Summary of Catastrophic Failure Of Llm Unlearning Via Quantization, by Zhiwei Zhang et al.
Catastrophic Failure of LLM Unlearning via Quantization
by Zhiwei Zhang, Fali Wang, Xiaomin Li, Zongyu Wu, Xianfeng Tang, Hui Liu, Qi He, Wenpeng Yin, Suhang Wang
First submitted to arxiv on: 21 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- 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 This paper investigates the effectiveness of machine unlearning methods for large language models (LLMs) in removing unwanted behaviors and sensitive content acquired during training. Existing unlearning methods are found to be flawed, as they do not truly forget information but rather hide it. The study reveals that applying quantization to unlearned models can restore “forgotten” knowledge. Comprehensive experiments are conducted using various quantization techniques across multiple precision levels, showing that unlearned models retain an average of 21% of forgotten knowledge in full precision, increasing to 83% after 4-bit quantization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models (LLMs) have become very good at generating text by learning from a huge amount of text data. However, this training can also make them learn things they shouldn’t know, like private or copyrighted content. To solve this problem, researchers developed “machine unlearning”, which tries to remove this unwanted knowledge without retraining the model. But current methods don’t really work as expected – they just hide the information instead of truly forgetting it. This study explores why that is and finds a way to make machine unlearning actually forget what it’s supposed to. |
Keywords
» Artificial intelligence » Precision » Quantization
