Deep Learning based Backward Stochastic Differential Equation

The repo is a Tensorflow implementation of deep learning based BSDE.

• Two versions of algorithms, based on paper Solving high-dimensional partial differential equations using deep learning and Machine Learning for Semi Linear PDEs respectively, have been implemented.
• A Black-Schole Call option BSDE has been added as an example.
• A ModelRunner class is added to control the pipeline of model training and evaluation: evaluate the performance on testset only when the lowest-by-far validation loss has been achieved.

How to Run

By default we use method mentioned in Solving high-dimensional partial differential equations using deep learning

python main.py 

If we want to use the method mentioned in Machine Learning for Semi Linear PDEs,

python main.py --network_type merged

Requirement

tensorflow==1.13.1

Reference

• DeepBSDE