Poster
PrE-Text: Training Language Models on Private Federated Data in the Age of LLMs
Charlie Hou · Akshat Shrivastava · Hongyuan Zhan · Rylan Conway · Trang Le · Adithya Sagar · Giulia Fanti · Daniel Lazar
Hall C 4-9 #2307
[
Abstract
]
[ Paper PDF ]
Oral
presentation:
Oral 2C Privacy
Tue 23 Jul 7:30 a.m. PDT — 8:30 a.m. PDT
Tue 23 Jul 4:30 a.m. PDT
— 6 a.m. PDT
Tue 23 Jul 7:30 a.m. PDT — 8:30 a.m. PDT
Abstract:
On-device training is currently the most common approach for training machine learning (ML) models on private, distributed user data. Despite this, on-device training has several drawbacks: (1) most user devices are too small to train large models on-device, (2) on-device training is communication- and computation-intensive, and (3) on-device training can be difficult to debug and deploy. To address these problems, we propose Private Evolution-Text (PrE-Text), a method for generating differentially private (DP) synthetic textual data. First, we show that across multiple datasets, training small models (models that fit on user devices) with PrE-Text synthetic data outperforms small models trained on-device under practical privacy regimes ($\epsilon=1.29$, $\epsilon=7.58$). We achieve these results while using 9$\times$ fewer rounds, 6$\times$ less client computation per round, and 100$\times$ less communication per round. Second, finetuning large models on PrE-Text's DP synthetic data improves large language model (LLM) performance on private data across the same range of privacy budgets. Altogether, these results suggest that training on DP synthetic data can be a better option than training a model on-device on private distributed data. Code is available at https://github.com/houcharlie/PrE-Text.
Live content is unavailable. Log in and register to view live content