Speaker
Description
Artificial neural networks are well suited tools for solving complex, multi-parameter problems and have been realized using light, termed optical neural networks (ONNs) or deep diffractive neural networks (D2NN). Their conception has been motivated by the decreased energy demands of optical systems, where nature itself is able to take on some of the computational burden. These systems have demonstrated the ability to tackle an array of tasks, including the classification of light with encoded spatial patterns. Scalability, however, remains a challenge, particularly when dealing with large numbers of spatial modes or non-orthogonal input states. Here, we demonstrate a compact and reprogrammable D2NN capable of classifying and detecting orbital angular momentum (OAM) modes, including modes from non-orthogonal sets. The network is implemented using trainable phase modulation layers optimized via stochastic gradient descent. We evaluate the implementation through numerical simulation and present the results. This method underscores the potential for hybrid classical-quantum optical systems to provide a framework for scalable, cost-effective, low-latency platforms for quantum inspired computations.
| Apply for student award at which level: | MSc |
|---|---|
| Consent on use of personal information: Abstract Submission | Yes, I ACCEPT |