Summary of You Never Know: Quantization Induces Inconsistent Biases in Vision-language Foundation Models, by Eric Slyman et al.
You Never Know: Quantization Induces Inconsistent Biases in Vision-Language Foundation Models
by Eric Slyman, Anirudh Kanneganti, Sanghyun Hong, Stefan Lee
First submitted to arxiv on: 26 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Computers and Society (cs.CY); 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 A medium-difficulty summary of the abstract is as follows: The paper investigates how a common practice in compressing foundation vision-language models, known as quantization, affects their ability to produce socially-fair outputs. Unlike previous findings with unimodal models that compression consistently amplifies social biases, this study’s extensive evaluation across four quantization settings, three datasets, and three CLIP variants reveals a surprising result: while individual compressed models exhibit bias, there is no consistent change in the magnitude or direction of bias across a population of compressed models due to quantization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A low-difficulty summary of the abstract is as follows: This study looks at how compressing special computer models called vision-language models affects their ability to make fair decisions. The researchers found that when they used a common method to shrink these models, it didn’t make them more or less unfair on average. However, some individual models were still biased in certain ways. |
Keywords
* Artificial intelligence * Quantization
