Likelihood training of Schrödinger bridge using Forward-backward SDEs theory

This is the repository associated with the project for the MVA course Generative Modelling.

This is a group project realised by :

Benjamin Maurel
Jérémie Stym-Popper
Gabriel Watkinson

Don't hesitate to contact any of us if you have any questions.

This experiments presented here and in the report comes from modification of https://github.com/ghliu/DeepGSB.

Installation

Clone the repository.

git clone https://github.com/gwatkinson/mva_sb_generative

Install the project and dependencies, creating a virtual environment with conda

cd mva_sb_generative
conda env create --file requirements.yaml
conda activate deepgsb

Reproduce the experiments

To reproduce the experiments, you can run the following commands:

cd DeepGSB
bash run.sh Evac

Results

The report can be found in the root of the repo in report.pdf.

The slides of the presentation are also available in the root of the repo in presentation.pdf

Lastly, some images of our experiments can be found in the images folder.

Those images show the impact of an obstacle in front of a door, simulating an evacuation or a crowd movement.

Credits

@inproceedings{liu2022deep,
  title={Deep Generalized Schr{\"o}dinger Bridge},
  author={Liu, Guan-Horng and Chen, Tianrong and So, Oswin and Theodorou, Evangelos A},
  booktitle={Advances in Neural Information Processing Systems},
  year={2022}
}

@inproceedings{chen2022likelihood,
  title={Likelihood Training of Schr{\"o}dinger Bridge using Forward-Backward SDEs Theory},
  author={Chen, Tianrong and Liu, Guan-Horng and Theodorou, Evangelos A},
  booktitle={International Conference on Learning Representations},
  year={2022}
}