Conditional Recurrent Neural Networks for broad applications in nonlinear optics

Abstract

We present a novel implementation of conditional Long Short-Term Memory Recurrent Neural Networks that successfully predict the spectral evolution of a pulse in nonlinear periodically-poled waveguides. The developed networks offer large flexibility by allowing the propagation of optical pulses with ranges of energies and temporal widths in waveguides with different poling periods. The results show very high agreement with the traditional numerical models. Moreover, we are able to use a single network to calculate both the real and imaginary parts of the pulse complex envelope, allowing for successfully retrieving the pulse temporal and spectral evolution using the same network.

References in corpus (5)

• Deep learning in nano-photonics: inverse design and beyond
• Artificial Neural Networks for Photonic Applications: From Algorithms to Implementation
• A feed-forward neural network as a nonlinear dynamics integrator for supercontinuum generation
• Mixing second and third-order nonlinear interactions in nanophotonic lithium-niobate waveguides
• Trapped Bose-Einstein condensates in the presence of a current nonlinearity