The team constructed a modular quantum convolutional neural network (QCNN) utilizing single photons from a quantum-dot source and two integrated quantum photonic processors. The system processes data in stages similar to classical convolutional neural networks. After the first stage, part of the light signal is measured. Depending on this result, the arrangement either injects a new photon or transmits the signal onward, steering computational outcomes as needed.
Because today's photonic devices cannot reliably switch light in real time without loss, the researchers emulated the adaptive step in the laboratory using a carefully controlled method to achieve the theoretical effect. Testing involved encoding simple 4 + 4 images - horizontal and vertical bar patterns - within the quantum neural network. The experimental results matched theoretical expectations, and the QCNN achieved over 92 percent classification accuracy, consistent with numerical models.
The approach demonstrates scalability prospects for quantum photonic systems. The researchers report that future hardware with rapid switching could permit larger QCNNs capable of outperforming classical machine learning for some tasks.
"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."
The addition of an adaptive step, which is feasible with current photonic technology, may further the advancement of practical quantum processors for artificial intelligence and data processing.
Research Report:Photonic quantum convolutional neural networks with adaptive state injection
| The content herein, unless otherwise known to be public domain, is Copyright 1995-2026 Space Media Network. All websites are published in Australia and are solely subject to Australian law and governed by Fair Use principles for news reporting and research purposes. |
