Summary of Overcoming Growth-induced Forgetting in Task-agnostic Continual Learning, by Yuqing Zhao et al.
Overcoming Growth-Induced Forgetting in Task-Agnostic Continual Learning
by Yuqing Zhao, Divya Saxena, Jiannong Cao, Xiaoyun Liu, Changlin Song
First submitted to arxiv on: 20 Aug 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 presents a novel approach to continual learning (CL) that addresses growth-induced forgetting (GIFt), a phenomenon where improper model growth leads to severe degradation of previously learned knowledge. The authors identify layer expansion as the primary cause of GIFt and propose a SparseGrow method that controls efficient parameter usage during growth, reducing GIFt while enhancing adaptability over new data. The approach combines sparse growth with on-data initialization at training late-stage to create partially 0-valued expansions that fit learned distributions. The paper demonstrates the necessity of layer expansion and showcases the effectiveness of SparseGrow in overcoming GIFt through experiments across various datasets and settings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Continual learning helps machines learn new tasks over time. However, this process can forget previously learned information if not done correctly. Researchers have found that expanding layers in a model to help it grow is actually making things worse. They propose a new way to control this growth called SparseGrow, which uses data-driven methods to decide how many parameters to add and when. This approach helps machines learn new tasks while keeping the knowledge they’ve gained so far. |
Keywords
* Artificial intelligence * Continual learning
