Training Data Leakage Analysis in Language Models
This work addresses privacy threats for users when language models are trained on confidential data, offering a way to measure and compare leakage risks, though it is incremental in building on existing concerns about memorization.
The paper tackles the problem of training data leakage in language models by introducing a methodology to identify user content that could be leaked under a strong threat model, proposing two metrics to quantify user-level data leakage and demonstrating their application through numerical studies on RNN and Transformer models.
Recent advances in neural network based language models lead to successful deployments of such models, improving user experience in various applications. It has been demonstrated that strong performance of language models comes along with the ability to memorize rare training samples, which poses serious privacy threats in case the model is trained on confidential user content. In this work, we introduce a methodology that investigates identifying the user content in the training data that could be leaked under a strong and realistic threat model. We propose two metrics to quantify user-level data leakage by measuring a model's ability to produce unique sentence fragments within training data. Our metrics further enable comparing different models trained on the same data in terms of privacy. We demonstrate our approach through extensive numerical studies on both RNN and Transformer based models. We further illustrate how the proposed metrics can be utilized to investigate the efficacy of mitigations like differentially private training or API hardening.