Summary of Downstream-pretext Domain Knowledge Traceback For Active Learning, by Beichen Zhang et al.
Downstream-Pretext Domain Knowledge Traceback for Active Learning
by Beichen Zhang, Liang Li, Zheng-Jun Zha, Jiebo Luo, Qingming Huang
First submitted to arxiv on: 20 Jul 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 proposes a novel active learning (AL) method, called downstream-pretext domain knowledge traceback (DOKT), which leverages pre-training for robust feature learning in AL. Unlike previous methods that directly use pre-trained representations in AL, DOKT traces the interactions between low-level pretext tasks and high-level annotated data to select diverse and instructive samples near the decision boundary. The method consists of a diversity indicator, which constructs two feature spaces to locate neighbors of unlabeled data from the downstream space in the pretext space, and an uncertainty estimator that uses domain mixing to measure the divergence of perturbed samples. Experiments on ten datasets show that DOKT outperforms state-of-the-art methods and generalizes well to various application scenarios such as semantic segmentation and image captioning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to choose which data points to label when you’re training an artificial intelligence model. They use a special technique called “active learning” that helps the model learn from the right examples. The method, called DOKT, looks at how different pieces of information relate to each other to pick the most useful samples for labeling. This approach is better than previous methods and works well in different situations like image recognition and natural language processing. |
Keywords
» Artificial intelligence » Active learning » Image captioning » Natural language processing » Semantic segmentation
