Summary of Delt: a Simple Diversity-driven Earlylate Training For Dataset Distillation, by Zhiqiang Shen and Ammar Sherif and Zeyuan Yin and Shitong Shao
DELT: A Simple Diversity-driven EarlyLate Training for Dataset Distillation
by Zhiqiang Shen, Ammar Sherif, Zeyuan Yin, Shitong Shao
First submitted to arxiv on: 29 Nov 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 new approach to dataset distillation, specifically addressing the issue of diversity in synthetic images generated through batch-to-global matching. The conventional method uses bi-level optimization methods, whereas this novel technique, called Diversity-driven EarlyLate Training (DELT), partitions samples into smaller subtasks and employs local optimizations to enhance image diversity while reducing computation. This approach is shown to improve generalization performance on various datasets, including CIFAR, Tiny-ImageNet, ImageNet-1K, and their sub-datasets, with an average improvement of 2-5% in accuracy and a reduction in synthesis time by up to 39.3%. The code for this method is available on GitHub. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates new synthetic images that are more diverse and can help computers learn better. Right now, there’s a problem when we try to create lots of synthetic images at once – they all look very similar! This paper solves this by breaking down the task into smaller parts and optimizing each part separately. This makes the synthetic images much more varied and helps computers learn even better. The researchers tested their method on several different datasets and found that it was significantly better than other methods, especially when there are a lot of classes to choose from. |
Keywords
» Artificial intelligence » Distillation » Generalization » Optimization
