Stochastic Volterra integral equations model random systems where the current state depends on the entire past. A new preprint on arXiv investigates these equations from two angles: a Wong-Zakai convergence theorem and a Fourier neural operator (FNO) approach.

Wong-Zakai convergence is a technique for approximating rough stochastic noise with smoother processes. The paper proves such a convergence result for SVIEs, giving a pathwise justification for replacing noise processes in these memory-dependent systems.

Separately, the authors train Fourier neural operators to learn the solution map of the SVIEs. FNOs are a class of deep learning models designed for equations that depend on spatial or temporal structure, and here they are applied to random systems with memory.

Because the abstract is short, the full scope of the numerical results is not available from the source alone. The contribution appears to be in linking a rigorous convergence theorem with a data-driven operator learning method.