Summary of Domain-inspired Sharpness-aware Minimization Under Domain Shifts, by Ruipeng Zhang et al.
Domain-Inspired Sharpness-Aware Minimization Under Domain Shifts
by Ruipeng Zhang, Ziqing Fan, Jiangchao Yao, Ya Zhang, Yanfeng Wang
First submitted to arxiv on: 29 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: 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 Medium Difficulty summary: This paper presents a novel optimization algorithm called Domain-Inspired Sharpness-Aware Minimization (DISAM) designed to address the issue of inconsistent convergence degree across different domains in optimization under domain shifts. The algorithm introduces a constraint that minimizes variance in the domain loss, allowing for elastic gradient calibration and automatic weakening of perturbations towards less optimized domains. DISAM is theoretically shown to achieve faster overall convergence and improved generalization when inconsistent convergence emerges. Experimental results on various domain generalization benchmarks demonstrate the superiority of DISAM over state-of-the-art methods. Additionally, the algorithm is shown to be efficient in parameter-efficient fine-tuning combined with pretraining models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: This paper develops a new way to make computer algorithms more consistent when they’re learning from different sources of data. The problem is that current algorithms can get stuck on certain types of data and forget about others. To solve this, the authors created an algorithm called DISAM that adjusts how much it “perturbs” (or changes) its calculations based on which type of data it’s working with. This helps it learn more quickly and accurately from a variety of sources. The authors tested their algorithm on many different types of data and found that it worked better than other algorithms in many cases. |
Keywords
» Artificial intelligence » Domain generalization » Fine tuning » Generalization » Optimization » Parameter efficient » Pretraining
