Summary of Social Debiasing For Fair Multi-modal Llms, by Harry Cheng et al.
Social Debiasing for Fair Multi-modal LLMs
by Harry Cheng, Yangyang Guo, Qingpei Guo, Ming Yang, Tian Gan, Liqiang Nie
First submitted to arxiv on: 13 Aug 2024
Categories
- Main: Computation and Language (cs.CL)
- 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 The paper tackles the issue of social biases in Multi-modal Large Language Models (MLLMs), which often inherit biases from their training datasets. To address this, the authors introduce a new Counterfactual dataset with Multiple Social Concepts (CMSC) and an Anti-Stereotype Debiasing strategy (ASD). The CMSC dataset provides a more diverse and extensive training set compared to existing datasets, while ASD works by revisiting the MLLM training process, rescaling the autoregressive loss function, and improving data sampling methods. Through experiments on various MLLMs, the authors demonstrate a significant reduction in social biases while maintaining original performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper tries to fix a problem with big language models that learn from the internet but often make unfair judgments based on things like race or gender. The solution involves creating a new training dataset and a special way of adjusting the model’s learning process. This helps reduce bias in the model without making it perform worse. |
Keywords
» Artificial intelligence » Autoregressive » Loss function » Multi modal
