Summary of Parameter-efficient Fine-tuning For Continual Learning: a Neural Tangent Kernel Perspective, by Jingren Liu et al.
Parameter-Efficient Fine-Tuning for Continual Learning: A Neural Tangent Kernel Perspective
by Jingren Liu, Zhong Ji, YunLong Yu, Jiale Cao, Yanwei Pang, Jungong Han, Xuelong Li
First submitted to arxiv on: 24 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 parameter-efficient fine-tuning for continual learning (PEFT-CL) by analyzing its dynamics using Neural Tangent Kernel (NTK) theory. It identifies three key factors influencing PEFT-CL performance: training sample size, task-level feature orthogonality, and regularization. The authors introduce NTK-CL, a framework that eliminates task-specific parameter storage while generating task-relevant features. This approach achieves state-of-the-art performance on established PEFT-CL benchmarks by fine-tuning optimizable parameters with appropriate regularization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper studies how to improve machine learning models so they can learn new things without forgetting what they already know. It uses a special tool called Neural Tangent Kernel (NTK) to understand why some methods work better than others. The authors find three important factors that affect how well these methods do: the size of the training data, how different each task is from the others, and how much regularization is used. They then create a new method called NTK-CL that does better than other methods by generating more features for each sample and using constraints to make sure the model doesn’t forget what it learned. |
Keywords
* Artificial intelligence * Continual learning * Fine tuning * Machine learning * Parameter efficient * Regularization
