Differentially Private Deep Learning with Direct Feedback Alignment

TL;DR

This paper introduces a novel differentially private training method for deep neural networks using direct feedback alignment, which improves accuracy over traditional backpropagation-based methods under privacy constraints.

Contribution

It presents the first differentially private training approach utilizing direct feedback alignment, demonstrating significant accuracy improvements across various architectures and datasets.

Findings

Achieves 10-20% higher accuracy than backprop-based private training.

Effective across fully connected and convolutional neural networks.

Provides a new avenue for privacy-preserving deep learning.

Abstract

Standard methods for differentially private training of deep neural networks replace back-propagated mini-batch gradients with biased and noisy approximations to the gradient. These modifications to training often result in a privacy-preserving model that is significantly less accurate than its non-private counterpart. We hypothesize that alternative training algorithms may be more amenable to differential privacy. Specifically, we examine the suitability of direct feedback alignment (DFA). We propose the first differentially private method for training deep neural networks with DFA and show that it achieves significant gains in accuracy (often by 10-20%) compared to backprop-based differentially private training on a variety of architectures (fully connected, convolutional) and datasets.