Description
Future fusion devices face a critical challenge in managing extreme heat loads at the plasma edge and plasma-facing components. While advanced kinetic simulation codes can model non-local heat transport and plasma-wall interactions, solving high-dimensional partial differential equations, such as the Vlasov-Fokker-Planck equation computationally expensive. A common approach to solving this problem is using standard Neural Network (NN) surrogates; however, here we introduce a Gaussian Process (GP) regression framework to accelerate these non-local transport simulations. Compared to standard NNs, the GP approach provides a highly flexible, non-parametric method that intrinsically quantifies predictive uncertainty. Furthermore, the GP surrogate requires significantly less training data while maintaining high fidelity to the underlying kinetic physics. This framework presents a robust alternative pathway for the fast, predictive modeling of edge plasma transport, providing essential statistical confidence alongside its predictions.