a photonic QCNN,” said senior author Fabio Sciarrino. “We expect these results to serve as a starting point for developing new quantum machine learning methods.” This research was published in Advanced Photonics (www.doi.org/10.1117/1.AP.7.6.066012).