vix.ing · top · new · best · stats · spec

Planting and Mitigating Memorized Content in Predictive-Text Language Models

2022/12/16 by C. M. Downey, Downey, C. M., Wei Dai +9
Computer Science · #Computation and Language (cs.CL) #Cryptography and Security (cs.CR) #FOS: Computer and information sciences #Privacy-Preserving Technologies in Data #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2212.08619

openalex publication_date 2022/12/16 · openalex created_date 2023/01/03 · openalex updated_date 2026/07/28

Abstract

Language models are widely deployed to provide automatic text completion services in user products. However, recent research has revealed that language models (especially large ones) bear considerable risk of memorizing private training data, which is then vulnerable to leakage and extraction by adversaries. In this study, we test the efficacy of a range of privacy-preserving techniques to mitigate unintended memorization of sensitive user text, while varying other factors such as model size and adversarial conditions. We test both "heuristic" mitigations (those without formal privacy guarantees) and Differentially Private training, which provides provable levels of privacy at the cost of some model performance. Our experiments show that (with the exception of L2 regularization), heuristic mitigations are largely ineffective in preventing memorization in our test suite, possibly because they make too strong of assumptions about the characteristics that define "sensitive" or "private" text. In contrast, Differential Privacy reliably prevents memorization in our experiments, despite its computational and model-performance costs.

Related