Summary of Eta: Evaluating Then Aligning Safety Of Vision Language Models at Inference Time, by Yi Ding et al.
ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time
by Yi Ding, Bolian Li, Ruqi Zhang
First submitted to arxiv on: 9 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Computation and Language (cs.CL); 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 This paper proposes a novel approach to improving the safety of Vision Language Models (VLMs) in multimodal intelligence applications. Despite their effectiveness, VLMs are limited by significant safety challenges that can be exploited by adversarial visual inputs. Existing defense methods often require substantial resources or fail to balance safety and usefulness in responses. The proposed Evaluating Then Aligning (ETA) framework addresses these limitations by first evaluating input visual contents and output responses to establish a robust safety awareness, and then aligning unsafe behaviors at both shallow and deep levels. ETA is shown to outperform baseline methods in terms of harmlessness, helpfulness, and efficiency, reducing the unsafe rate by 87.5% in cross-modality attacks and achieving 96.6% win-ties in GPT-4 helpfulness evaluation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to make vision language models safer. These models are very good at understanding text and images, but they can be tricked into doing bad things if given the right image input. The researchers created a new system that first checks what’s in an image and then makes sure the model doesn’t do anything harmful with it. This system works better than other ways to keep models safe and helps them give helpful answers more often. |
Keywords
» Artificial intelligence » Gpt
