Deep Backward Dynamic Programming (DBDP) is an open-source Python library currently under active development designed for solving high-dimensional nonlinear partial differential equations (PDEs).
The project implements deep learning methods described in the research paper by Côme Huré, Huyên Pham, and Xavier Warin, leveraging the classical backward stochastic differential equation (BSDE) representation to efficiently approximate PDE solutions.
Clone the repository and install the required dependencies:
git clone https://github.com/RyanTmi/dbdp
cd dbdp
pip install -r requirements.txtdocs/ includes the original research paper.
notebooks/ contains Jupyter notebooks that replicate results from the research paper.
models/ provides pretrained networks and saved model checkpoints.
This project is collaboratively developed by :
- Ryan Timeus
- Paul-Antoine Charbit
- Jeremie Vilpellet