Skip to content

PyTorch implementation for our ICLR 2024 paper "Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization"

License

Notifications You must be signed in to change notification settings

zdhNarsil/Diffusion-Generative-Flow-Samplers

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Diffusion Generative Flow Samplers

Diffusion Generative Flow Samplers (DGFS): Improving learning signals through partial trajectory optimization

Dinghuai Zhang, Ricky Tian Qi Chen, Cheng-Hao Liu, Aaron Courville, Yoshua Bengio.

We propose a novel DGFS sampler for continuous space sampling from given unnormalized densities based on stochastic optimal control 🤖 formulation and the probabilistic 🎲 GFlowNet framework.

pPOmv7T.png

target/ has the target distribution code. gflownet/ contains the DGFS algorithm code.

Examples

python -m gflownet.main target=gm dt=0.05
python -m gflownet.main target=funnel
python -m gflownet.main target=wells

Dependency

Apart from commonly used torch, torchvision, numpy, scipy, matplotlib, we use the following packages:

pip install hydra-core omegaconf submitit hydra-submitit-launcher
pip install wandb tqdm einops seaborn ipdb

About

PyTorch implementation for our ICLR 2024 paper "Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages