Differentially Private Language Models for Secure Data Sharing

by   Justus Mattern, et al.

To protect the privacy of individuals whose data is being shared, it is of high importance to develop methods allowing researchers and companies to release textual data while providing formal privacy guarantees to its originators. In the field of NLP, substantial efforts have been directed at building mechanisms following the framework of local differential privacy, thereby anonymizing individual text samples before releasing them. In practice, these approaches are often dissatisfying in terms of the quality of their output language due to the strong noise required for local differential privacy. In this paper, we approach the problem at hand using global differential privacy, particularly by training a generative language model in a differentially private manner and consequently sampling data from it. Using natural language prompts and a new prompt-mismatch loss, we are able to create highly accurate and fluent textual datasets taking on specific desired attributes such as sentiment or topic and resembling statistical properties of the training data. We perform thorough experiments indicating that our synthetic datasets do not leak information from our original data and are of high language quality and highly suitable for training models for further analysis on real-world data. Notably, we also demonstrate that training classifiers on private synthetic data outperforms directly training classifiers on real data with DP-SGD.


page 1

page 2

page 3

page 4


Differential Privacy Made Easy

Data privacy is a major issue for many decades, several techniques have ...

The Limits of Word Level Differential Privacy

As the issues of privacy and trust are receiving increasing attention wi...

Privately generating tabular data using language models

Privately generating synthetic data from a table is an important brick o...

Differentially Private Distributed Learning for Language Modeling Tasks

One of the big challenges in machine learning applications is that train...

Optimal Local Bayesian Differential Privacy over Markov Chains

In the literature of data privacy, differential privacy is the most popu...

Chasing Accuracy and Privacy, and Catching Both: A Literature Survey on Differentially Private Histogram Publication

Histograms and synthetic data are of key importance in data analysis. Ho...

Identification and Formal Privacy Guarantees

Empirical economic research crucially relies on highly sensitive individ...

Please sign up or login with your details

Forgot password? Click here to reset