Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Synthetic Text Generation with Differential Privacy: A Simple and Practical Recipe

About

Privacy concerns have attracted increasing attention in data-driven products due to the tendency of machine learning models to memorize sensitive training data. Generating synthetic versions of such data with a formal privacy guarantee, such as differential privacy (DP), provides a promising path to mitigating these privacy concerns, but previous approaches in this direction have typically failed to produce synthetic data of high quality. In this work, we show that a simple and practical recipe in the text domain is effective: simply fine-tuning a pretrained generative language model with DP enables the model to generate useful synthetic text with strong privacy protection. Through extensive empirical analyses on both benchmark and private customer data, we demonstrate that our method produces synthetic text that is competitive in terms of utility with its non-private counterpart, meanwhile providing strong protection against potential privacy leakages.

Xiang Yue, Huseyin A. Inan, Xuechen Li, Girish Kumar, Julia McAnallen, Hoda Shajari, Huan Sun, David Levitan, Robert Sim• 2022

Related benchmarks

TaskDatasetResultRank
Structured text synthesisWater
RoBERTa ROC-AUC83.69
24
Structured text synthesisLOAN
RoBERTa ROC-AUC54.76
24
Structured text synthesisMIMIC
RoBERTa ROC-AUC56.22
24
Distributional and Semantic Similarity EvaluationREDDIT
FID0.06
12
Lexical DiversityREDDIT
Self-BLEU22
12
Next-word predictionREDDIT
Accuracy24.4
12
Membership Inference AttackReddit (test)
AUC (PPL)44.3
12
Privacy EvaluationWater
NRS1
5
Privacy EvaluationMIMIC
NRS1
5
Privacy EvaluationLOAN
NRS1
5
Showing 10 of 10 rows

Other info

Follow for update