Summary of Probabilistic Language-image Pre-training, by Sanghyuk Chun and Wonjae Kim and Song Park and Sangdoo Yun
Probabilistic Language-Image Pre-Training
by Sanghyuk Chun, Wonjae Kim, Song Park, Sangdoo Yun
First submitted to arxiv on: 24 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
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 This paper introduces Probabilistic Language-Image Pre-training (ProLIP), a novel vision-language model that embeds aligned image-text pairs into a joint space using probabilistic objectives. ProLIP achieves strong zero-shot capability, achieving 74.6% ImageNet accuracy with ViT-B/16. The model efficiently estimates uncertainty using an “uncertainty token” without extra parameters and introduces a novel inclusion loss to enforce distributional relationships between image-text pairs. Experimental results demonstrate that ProLIP benefits downstream tasks by leveraging uncertainty estimates, aligning with intuitive notions of uncertainty. For example, shorter texts are more uncertain and more general inputs include specific ones. The code is available at https://github.com/naver-ai/prolip. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper makes language and image models work better together! Right now, these models assume that each picture has one special description, but in real life, many pictures have lots of descriptions. To fix this problem, the researchers created a new model called ProLIP (Probabilistic Language-Image Pre-training). ProLIP is special because it uses math to figure out when things might not be exactly right, like if an image has many different descriptions or vice versa. This helps make the model more useful for everyday tasks. |
Keywords
» Artificial intelligence » Language model » Token » Vit » Zero shot
