Summary of Manipulating and Mitigating Generative Model Biases Without Retraining, by Jordan Vice et al.
Manipulating and Mitigating Generative Model Biases without Retraining
by Jordan Vice, Naveed Akhtar, Richard Hartley, Ajmal Mian
First submitted to arxiv on: 3 Apr 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 proposed dynamic bias manipulation technique for text-to-image generative models leverages their rich language embedding spaces without requiring retraining. This approach allows for convenient control over model outputs and class distributions by exploiting foundational vector algebra. The method is demonstrated by balancing the frequency of social classes in generated images, effectively controlling three social bias dimensions. However, this technique can also be framed as a backdoor attack with severity control using semantically-null input triggers, achieving up to 100% attack success rate. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper introduces a new way to control the biases of text-to-image generative models without needing to retrain them. The method works by manipulating the language embeddings used by the model, which allows for precise control over the images generated. This can be useful for balancing out social biases in generated images. However, it also raises concerns about potential misuse, as this technique could be used to create fake images that are difficult to detect. |
Keywords
» Artificial intelligence » Embedding
