Skip to content

fredwang25/DiME

Repository files navigation

Sample-efficient evidence estimation of score based priors for model selection (ICLR 2026)

Frederic Wang, Katherine L. Bouman

Model evidence of the April 6, 2017, M87* black hole observations under five different diffusion priors.

Model evidence of the April 6, 2017, M87 black hole observations under five different diffusion priors.

Webpage

Paper PDF

Getting Started

Install required packages with:

pip install -r requirements.txt

Download pretrained checkpoints here and place them in the checkpoints/ folder:

MNIST models

Black hole models

Run experiments and baselines with

python experiments/DIME_gaussian_mixture.py  
python experiments/evaluate_MNIST_gaussian.py
python experiments/evaluate_MNIST_fourier.py
python experiments/black_hole_evidence.py
python smc_baselines/MNIST_SMC_adaptive_sweep.py
python smc_baselines/MNIST_SMC_fixed_sweep.py

The training scripts can be found at train_blackhole64x64.py, train_mnist28x28.py, and a notebook with example unconditional sampling can be found at unconditional_sampling.ipynb.

Citation

@article{wang2026sample,
  title={Sample-efficient evidence estimation of score based priors for model selection},
  author={Wang, Frederic and Bouman, Katherine L},
  booktitle={International Conference on Learning Representations}
  year={2026}
}

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors