Summary of Meow: Memory Supervised Llm Unlearning Via Inverted Facts, by Tianle Gu et al.
MEOW: MEMOry Supervised LLM Unlearning Via Inverted Facts
by Tianle Gu, Kexin Huang, Ruilin Luo, Yuanqi Yao, Yujiu Yang, Yan Teng, Yingchun Wang
First submitted to arxiv on: 18 Sep 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 The proposed Large Language Models (LLMs) unlearning method, LLM Unlearning, addresses concerns about memorizing sensitive information by removing this information from trained LLMs. The existing approaches face challenges regarding utility, efficiency, and robustness. To overcome these limitations, the authors introduce MEOW, a gradient descent-based unlearning method that uses an offline LLM to generate inverted facts, designates MEMO as a metric for quantifying memorization, and fine-tunes the model based on these signals. The evaluation on the ToFU benchmark with Llama2-7B-Chat and Phi-1.5B shows significant improvement in forget quality without substantial loss in model utility. MEOW also exhibits no significant degradation in NLU or NLG capabilities, with a slight improvement in NLU performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a method to remove sensitive information from Large Language Models (LLMs) called LLM Unlearning. This is important because LLMs can memorize private data. The current methods have problems making them useful. MEOW is a new way to unlearn this information without losing the model’s abilities. It uses an offline LLM, generates special facts, and adjusts the model based on these signals. The results show that MEOW works well and doesn’t hurt the model’s performance. |
Keywords
» Artificial intelligence » Gradient descent
