Prompt Reverse Learning: Enhancing Visual Language Models for Rare Image Recognition

Shuhan Zhu, Yonggang Zhang, Xinmei Tian, Xiaoyan Sun

Keywords: Prompt Learning

Abstract: Large visual language models like CLIP have demonstrated impressive performance on various downstream tasks involving common data, e.g., natural images, by leveraging prompt learning. However, these models often falter when applied to tasks involving rare data, e.g., medical images. We provide an experimental insight into this phenomenon: CLIP is insensitive to the class names of rare images. For instance, replacing the class name “medulloblastoma” (a type of brain tumor) with “dog” in prompts has minimal impact on performance, a phenomenon not observed with common images. This highlights the disparity in representation learning between common and rare data. To realign prompt learning with rare image recognition, we propose a novel prompt learning strategy, termed prompt reverse learning (PeLen). Different from the existing methods that adapt CLIP's representations to downstream tasks, PeLen adapts task-specific representations to CLIP's representations. Built upon the insensitivity to the class names of rare images, PeLen designates common images and their class names to represent a specific class of rare images and class names, e.g., allowing the image and text of a dog to correspond to the image and text of medulloblastoma. Consequently, PeLen learns prompts to align the representations between the rare images and the visual and textual representations of common images. Our experiments on three types of rare images demonstrate the efficacy of PeLen for rare image recognition.

Primary Area: transfer learning, meta learning, and lifelong learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3052