Revisiting the Robust Generalization of Adversarial Prompt Tuning
This work addresses adversarial robustness for vision-language models, which is crucial for reliable zero-shot generalization in downstream tasks, but it is incremental as it builds on existing prompt learning strategies.
The paper tackled the problem of over-fitting in adversarial prompt tuning for vision-language models like CLIP, proposing the CAPT framework which improved robust generalization on adversarial examples while maintaining accuracy on clean ones, as shown through experiments across 14 datasets and 4 data sparsity schemes.
Understanding the vulnerability of large-scale pre-trained vision-language models like CLIP against adversarial attacks is key to ensuring zero-shot generalization capacity on various downstream tasks. State-of-the-art defense mechanisms generally adopt prompt learning strategies for adversarial fine-tuning to improve the adversarial robustness of the pre-trained model while keeping the efficiency of adapting to downstream tasks. Such a setup leads to the problem of over-fitting which impedes further improvement of the model's generalization capacity on both clean and adversarial examples. In this work, we propose an adaptive Consistency-guided Adversarial Prompt Tuning (i.e., CAPT) framework that utilizes multi-modal prompt learning to enhance the alignment of image and text features for adversarial examples and leverage the strong generalization of pre-trained CLIP to guide the model-enhancing its robust generalization on adversarial examples while maintaining its accuracy on clean ones. We also design a novel adaptive consistency objective function to balance the consistency of adversarial inputs and clean inputs between the fine-tuning model and the pre-trained model. We conduct extensive experiments across 14 datasets and 4 data sparsity schemes (from 1-shot to full training data settings) to show the superiority of CAPT over other state-of-the-art adaption methods. CAPT demonstrated excellent performance in terms of the in-distribution performance and the generalization under input distribution shift and across datasets.