Summary of Fine-tuning Is Fine, If Calibrated, by Zheda Mai et al.
Fine-Tuning is Fine, if Calibrated
by Zheda Mai, Arpita Chowdhury, Ping Zhang, Cheng-Hao Tu, Hong-You Chen, Vardaan Pahuja, Tanya Berger-Wolf, Song Gao, Charles Stewart, Yu Su, Wei-Lun Chao
First submitted to arxiv on: 24 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
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 explores the limitations of fine-tuning pre-trained models for downstream applications, specifically examining what is lost or gained during this process. The authors identify that the primary issue lies not in forgetting relationships among classes or degrading features, but rather in the discrepancy in logit scales between fine-tuning classes and other classes. They propose a simple post-processing calibration to restore the pre-trained model’s capabilities while revealing feature improvements across all classes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how we can make computers better at recognizing things they’ve learned before. Sometimes, we teach a computer to recognize lots of different things, but then we want it to focus on just a few specific things. This process is called fine-tuning. The problem is that when we do this, the computer often forgets what it knew about other things. The authors of this paper looked into why this happens and found that the issue isn’t that the computer forgets or loses its ability to recognize certain things, but rather that it has trouble comparing how likely something is to belong to one category versus another. They suggest a simple way to fix this problem, which could help us make computers even better at recognizing things. |
Keywords
* Artificial intelligence * Fine tuning
