Summary of Improving Diffusion-based Generative Models Via Approximated Optimal Transport, by Daegyu Kim et al.
Improving Diffusion-Based Generative Models via Approximated Optimal Transport
by Daegyu Kim, Jooyoung Choi, Chaehun Shin, Uiwon Hwang, Sungroh Yoon
First submitted to arxiv on: 8 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The Approximated Optimal Transport (AOT) technique is a novel training scheme for diffusion-based generative models that aims to integrate optimal transport into the training process, improving the accuracy of denoiser outputs and reducing truncation errors during sampling. This approach achieves superior image quality and reduces sampling steps by employing AOT in training. The paper reports FID scores of 1.88 with 27 NFEs and 1.73 with 29 NFEs for unconditional and conditional generations, respectively. When applying AOT to train the discriminator, it establishes new state-of-the-art FID scores of 1.68 and 1.58 for unconditional and conditional generations, respectively, each with 29 NFEs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Diffusion models are a type of machine learning algorithm that can generate images. The authors of this paper created a new way to train these models called Approximated Optimal Transport (AOT). AOT makes the models better at generating realistic images and reduces the number of steps needed to do so. This is important because it means we can get good results with fewer calculations, which can be useful for large datasets. |
Keywords
* Artificial intelligence * Diffusion * Machine learning
