Summary of Undial: Self-distillation with Adjusted Logits For Robust Unlearning in Large Language Models, by Yijiang River Dong et al.
UNDIAL: Self-Distillation with Adjusted Logits for Robust Unlearning in Large Language Models
by Yijiang River Dong, Hongzhou Lin, Mikhail Belkin, Ramon Huerta, Ivan Vulić
First submitted to arxiv on: 15 Feb 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 introduces a novel method for mitigating the retention of sensitive information in large language models, called UnDIAL (Unlearning via Self-Distillation on Adjusted Logits). Existing methods often become unstable when fine-tuning models to remove unwanted information, leading to over-unlearning. The proposed approach leverages self-distillation to adjust logits and selectively reduce the influence of targeted tokens, ensuring smooth convergence and avoiding catastrophic forgetting. UnDIAL achieves robustness in unlearning and scalability while maintaining stable training dynamics and resilience to hyperparameter tuning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models need to be careful not to remember things they’re supposed to forget! This paper shows how to make sure these models don’t retain sensitive or private information. The problem is that existing ways of “unlearning” this information can actually make the model worse, not better. So, the authors came up with a new way called UnDIAL. It uses something called self-distillation to help the model forget what it’s supposed to forget without messing things up. This makes it more stable and reliable. |
Keywords
» Artificial intelligence » Distillation » Fine tuning » Hyperparameter » Logits
