Summary of Exploring Learning Complexity For Efficient Downstream Dataset Pruning, by Wenyu Jiang et al.
Exploring Learning Complexity for Efficient Downstream Dataset Pruning
by Wenyu Jiang, Zhenlong Liu, Zejian Xie, Songxin Zhang, Bingyi Jing, Hongxin Wei
First submitted to arxiv on: 8 Feb 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 This paper tackles the challenge of reducing the size of datasets used to fine-tune large-scale pre-trained models without sacrificing task performance. Existing methods require training on the entire dataset, which is impractical for large models. The authors propose a novel, training-free hardness score called Distorting-based Learning Complexity (DLC) that efficiently identifies informative images and instructions from downstream datasets. They define Learning Complexity to quantify sample hardness and use a lightweight weights masking process for fast estimation. The proposed approach, dubbed FlexRand, is designed to alleviate the subset distribution shift problem and achieves state-of-the-art performance with reduced pruning time by 35x in image pruning benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps make big pre-trained models smaller without losing their abilities. To do this, they create a new way to figure out which parts of the data are most important for training. This method doesn’t require training on all the data, making it much faster and more efficient. The authors test their approach on several datasets and show that it works well, reducing the time it takes to train by 35 times. |
Keywords
* Artificial intelligence * Pruning
