1: \begin{abstract}

2: We report two methods for solving FBSDEs of path dependent types of high dimensions. Inspired by the work from \cite{haojie}, \cite{weinan1} and \cite{lyons1}, we propose a deep learning framework for solving such problems using path signatures as underlying features. Our two methods (forward/backward) demonstrate comparable/better accuracy and efficiency compared to the state of the art \cite{qifeng1}, \cite{qifeng2} and \cite{come1}. More importantly, leveraging the techniques developed in \cite{lyons5} we are able to solve the problem of high dimension which is an limitation in \cite{qifeng1} and \cite{qifeng2}. We also provide convergence proof for both methods with the proof of the backward methods inspired from \cite{ruimeng} in the Markovian case. 

3: \end{abstract}

4: