Summary of Open-vocabulary Calibration For Fine-tuned Clip, by Shuoyuan Wang et al.
Open-Vocabulary Calibration for Fine-tuned CLIP
by Shuoyuan Wang, Jindong Wang, Guoqing Wang, Bob Zhang, Kaiyang Zhou, Hongxin Wei
First submitted to arxiv on: 7 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 investigates the confidence calibration problem in fine-tuned vision-language models (VLMs) and presents a simple yet effective approach called Distance-Aware Calibration (DAC). The authors demonstrate that existing calibration methods are insufficient to address this issue, especially in open-vocabulary settings. They propose DAC as an alternative solution, which uses distance-based temperature scaling for reliable model deployment. The results show high efficacy without compromising inference speed across 11 diverse datasets and 7 prompt learning methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at a problem that makes it hard to trust the answers from certain artificial intelligence models. These models are great at doing things like recognizing images and generating text, but they can be wrong sometimes. The authors of this paper found out that there’s an easy way to make these models more reliable by changing how they’re trained. They tested their method on lots of different datasets and showed it works well without slowing down the model too much. |
Keywords
* Artificial intelligence * Inference * Prompt * Temperature
