Summary of Unlearn Efficient Removal Of Knowledge in Large Language Models, by Tyler Lizzo and Larry Heck
UNLEARN Efficient Removal of Knowledge in Large Language Models
by Tyler Lizzo, Larry Heck
First submitted to arxiv on: 8 Aug 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 novel UNLEARN method dynamically forgets specific knowledge in large language models, allowing private or proprietary information to be removed without retraining the model. By leveraging subspace methods, UNLEARN identifies and targets the removal of unwanted knowledge, achieving 96% targeted forgetting while maintaining performance on other knowledge within 2.5% of the original model. This outperforms previous state-of-the-art discriminatory abilities. The proposed LEARN method also efficiently adds targeted knowledge, matching fine-tuning accuracy with Low-Rank Adaptation (LoRA) without affecting similar tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can forget specific knowledge, like private information, without retraining. A new way to do this is called UNLEARN. It uses a special kind of math to find and remove the unwanted knowledge. This helps keep the model’s overall performance good while removing the unwanted parts. In tests, UNLEARN removed 96% of targeted knowledge without harming other parts of the model. Another method, LEARN, adds new knowledge in a way that is similar to how humans learn. |
Keywords
» Artificial intelligence » Fine tuning » Lora » Low rank adaptation
