DP-MemArc: Differential Privacy Transfer Learning for Memory Efficient Language Models

2024-06-16

AI Summary

To address the dual challenges of excessive memory overhead and user privacy leakage in large language model (LLM) deployment, this paper proposes the first differentially private (DP) fine-tuning framework integrating side networks and invertible architectures. Our method employs memory-aware gradient computation and a privacy-accuracy co-optimization strategy, achieving significant GPU memory reduction under strict DP guarantees (ε ≤ 2). Compared to standard DP-SGD, it overcomes inherent bottlenecks of high memory consumption and training inefficiency: across multiple tasks, it attains up to 58% memory compression while incurring only a 1.2-point drop in GLUE average score—preserving accuracy comparable to non-private baselines. The core contribution lies in the novel integration of invertible networks and side networks into DP fine-tuning, enabling, for the first time, simultaneous optimization of strong privacy protection and memory efficiency.

Technology Category

• Machine Learning: Privacy
• Natural Language Processing: Safety and Robustness
• Search and Optimization: Learning to Search

Application Category

• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Large language models have repeatedly shown outstanding performance across diverse applications. However, deploying these models can inadvertently risk user privacy. The significant memory demands during training pose a major challenge in terms of resource consumption. This substantial size places a heavy load on memory resources, raising considerable practical concerns. In this paper, we introduce DP-MemArc, a novel training framework aimed at reducing the memory costs of large language models while emphasizing the protection of user data privacy. DP-MemArc incorporates side network or reversible network designs to support a variety of differential privacy memory-efficient fine-tuning schemes. Our approach not only achieves in memory optimization but also ensures robust privacy protection, keeping user data secure and confidential. Extensive experiments have demonstrated that DP-MemArc effectively provides differential privacy-efficient fine-tuning across different task scenarios.

Problem

• Reduces memory costs for large language models
• Ensures robust protection of user data privacy
• Supports differential privacy-efficient fine-tuning schemes

Innovation

• Differential privacy transfer learning
• Memory efficient language models
• Reversible network designs