Diffusion models for supervised learning
Title: Diffusion models for supervised learning
DNr: Berzelius-2024-58
Project Type: LiU Berzelius
Principal Investigator: Zheng Zhao <zheng.zhao@it.uu.se>
Affiliation: Uppsala universitet
Duration: 2024-03-01 – 2024-09-01
Classification: 10201


Diffusion models are a recent state-of-the-art framework for many machine learning problems, e.g., image processing. In this project, we plan to explore the fundamental connections between diffusion models and supervised learning methods, and see how could one accelerates the computation of the another. We also will demonstrate the method via image datasets. In order to carry out the computation, it is necessary for us to use GPUs because of the training of neural networks for diffusion models. We have demonstrated that we can effectively use the resources of Berzelius in our previous project under the number Berzelius-2023-194, and that we have published a paper in a top-tier machine learning conference.