Summary of Groot: Adversarial Testing For Generative Text-to-image Models with Tree-based Semantic Transformation, by Yi Liu et al.
Groot: Adversarial Testing for Generative Text-to-Image Models with Tree-based Semantic Transformation
by Yi Liu, Guowei Yang, Gelei Deng, Feiyue Chen, Yuqi Chen, Ling Shi, Tianwei Zhang, Yang Liu
First submitted to arxiv on: 19 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Software Engineering (cs.SE)
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 Groot, an automated framework for testing the safety of text-to-image generative models. The existing adversarial testing techniques have limitations, including low success rates and inefficiency. To address these challenges, Groot uses tree-based semantic transformation to systematically refine adversarial prompts. This approach employs semantic decomposition and sensitive element drowning strategies in conjunction with Large Language Models (LLMs). In a comprehensive evaluation, Groot outperforms current state-of-the-art approaches, achieving a remarkable success rate of 93.66% on models like DALL-E 3 and Midjourney. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Groot is a new way to test if text-to-image models can make naughty pictures. These models are very good at making images from words, but they sometimes create things that shouldn’t be shown to everyone. To help keep people safe online, researchers need better ways to test these models and see what they can do. Groot uses special tricks to ask the models questions in a way that makes them show us what’s inside. It works really well, even better than the best methods we have now. |
