Summary of Free: Faster and Better Data-free Meta-learning, by Yongxian Wei et al.
FREE: Faster and Better Data-Free Meta-Learning
by Yongxian Wei, Zixuan Hu, Zhenyi Wang, Li Shen, Chun Yuan, Dacheng Tao
First submitted to arxiv on: 2 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 proposes Faster and Better Data-Free Meta-Learning (FREE), a framework that extracts knowledge from pre-trained models without requiring original data. FREE addresses slow recovery speed and overlooks gaps in heterogeneous models by introducing a meta-generator for rapid task recovery and a meta-learner for generalizing to new tasks. The Faster Inversion via Meta-Generator module accelerates data recovery, while the Better Generalization via Meta-Learner module optimizes the meta-learner using an implicit gradient alignment algorithm. Experiments on multiple benchmarks demonstrate a 20x speed-up and performance enhancement of 1.42% to 4.78% compared to state-of-the-art methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps computers learn new skills without needing all the old data. Current methods are slow and don’t work well with different types of old data. The authors propose a faster and better way to do this, called FREE. FREE has two parts: one that quickly learns new tasks from old data and another that improves how it generalizes to new tasks. Tests show that FREE is 20 times faster and does 1-5% better than other methods. |
Keywords
» Artificial intelligence » Alignment » Generalization » Meta learning
