Summary of Why Fine-grained Labels in Pretraining Benefit Generalization?, by Guan Zhe Hong et al.
Why Fine-grained Labels in Pretraining Benefit Generalization?
by Guan Zhe Hong, Yin Cui, Ariel Fuxman, Stanley Chan, Enming Luo
First submitted to arxiv on: 30 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
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 Recent studies demonstrate that pretraining deep neural networks with fine-grained labeled data and subsequent fine-tuning on coarse-labeled data for downstream tasks typically leads to better generalization. Despite empirical evidence supporting this, the theoretical justification remains an open problem. This paper addresses the gap by introducing a “hierarchical multi-view” structure confining input data distribution. The authors prove that coarse-grained pretraining only allows neural networks to learn common features well, while fine-grained pretraining enables learning rare features in addition to common ones, resulting in improved accuracy on challenging downstream test samples. Furthermore, they employ the hierarchical multi-view framework to derive theoretical bounds for generalization performance. By doing so, this work provides a solid theoretical foundation for understanding the benefits of fine-grained pretraining. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This study explores why some deep neural networks perform better when trained on small amounts of labeled data and then adjusted for specific tasks. Researchers found that using too little information can actually make the network worse at recognizing patterns in new data. To explain this, they developed a special framework called “hierarchical multi-view” that helps the network understand different types of input. By showing how this framework works, the authors prove that fine-tuning networks with more detailed information can improve their ability to recognize rare or unusual patterns. This discovery provides a deeper understanding of why some neural networks perform better than others. |
Keywords
» Artificial intelligence » Fine tuning » Generalization » Pretraining
