Differentially Private Federated Low Rank Adaptation Beyond Fixed-Matrix

TL;DR

This paper introduces FedASK, a novel federated LoRA framework that enables effective, privacy-preserving updating of low-rank adapters in large language models using a double sketching approach inspired by randomized SVD.

Contribution

FedASK is the first framework to allow robust differential privacy for updating both adapters in federated LoRA, overcoming previous limitations with a two-stage sketching pipeline.

Findings

FedASK provides strong differential privacy guarantees.

It outperforms baseline methods across various privacy settings.

The approach effectively updates both adapters in federated LLM fine-tuning.

Abstract

Large language models (LLMs) typically require fine-tuning for domain-specific tasks, and LoRA offers a computationally efficient approach by training low-rank adapters. LoRA is also communication-efficient for federated LLMs when multiple users collaboratively fine-tune a global LLM model without sharing their proprietary raw data. However, even the transmission of local adapters between a server and clients risks serious privacy leakage. Applying differential privacy (DP) to federated LoRA encounters a dilemma: adding noise to both adapters amplifies synthetic noise on the model, while fixing one adapter impairs the learnability of fine-tuning. In this paper, we propose FedASK (Differentially Private Federated Low Rank Adaptation with Double Sketching) , a novel federated LoRA framework to enable effective updating of both low-rank adapters with robust differential privacy. Inspired by randomized SVD, our key idea is a two-stage sketching pipeline. This pipeline first aggregates carefully sketched, privacy-preserving local updates, and then reconstructs the global matrices on the server to facilitate effective updating of both adapters. We theoretically prove FedASK's differential privacy guarantee and its exact aggregation property. Comprehensive experiments demonstrate that FedASK consistently outperforms baseline methods across a variety of privacy settings and data distributions.