Summary of Mitigating Memorization in Language Models, by Mansi Sakarvadia et al.
Mitigating Memorization In Language Models
by Mansi Sakarvadia, Aswathy Ajith, Arham Khan, Nathaniel Hudson, Caleb Geniesse, Kyle Chard, Yaoqing Yang, Ian Foster, Michael W. Mahoney
First submitted to arxiv on: 3 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 investigates methods to mitigate language model memorization, which can be problematic when training data are private or sensitive. The authors introduce a suite of small, computationally-efficient LMs called TinyMem for rapid development and evaluation of mitigation methods. They compare regularizer-based, fine-tuning-based, and machine unlearning-based methods, finding that the latter are more effective at curbing memorization while preserving performance on target tasks. Specifically, they propose an unlearning method called BalancedSubnet that outperforms other methods in removing memorized information. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at ways to stop language models from remembering training data. This can be a problem when the data is private or sensitive. The authors create some small and fast language models called TinyMem to test these methods. They look at three kinds of methods: ones that add regularizers, ones that fine-tune the model, and ones that use machine unlearning. They find that the unlearning methods are the most effective in stopping memorization while keeping the model’s performance good. |
Keywords
» Artificial intelligence » Fine tuning » Language model
