Mind the Privacy Unit! User-Level Differential Privacy for Language Model Fine-Tuning

arXiv:2406.14322v1 Announce Type: new
Abstract: Large language models (LLMs) have emerged as powerful tools for tackling complex tasks across diverse domains, but they also raise privacy concerns when fine-tuned on sensitive data due to potential memorization. While differential privacy (DP) offers a promising solution by ensuring models are `almost indistinguishable' with or without any particular privacy unit, current evaluations on LLMs mostly treat each example (text record) as the privacy unit. This leads to uneven user privacy guarantees when contributions …

abstract arxiv concerns cs.cl cs.cr cs.lg data differential differential privacy diverse domains fine-tuning language language model language models large language large language models llms mind model fine-tuning potential privacy raise solution tasks tools tuning type while