A derivative-free deep neural network method for semi-linear elliptic and parabolic equations

Speaker: Dr. Mihai Nica (UofT)
Title: A derivative-free deep neural network method for semi-linear elliptic and parabolic equations
Abstract: We create a new numerical method for solving semi-linear elliptic and parabolic PDEs. The method solves the “weak form” of the equation using a connection to Brownian motion, and therefore does not explicitly compute derivatives. This makes it particularly suited for solving complicated boundary conditions on irregular boundaries. Based on joint work with J. Han and A. Stinchcombe.