Differentially Private Optimization on Large Model at Small Cost

TL;DR

This paper introduces a novel Book-Keeping technique that significantly reduces the computational cost of differentially private optimization for large neural networks, making privacy-preserving training more practical and efficient.

Contribution

The authors develop the Book-Keeping method that enables efficient DP training on large models with minimal overhead, matching the accuracy of existing DP optimizers while drastically reducing computational costs.

Findings

BK achieves near-standard training speed with minimal overhead.

It enables DP training on large models previously limited by memory and speed constraints.

Experimental results show state-of-the-art accuracy with significantly reduced computational costs.

Abstract

Differentially private (DP) optimization is the standard paradigm to learn large neural networks that are accurate and privacy-preserving. The computational cost for DP deep learning, however, is notoriously heavy due to the per-sample gradient clipping. Existing DP implementations are 2-1000X more costly in time and space complexity than the standard (non-private) training. In this work, we develop a novel Book-Keeping (BK) technique that implements existing DP optimizers (thus achieving the same accuracy), with a substantial improvement on the computational cost. Specifically, BK enables DP training on large models and high dimensional data to be roughly as fast and memory-saving as the standard training, whereas previous DP algorithms can be inefficient or incapable of training due to memory error. The computational advantage of BK is supported by the complexity analysis as well as extensive experiments on vision and language tasks. Our implementation achieves state-of-the-art (SOTA) accuracy with very small extra cost: on GPT2 and at almost the same memory cost (<1% overhead), BK has 1.03X the time complexity of the standard training (0.83X training speed in practice), and 0.61X the time complexity of the most efficient DP implementation (1.36X training speed in practice). We open-source the codebase for the BK algorithm at the FastDP library (https://github.com/awslabs/fast-differential-privacy).