MLMCforDMs

Code for Shaw, Luke, Abdul-Lateef Haji-Ali, Marcelo Pereyra, and Konstantinos Zygalakis. "Bayesian computation with generative diffusion models by Multilevel Monte Carlo." Philosophical Transactions A 383, no. 2299 (2025): 20240333. (https://arxiv.org/abs/2409.15511) adapted from various sources.

Denoising

Adapted from https://github.com/yang-song/score_sde_pytorch associated to

@inproceedings{
  song2021scorebased,
  title={Score-Based Generative Modeling through Stochastic Differential Equations},
  author={Yang Song and Jascha Sohl-Dickstein and Diederik P Kingma and Abhishek Kumar and Stefano Ermon and Ben Poole},
  booktitle={International Conference on Learning Representations},
  year={2021},
  url={https://openreview.net/forum?id=PxTIG12RRHS}
}

Used under Apache-2.0 license.

Make sure that you download CIFAR10 to the appropriate directory (see main.py) and then run e.g.

python main.py --workdir exp/checkpoints/cifar10_ddpmpp_continuous --acc 0.0086 --config configs/vp/cifar10_ddpmpp_continuous.py --eval_folder exp/eval/cifar10DenoisingSecondmoment_0.8 --MLMCsampler EXPINT --probflow=False --conditional=True --payoff secondmoment --conditional_noise .8

Superresolution

Adapted from https://github.com/Janspiry/Image-Super-Resolution-via-Iterative-Refinement associated to

Paper | Project

Run e.g.

python MLMC.py -acc 0.018 -gpu 1

Inpainting

Adapted from https://github.com/NVlabs/I2SB associated to

@article{liu2023i2sb,
  title={I{$^2$}SB: Image-to-Image Schr{\"o}dinger Bridge},
  author={Liu, Guan-Horng and Vahdat, Arash and Huang, De-An and Theodorou, Evangelos A and Nie, Weili and Anandkumar, Anima},
  journal={arXiv preprint arXiv:2302.05872},
  year={2023},
}

Used under the licence: Copyright © 2023, NVIDIA Corporation. All rights reserved.

This work is made available under the Nvidia Source Code License-NC.

The model checkpoints are shared under CC-BY-NC-SA-4.0. If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original.

Run e.g.

python mlmc.py --dataset-dir datasetdir --ckpt inpaint-center --batch-size 12