Summary of Safety Alignment For Vision Language Models, by Zhendong Liu et al.
Safety Alignment for Vision Language Models
by Zhendong Liu, Yuanbi Nie, Yingshui Tan, Xiangyu Yue, Qiushi Cui, Chongjun Wang, Xiaoyong Zhu, Bo Zheng
First submitted to arxiv on: 22 May 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 The paper proposes an enhanced Vision Language Model (VLM) that addresses the vulnerability of its visual modality to attacks by incorporating safety modules. The authors build upon existing models, such as LLaVA-v1.5, and achieve a safety score of 8.26 on the Red Teaming Visual Language Models (RTVLM) benchmark, surpassing GPT-4V. The method utilizes a two-stage training process, including a safety projector, safety tokens, and a safety head, to improve defense against risky images. This approach offers ease of use, high flexibility, and strong controllability while maintaining minimal impact on the model’s general performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper is about making language models safer by adding special features that can spot when an image might be trying to trick it. Right now, these models are good at understanding what they see, but bad at telling if something is off. The authors create a new kind of model that can do both things well – understand images and detect potential tricks. They test this model on some benchmark tests and find that it does better than other similar models in keeping itself safe. |
Keywords
» Artificial intelligence » Gpt » Language model
