Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
TUE: Tuesday Contributed Sessions
TUE 11: Quantum Optics and Quantum Computation
TUE 11.2: Vortrag
Dienstag, 9. September 2025, 14:30–14:45, ZHG104
Training non-linear optical neural networks with Scattering Backpropagation — •Nicola Dal Cin1,2, Florian Marquardt1,2, and Clara Wanjura1 — 1Max Planck Institute for the Science of Light, Staudtstraße 2, 91058 Erlangen, Germany — 2Department of Physics, University of Erlangen-Nuremberg, 91058 Erlangen, Germany
As deep learning applications continue to deploy increasingly large artificial neural networks, the associated high energy demands are creating a need for alternative neuromorphic approaches. Optics and photonics are particularly compelling as they offer high speeds and energy efficiency. Neuromorphic systems based on non-linear optics promise high expressivity with a minimal number of parameters. However, so far, there is no efficient and generic physics-based training method with gradients for non-linear optical systems with dissipation. In this work, we present ``Scattering Backpropagation", the first efficient physics-inspired method for experimentally measuring approximated gradients for nonlinear optical neural networks. Remarkably, our approach does not require a mathematical model of the physical nonlinearity, and only involves two measurements of the system to compute all gradient approximations. In addition, the estimation precision depends on the deviation from reciprocity. We successfully apply our method to well-known benchmarks such as XOR and MNIST. Our method is widely applicable to existing state-of-the-art, scalable platforms, such as optics, microwave, and also extends to other physical platforms such as electrical circuits.
Keywords: Neuromorphic Computing; Nonlinear optics; Physical learning; Gradient Backpropagation; Non-reciprocity