A Generative Model that outperform GANs in terms of compute and benchmarks.
Predecessor of many state of the art generative models such as DALLE and Stable Diffusion.
-
Learning: Changing loss function and variance formula
-
Cosine Schedule
-
Faster Sampling by changing number of timesteps
-
Scalable transformer for diffusion
-
Importance Sampling
@article{ho2020denoising,
title={Denoising diffusion probabilistic models},
author={Ho, Jonathan and Jain, Ajay and Abbeel, Pieter},
journal={Advances in neural information processing systems},
volume={33},
pages={6840--6851},
year={2020}
}@inproceedings{nichol2021improved,
title={Improved denoising diffusion probabilistic models},
author={Nichol, Alexander Quinn and Dhariwal, Prafulla},
booktitle={International conference on machine learning},
pages={8162--8171},
year={2021},
organization={PMLR}
}@inproceedings{peebles2023scalable,
title={Scalable diffusion models with transformers},
author={Peebles, William and Xie, Saining},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={4195--4205},
year={2023}
}