Differentially Private Learning Needs Better Model Initialization and Self-Distillation

TL;DR

This paper introduces DPRefine, a three-phase method that enhances differentially private language model training by improving initialization and output quality, significantly outperforming standard DPSGD in utility and linguistic accuracy.

Contribution

The paper proposes DPRefine, a novel three-phase approach combining data synthesis, DP finetuning, and self-distillation to improve privacy-preserving language model training.

Findings

DPRefine outperforms vanilla DPSGD in human evaluations.

Reduces linguistic errors in generated text by 84%.

Small models like GPT-2 are effective for initialization and distillation.

Abstract

Differentially private SGD (DPSGD) enables privacy-preserving training of language models, but often reduces utility, diversity, and linguistic quality. We introduce DPRefine, a three-phase method that initializes a model using data synthesis from a small pre-trained LM with rigorous filtering, applies DP finetuning on private data, and performs self-distillation to refine outputs. This approach significantly outperforms vanilla DPSGD, with AlpacaEval preferring DPRefine's generations in 78.4% of cases across all datasets. Our analysis reveals that DPRefine reduces linguistic errors in generated text by 84.0%, mitigating grammar and spelling errors, commonly associated with DPSGD. It also reduces inconsistencies of non-private models, such as hallucinated details and misattributed quotes. We find that small models like GPT-2 can be effective for initialization and distillation, highlighting their potential in enabling scalable and efficient deployment of privacy-preserving language.