Summary of Adapting to Distribution Shift by Visual Domain Prompt Generation, By Zhixiang Chi et al.
Adapting to Distribution Shift by Visual Domain Prompt Generation
by Zhixiang Chi, Li Gu, Tao Zhong, Huan Liu, Yuanhao Yu, Konstantinos N Plataniotis, Yang Wang
First submitted to arxiv on: 5 May 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 proposes an approach to adapt a model at test-time using a few unlabeled data points, addressing distribution shifts. It leverages correlated information from pre-trained backbones and source domains to extract domain knowledge from limited data. The method utilizes recent foundation models with strong out-of-distribution generalization, unlike previous studies. A key innovation is the integration of modelling source domains and learning to adapt into a single training stage. The approach consists of building a knowledge bank, generating a domain prompt based on target data, and directing visual features using a guidance module. Additionally, a domain-aware contrastive loss is proposed, along with meta-learning for domain knowledge extraction. Experimental results demonstrate the effectiveness of this method, outperforming previous work on 5 large-scale benchmarks including WILDS and DomainNet. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps machines learn to adapt to new situations by using a small amount of data from that situation. It’s like learning a new language or understanding cultural differences. The researchers use powerful pre-trained models and combine them with information from different domains (like animals, vehicles, or buildings) to create a “knowledge bank”. When the model is shown some target data from a new domain, it generates a special prompt to help it understand that domain better. This approach outperforms previous methods on several large-scale benchmarks. |
Keywords
» Artificial intelligence » Contrastive loss » Generalization » Meta learning » Prompt
