English

Provably Confidential Language Modelling

Computation and Language 2022-06-27 v2 Cryptography and Security Machine Learning

Abstract

Large language models are shown to memorize privacy information such as social security numbers in training data. Given the sheer scale of the training corpus, it is challenging to screen and filter these privacy data, either manually or automatically. In this paper, we propose Confidentially Redacted Training (CRT), a method to train language generation models while protecting the confidential segments. We borrow ideas from differential privacy (which solves a related but distinct problem) and show that our method is able to provably prevent unintended memorization by randomizing parts of the training process. Moreover, we show that redaction with an approximately correct screening policy amplifies the confidentiality guarantee. We implement the method for both LSTM and GPT language models. Our experimental results show that the models trained by CRT obtain almost the same perplexity while preserving strong confidentiality.

Keywords

Cite

@article{arxiv.2205.01863,
  title  = {Provably Confidential Language Modelling},
  author = {Xuandong Zhao and Lei Li and Yu-Xiang Wang},
  journal= {arXiv preprint arXiv:2205.01863},
  year   = {2022}
}

Comments

NAACL 2022

R2 v1 2026-06-24T11:06:38.613Z