Controlling supercontinuum generation with tandem neural network–based models

White light continuum (WLC) generation in nonlinear media is governed by strongly nonlinear and coupled physical processes, making the selection of experimental parameters to achieve a desired output challenging. In our experiments, we generate WLC in an undoped YAG crystal using tunable subpicosecond, 1-µm laser pulses. We demonstrate a tandem neural network framework as an inverse model for WLC generation and find that the model successfully converges to reproduce desired outputs, despite nonunique experimental parameters. We further used the model to study the range over which we can optimize the system. These results highlight both the utility and limitations of neural-network–based inverse design for nonlinear optical systems and provide physical insight into WLC generation.