Summary of Bias Begets Bias: the Impact Of Biased Embeddings on Diffusion Models, by Sahil Kuchlous et al.
Bias Begets Bias: The Impact of Biased Embeddings on Diffusion Models
by Sahil Kuchlous, Marvin Li, Jeffrey G. Wang
First submitted to arxiv on: 15 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV); Computers and Society (cs.CY)
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 social biases in Text-to-Image (TTI) systems, focusing on diffusion models that generate images based on input prompts. The authors propose statistical group fairness criteria to evaluate the representation of the world within these models. They find that an unbiased text embedding space is necessary for representationally balanced diffusion models, which satisfy diversity requirements with respect to protected attributes. Additionally, they examine how biased embeddings affect evaluating the alignment between generated images and prompts, showing that biases can result in lower alignment scores for fair TTI models. The authors develop a theoretical framework to study and mitigate these biases, providing new fairness conditions for diffusion model development and evaluation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how text-to-image systems create pictures based on what people write. It’s like when you ask an AI to draw a picture of your favorite animal, but instead, it draws something that doesn’t look like the animal at all. The researchers want to make sure these systems are fair and don’t show biases against certain groups. They came up with new ways to measure fairness in these systems and showed that if the system is biased, it will create pictures that are not aligned with what people wrote. They’re working on fixing this problem so that AI can be more honest and unbiased. |
Keywords
» Artificial intelligence » Alignment » Diffusion » Diffusion model » Embedding space
