Prediction of ultrafast phenomena and nonlinear frequency conversion in hollow-core fibers using deep learning

In this presentation, we introduce FusionNet, a multi-modality deep learning network tailored to integrate and analyze temporal, frequency, and experimental parameters for predicting and understanding ultrafast nonlinear phenomena. Specifically, the model is benchmarked in gas-filled hollow-core fibers (HCFs), which serves as an ideal test that exploits unique nonlinear interactions – four-wave mixing, self-phase modulation, and cross-phase modulation – and advanced light-guiding principles to enable groundbreaking experiments in ultrafast science. Our method achieves an overall reduction in error by 73% and improves computing time by 83% compared to standard deep learning methods. Our work accelerates parametric simulations and experimental setup designs, including high-precision spectroscopy, quantum transduction, and distributed entangled interconnects.