Summary of Revisiting the Robust Generalization Of Adversarial Prompt Tuning, by Fan Yang et al.
Revisiting the Robust Generalization of Adversarial Prompt Tuning
by Fan Yang, Mingxuan Xia, Sangzhou Xia, Chicheng Ma, Hui Hui
First submitted to arxiv on: 18 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 a new framework for enhancing the robustness of pre-trained vision-language models like CLIP against adversarial attacks, which is crucial for ensuring zero-shot generalization capacity on various downstream tasks. The current state-of-the-art defense mechanisms rely on prompt learning strategies, but this approach can lead to over-fitting and impede further improvement in model performance. To address this issue, the authors introduce an adaptive Consistency-guided Adversarial Prompt Tuning (CAPT) framework that utilizes multi-modal prompt learning to align image and text features for adversarial examples and leverage the strong generalization of pre-trained CLIP. The CAPT framework also includes a novel adaptive consistency objective function to balance the consistency between fine-tuned and pre-trained models. Experimental results demonstrate the superiority of CAPT over other state-of-the-art adaptation methods, showing excellent performance in terms of in-distribution performance and generalization under input distribution shift and across datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper talks about how big AI models can be tricked into making mistakes. To stop this from happening, researchers developed ways to fine-tune these models for specific tasks. However, this process can make the models too specialized and not good at generalizing to new situations. The authors of this paper want to fix this problem by creating a new way to fine-tune models that keeps them robust against mistakes while still making them good at specific tasks. They call this method Consistency-guided Adversarial Prompt Tuning (CAPT). This method uses multiple ways to align the images and text inputs to make the model more accurate. The authors tested their method on 14 different datasets and showed that it works better than other methods. |
Keywords
» Artificial intelligence » Generalization » Multi modal » Objective function » Prompt » Zero shot
