Summary of On Large Language Model Continual Unlearning, by Chongyang Gao et al.
On Large Language Model Continual Unlearning
by Chongyang Gao, Lixu Wang, Kaize Ding, Chenkai Weng, Xiao Wang, Qi Zhu
First submitted to arxiv on: 14 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR)
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 While large language models (LLMs) have achieved impressive performance across various domains and tasks, their security issues have become increasingly severe. Machine unlearning has emerged as a representative approach for model safety and security by removing the influence of undesired data on the target model. However, existing LLM unlearning methods often ignore previous data access limitations due to privacy concerns and copyright protection. To overcome these challenges, we propose the OOO framework that includes an Orthogonal low-rank adapter (LoRA) for continually unlearning requested data and an Out-Of-Distribution (OOD) detector to measure the similarity between input and unlearning data. The orthogonal LoRA achieves parameter disentanglement among continual unlearning requests. Our proposed framework, OOO, consistently achieves the best unlearning effectiveness and utility preservation across three tasks and seven datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are getting better at understanding language, but they’re also becoming a security risk. To make them safer, scientists have developed a way to “forget” unwanted data. However, this method has some limitations when dealing with continuous requests to forget certain information. The new framework, called OOO, can continuously unlearn requested data without losing its ability to understand language. It uses two main components: an adapter that separates the data it needs to forget from the rest of the model’s knowledge and a detector that measures how similar the new data is to what it already knows. In tests, OOO performed better than existing methods in forgetting unwanted data while still keeping its language understanding abilities. |
Keywords
* Artificial intelligence * Language understanding * Lora
