Summary of Label Privacy in Split Learning For Large Models with Parameter-efficient Training, by Philip Zmushko et al.
Label Privacy in Split Learning for Large Models with Parameter-Efficient Training
by Philip Zmushko, Marat Mansurov, Ruslan Svirschevski, Denis Kuznedelev, Max Ryabinin, Aleksandr Beznosikov
First submitted to arxiv on: 21 Dec 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 As deep learning models become larger and more expensive, many practitioners turn to fine-tuning APIs, which allow for parameter-efficient fine-tuning (PEFT) between clients providing data and servers hosting the model. However, this raises concerns about privacy breaches during training. This study systematically searches for ways to fine-tune models over an API while keeping labels private, analyzing LoRA’s privacy properties and proposing P^3EFT, a multi-party split learning algorithm that maintains privacy at a lower performance overhead. We validate our algorithm by fine-tuning DeBERTa-v2-XXLarge, Flan-T5 Large, and LLaMA-2 7B on various NLP tasks, finding that P^3EFT is competitive with existing privacy-preserving methods in multi-party and two-party setups while achieving higher accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper studies how to fine-tune big models over the internet without sharing private information. Right now, many people use web services to fine-tune their models between a client providing data and a server hosting the model. But this raises concerns about keeping that data private during training. The researchers looked for ways to do this while still getting good results. They analyzed a popular approach called LoRA and came up with a new method called P^3EFT, which keeps privacy at a lower cost of accuracy. To test their idea, they tried fine-tuning three big models on different natural language processing tasks and found that their method worked well compared to existing methods. |
Keywords
» Artificial intelligence » Deep learning » Fine tuning » Llama » Lora » Natural language processing » Nlp » Parameter efficient » T5
