Abstract

The numerical simulation of stochastic differential equations (SDEs), particularly through methods like Euler-Maruyama and Milstein schemes, often involves a trade-off between simulation accuracy and computational cost due to time discretization. In this paper, we propose a novel method for accelerating SDE simulations by leveraging temporal convolutional networks (TCNs) to efficiently learn the underlying stochastic dynamics. This approach enables parallel computation of drift and diffusion terms, significantly reducing the computational burden compared to traditional Monte Carlo simulation. We demonstrate that the output of our TCN-based model converges to the target SDE dynamics in probability and evaluate its performance on two well-known processes: geometric Brownian motion and the Ornstein-Uhlenbeck process. The results show that the TCN not only achieves accurate time-series and cross-sectional characteristics but also improves simulation speed while maintaining robust statistical properties of the original SDEs.