Summary of Multi-objective Large Language Model Unlearning, by Zibin Pan et al.
Multi-Objective Large Language Model Unlearning
by Zibin Pan, Shuwen Zhang, Yuesheng Zheng, Chi Li, Yuheng Cheng, Junhua Zhao
First submitted to arxiv on: 29 Dec 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 machine unlearning in large language models (LLMs) to eliminate undesirable behaviors. The authors propose the Multi-Objective Large Language Model Unlearning (MOLLM) algorithm, which addresses two major challenges: gradient explosion and catastrophic forgetting. By formulating LLM unlearning as a multi-objective optimization problem, MOLLM modifies the cross-entropy loss to overcome gradient explosion. A common descent update direction is calculated, allowing the model to forget target data while preserving its utility. The authors demonstrate that MOLLM outperforms state-of-the-art (SOTA) GA-based LLM unlearning methods in terms of unlearning effect and model utility preservation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper talks about a new way to fix large language models so they don’t learn things we don’t want them to. This is important because these models can be very useful, but they can also learn bad habits or biases from the data they’re trained on. The authors propose an algorithm called MOLLM that helps these models forget what they’ve learned and start fresh. They show that this algorithm works better than other methods at removing unwanted knowledge while still keeping the model’s overall usefulness. |
Keywords
» Artificial intelligence » Cross entropy » Large language model » Optimization
