Summary of Analyzing and Reducing Catastrophic Forgetting in Parameter Efficient Tuning, by Weijieying Ren et al.
Analyzing and Reducing Catastrophic Forgetting in Parameter Efficient Tuning
by Weijieying Ren, Xinlong Li, Lei Wang, Tianxiang Zhao, Wei Qin
First submitted to arxiv on: 29 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 investigates the catastrophic forgetting problem in large language models (LLMs) when they are fine-tuned on complex and diverse downstream tasks. Existing strategies like memory replay, regularization, and parameter isolation have been explored to balance learning plasticity and memory stability. This work uncovers a geometric connection between different minima in LLMs continual learning scenarios through mode connectivity, which connects different minima by low-loss valleys. The authors propose an Interpolation-based LoRA (I-LoRA) method that constructs a dual-memory experience replay framework based on LoRA parameter interpolations. I-LoRA demonstrates significant performance gains of up to 11% over previous state-of-the-art approaches on eight domain-specific continual learning benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are very good at understanding and generating human languages. However, when they learn new things, they often forget the old things they knew. This is a problem because we want them to keep learning without forgetting what they already know. The researchers in this paper looked for ways to solve this problem by studying how different parts of the model are connected. They found that these connections can be used to help the model remember old things while still learning new ones. They also developed a new method called I-LoRA, which helps the model learn and remember better. |
Keywords
* Artificial intelligence * Continual learning * Lora * Regularization
