Summary of Lead: Learning Decomposition For Source-free Universal Domain Adaptation, by Sanqing Qu et al.
LEAD: Learning Decomposition for Source-free Universal Domain Adaptation
by Sanqing Qu, Tianpei Zou, Lianghua He, Florian Röhrbein, Alois Knoll, Guang Chen, Changjun Jiang
First submitted to arxiv on: 6 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); 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 a novel approach called LEArning Decomposition (LEAD) for universal domain adaptation (UniDA) in the presence of covariate and label shifts. Without access to source data, SF-UniDA aims to achieve UniDA by determining whether covariate-shifted samples belong to target-private unknown categories. Existing methods rely on hand-crafted thresholding or iterative clustering strategies, which can be time-consuming and impractical. LEAD decouples features into known and unknown components using orthogonal decomposition analysis, then builds instance-level decision boundaries to identify target-private data. The paper demonstrates the effectiveness of LEAD through extensive experiments across various UniDA scenarios, including the OPDA scenario on VisDA dataset, where it outperforms GLC by 3.5% overall H-score while reducing processing time by 75%. LEAD is also complementary to existing methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to transfer knowledge from one place to another when there are changes in what kind of data we’re working with. They call this “universal domain adaptation”. The challenge is figuring out if the new data belongs to categories we don’t know yet. Currently, people use complicated methods that take a long time or make decisions based on rules they set beforehand. This paper proposes a new approach called LEAD (Learning Decomposition) that breaks down features into things we do and don’t know about. It then uses this information to decide which data belongs to unknown categories. The results show that LEAD works well in different situations, including one where it beats another method by 3.5% while being much faster. |
Keywords
* Artificial intelligence * Clustering * Domain adaptation
