A surrogate matrix-space training stage followed by adjoint-based physical realization allows scalable design of photonic neural networks that match ideal accuracy on image tasks with orders-of-magnitude fewer electromagnetic simulations.
High computational density nanophotonic media for machine learning inference
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
physics.optics 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Scalable Photonic Neural Networks via Surrogate Scattering-Matrix Inverse Design
A surrogate matrix-space training stage followed by adjoint-based physical realization allows scalable design of photonic neural networks that match ideal accuracy on image tasks with orders-of-magnitude fewer electromagnetic simulations.