Summary of Rethinking Weight Decay For Robust Fine-tuning Of Foundation Models, by Junjiao Tian et al.
Rethinking Weight Decay for Robust Fine-Tuning of Foundation Models
by Junjiao Tian, Chengyue Huang, Zsolt Kira
First submitted to arxiv on: 3 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL); 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 Medium Difficulty Summary: Modern optimizers like AdamW, equipped with momentum and adaptive learning rates, are designed to escape local minima and explore the vast parameter space. This paper proposes a new weight decay technique called Selective Projection Decay (SPD) that selectively imposes strong penalties on certain layers while allowing others to change freely. SPD expands and contracts the parameter search space for layers with consistent and inconsistent loss reduction, respectively. The authors experimentally demonstrate that Adam equipped with SPD provides better in-distribution generalization and out-of-distribution robustness performance on multiple popular vision and language benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty Summary: This paper talks about how to make machine learning models work better. Right now, some optimizers (like AdamW) are good at finding new solutions, but they can also be too flexible and get stuck in bad places. The authors propose a new way to control this flexibility called Selective Projection Decay (SPD). SPD helps the model learn more efficiently by adjusting how much it changes its parameters. In experiments, using SPD with AdamW makes the models perform better on certain tasks. |
Keywords
» Artificial intelligence » Generalization » Machine learning
