A deep shotgun method for solving high-dimensional parabolic partial differential equations

Recent advances in deep learning makes solving parabolic partial differential equations (PDEs) in high dimensional spaces possible via forward-backward stochastic differential equation (FBSDE) formulations. The implementation of most existing methods requires simulating multiple trajectories of stochastic processes with a small step size of time discretization to ensure accuracy, hence having limited performance, especially when solving on a large time interval. To address such issue, we propose a deep "shotgun method" that does not exploit full trajectories, but only utilizes the data distribution of them. Numerical results including examples with dimensionality up to 10000 demonstrate the competitiveness of the proposed shotgun method in both performance and accuracy.