Summary of Label-augmented Dataset Distillation, by Seoungyoon Kang et al.
Label-Augmented Dataset Distillation
by Seoungyoon Kang, Youngsun Lim, Hyunjung Shim
First submitted to arxiv on: 24 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 A novel framework for dataset distillation, Label-Augmented Dataset Distillation (LADD), enhances image representation by incorporating dense labels with rich semantics. By sub-sampling synthetic images and generating additional labels, LADD achieves significant performance benefits with only a 2.5% increase in storage requirements. The label generation strategy can complement existing methods, enhancing training efficiency and accuracy. Experimental results demonstrate that LADD outperforms existing approaches in terms of computational overhead and accuracy, achieving an average gain of 14.9%. The method also improves cross-architecture robustness of the distilled dataset, which is crucial for real-world applications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Dataset distillation gets a boost with Label-Augmented Dataset Distillation (LADD), a new framework that adds dense labels to synthetic images. This makes image representation even better and helps training be more efficient. The extra labels only take up a little extra space – just 2.5% for ImageNet subsets! LADD works well across different datasets, distillation settings, and algorithms. It’s also good at keeping the distilled dataset working well with different models, which is important in real-world uses. |
Keywords
» Artificial intelligence » Distillation » Semantics
