-
Date: 30 Jul. (Thu.)
-
Place: On the Web
-
Time: 16:30-18:00
-
Speaker: Noufel Frikha (Paris VII)
-
Title: Yet another learning algorithm for Backward Stochastic Differential Equations
-
Abstract:
Backward Stochastic Differential Equations are stochastic processes that allow to represent the solution of semi-linear PDEs.
They are used to design numerical probabilistic methods for these PDEs. Recently, learning methods have proven to be successful in computing the solution of such equation in high dimensional settings.
We present a learning algorithm based on a stochastic gradient descent method and the use of sparse grids. We are able to prove precise rate of convergence for this numerical method, which allows to tame the curse of dimensionality in a smooth setting. We illustrate our theoretical results with some numerical experiments.
Joint work with J.-F. Chassagneux, J. Chen and C. Zhou.
日本語
English