Summary of Fairpivara: Reducing and Assessing Biases in Clip-based Multimodal Models, by Diego A. B. Moreira et al.
FairPIVARA: Reducing and Assessing Biases in CLIP-Based Multimodal Models
by Diego A. B. Moreira, Alef Iury Ferreira, Jhessica Silva, Gabriel Oliveira dos Santos, Luiz Pereira, João Medrado Gondim, Gustavo Bonil, Helena Maia, Nádia da Silva, Simone Tiemi Hashiguti, Jefersson A. dos Santos, Helio Pedrini, Sandra Avila
First submitted to arxiv on: 28 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 Despite significant advancements in vision-language models, there is a lack of research on their ethical implications. These models often rely on extensive training data from imbalanced datasets and raise concerns about biases. The CLIP model, initially trained in English, can be fine-tuned for other languages, introducing new biases. A CLIP-based Portuguese model, CAPIVARA, has shown strong zero-shot performance. This paper evaluates four types of discriminatory practices within visual-language models and proposes FairPIVARA, a method to reduce biases by removing feature embedding dimensions. Applying FairPIVARA reduces observed biases up to 98%, promoting a balanced word distribution. The authors’ model and code are available on GitHub. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at the ethical concerns surrounding vision-language models. These models need a lot of data to work well, but this data can be unbalanced and cause problems. When we fine-tune these models for other languages, it can introduce new biases. The researchers tested four different types of bias in these models and found that they all have discriminatory practices. To fix this, they developed FairPIVARA, a method to reduce these biases by changing how the model processes information. This helps to make the model more fair and balanced. |
Keywords
» Artificial intelligence » Embedding » Zero shot
