Summary of Heavy-tailed Diffusion Models, by Kushagra Pandey et al.
Heavy-Tailed Diffusion Models
by Kushagra Pandey, Jaideep Pathak, Yilun Xu, Stephan Mandt, Michael Pritchard, Arash Vahdat, Morteza Mardani
First submitted to arxiv on: 18 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 This paper investigates the ability of diffusion models to capture rare or extreme events in heavy-tailed distributions. Currently, state-of-the-art generation quality is achieved by diffusion models across many applications, but their performance in capturing heavy-tailed behavior remains unclear. The authors find that traditional diffusion and flow-matching models with standard Gaussian priors fail to capture heavy-tailed behavior. To address this, the paper introduces a tailored perturbation kernel and derives the denoising posterior based on the conditional Student-t distribution for the backward process. This framework enables controllable tail generation using only a single scalar hyperparameter, making it easily tunable for diverse real-world distributions. The authors also introduce t-EDM and t-Flow, extensions of existing diffusion and flow models that employ a Student-t prior. Experimental results show that these new frameworks outperform standard diffusion models in heavy-tail estimation on high-resolution weather datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us better understand how machines can generate realistic images and data. Currently, these models are really good at making things look normal, but they struggle with creating rare or extreme events. The researchers found that the traditional way of doing this doesn’t work well for unusual events. To solve this problem, they came up with a new approach that uses something called a Student-t distribution. This allows them to control how often these rare events happen and make it easier to use in real-world situations. They also developed two new models, t-EDM and t-Flow, which are better at capturing heavy-tailed distributions than the old way of doing things. |
Keywords
» Artificial intelligence » Diffusion » Hyperparameter
