Provably Confidential Language Modelling

By Xuandong Zhao, Lei Li, Yu-Xiang Wang. 2022

25 citations recorded by the source corpus.

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.

Subfields: Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Safety, Robustness & Interpretability

Citations in the corpus

Deduplicating Training Data Mitigates Privacy Risks in Language Models (2022, background): “(Zhao et al., 2022) propose provable confidentiality, a related guarantee that ensures that the content of particular secrets in the training data do not have a large effect on training.”
Preventing Verbatim Memorization in Language Models Gives a False Sense of Privacy (2022, background): “There is also a newly emerging line of works exploring differential-privacy (DP)-based definitions (Zhao et al., 2022; Stock et al., 2022), which relate to document-level DP guarantees in language modelling (Yu et al., 2021).”