DR-Encoder: Encode Low-rank Gradients with Random Prior for Large Language Models Differentially Privately
This addresses privacy leakage risks in federated learning for LLMs, offering a solution for secure fine-tuning, though it appears incremental as it builds on existing differential privacy and federated learning techniques.
The paper tackles the problem of ensuring end-to-end privacy during federated fine-tuning of large language models (FedLLM) by proposing DR-Encoder, which uses a two-stage randomness approach with gradient auto-encoders and Gaussian noise, achieving efficiency and accuracy gains as demonstrated on foundation models and benchmarks.
The emergence of the Large Language Model (LLM) has shown their superiority in a wide range of disciplines, including language understanding and translation, relational logic reasoning, and even partial differential equations solving. The transformer is the pervasive backbone architecture for the foundation model construction. It is vital to research how to adjust the Transformer architecture to achieve an end-to-end privacy guarantee in LLM fine-tuning. In this paper, we investigate three potential information leakage during a federated fine-tuning procedure for LLM (FedLLM). Based on the potential information leakage, we provide an end-to-end privacy guarantee solution for FedLLM by inserting two-stage randomness. The first stage is to train a gradient auto-encoder with a Gaussian random prior based on the statistical information of the gradients generated by local clients. The second stage is to fine-tune the overall LLM with a differential privacy guarantee by adopting appropriate Gaussian noises. We show the efficiency and accuracy gains of our proposed method with several foundation models and two popular evaluation benchmarks. Furthermore, we present a comprehensive privacy analysis with Gaussian Differential Privacy (GDP) and Renyi Differential Privacy (RDP).