GPriS: A Differentially Private Deep Learning Method with Gradient Sparsification

Huiqi Zhang

Advances in Computer and Materials Scienc Research Vol. 3 / No. 1 / pp. 205-211 / July 2, 2025

DOI 10.70114/acmsr.2025.3.1.P205

Full Screen
PDF

Abstract

Privacy protection has become a core issue in machine learning, with differential privacy widely regarded as an effective method for preserving privacy in stochastic gradient descent. We propose the GPriS (Gradient Privacy Sparse) method, a novel method that enhances differential privacy in deep learning by incorporating sparsity-based pruning. By leveraging the Lottery Hypothesis, GPriS identifies and preserves a lucky “sub-network” of critical parameters, reducing the number of parameters and improving both efficiency and privacy. Experiments on the MNIST dataset demonstrate that GPriS achieves 98.83% test accuracy with only 60% of gradient updates under a privacy budget of ε=4, a performance level comparable to non-privacy settings. Compared to traditional DP-SGD, GPriS offers a better trade-off between privacy, model accuracy, and training efficiency. Our results show that GPriS effectively allocates privacy budgets, ensuring optimal performance even under tight privacy constraints

Keywords

Differential Privacy / Deep Neural Networks / Gradient Sparsification / Importance-based Updates

2025 International Conference on Power Electronics, Biology, Building and Materials Science

Full Text

Download PDF

More in This Issue

  1. 1 Intelligent rubbish Bin Based on KJ-KANO Model Innovative Design Research pp. 1–10
  2. 11 Construction and Field Application Practice of Cement Slurry System Suitable for Ultra-Deep Wells with Ultra-Long Cementing Sections and Large Temperature Differences pp. 11–18
  3. 19 Layout Design of Nuclear Island for HPR1000 pp. 19–25
  4. 26 The Effect of PEW on Properties of Solid State Drawn UHMWPE pp. 26–32
Full Issue

Cite This Article