Summary of Em Distillation For One-step Diffusion Models, by Sirui Xie et al.
EM Distillation for One-step Diffusion Models
by Sirui Xie, Zhisheng Xiao, Diederik P Kingma, Tingbo Hou, Ying Nian Wu, Kevin Patrick Murphy, Tim Salimans, Ben Poole, Ruiqi Gao
First submitted to arxiv on: 27 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
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 proposed EM Distillation (EMD) approach enables efficient sampling from complex distributions using a maximum likelihood-based method that distills a diffusion model to a one-step generator model while maintaining perceptual quality. This is achieved by updating the generator parameters through an Expectation-Maximization algorithm, which incorporates samples from the joint distribution of the diffusion teacher prior and inferred generator latents. The approach also includes a reparametrized sampling scheme and noise cancellation technique that stabilizes the distillation process. EMD outperforms existing one-step generative methods in terms of FID scores on ImageNet-64 and ImageNet-128, and compares favorably with prior work on distilling text-to-image diffusion models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary EM Distillation is a new way to make computers generate images or other data without needing a lot of computer power. This is useful because it can help us create more realistic pictures or sounds. The method uses an old idea called Expectation-Maximization to figure out how to make the generator model work better. It also includes some special tricks to keep the process stable and working well. EM Distillation works better than other methods that try to do similar things, especially when we want to create images or data that are very realistic. |
Keywords
» Artificial intelligence » Diffusion » Diffusion model » Distillation » Likelihood
