Summary of Private Fine-tuning Of Large Language Models with Zeroth-order Optimization, by Xinyu Tang et al.
Private Fine-tuning of Large Language Models with Zeroth-order Optimization
by Xinyu Tang, Ashwinee Panda, Milad Nasr, Saeed Mahloujifar, Prateek Mittal
First submitted to arxiv on: 9 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL); 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 This paper presents a new approach to training large language models while preserving user privacy. The authors introduce DP-ZO, a private fine-tuning framework that builds upon zeroth-order optimization methods. By privatizing only the scalar step size in these methods, DP-ZO achieves strong memory efficiency and a good trade-off between privacy and utility across different tasks. Compared to existing approaches like DP-SGD, DP-ZO offers significant advantages in terms of memory usage and obtains higher utility when using the Laplace mechanism. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure that large language models are trained in a way that keeps users’ information private. The authors came up with a new method called DP-ZO to do this. It’s like a shortcut that only privatizes the small part of the training process, which makes it more efficient and effective. This approach helps protect user privacy while still allowing the model to learn and improve. |
Keywords
* Artificial intelligence * Fine tuning * Optimization
