The field of smart photonics describes the use of advanced numerical algorithms and machine learning methods to create new functionality and improved performance in photonic systems. A particular area of interest is the use of machine learning techniques to determine the complex nonlinear transfer function of a multidimensional system, allowing its systematic control and optimization. One example of such a system is the generation of a broadband supercontinuum (SC) from nonlinear propagation of ultrashort pulses of light in optical fiber [1–3]. In particular, although supercontinuum sources are now well established in many fields with important applications, the laboratory optimization of SC spectral characteristics is usually performed by time-consuming trial and error, usually guided to some extent by computationally demanding numerical simulations of the underlying propagation model. Given that the potential parameter space is large, this approach is very inefficient.