Differentially Private Deep Learning with Direct Feedback Alignment
This work addresses the challenge of maintaining model accuracy while ensuring privacy for deep learning applications, representing an incremental improvement over existing private training methods.
The paper tackles the problem of training deep neural networks with differential privacy, which typically reduces accuracy, by proposing a method using direct feedback alignment (DFA) instead of backpropagation. The result shows significant accuracy gains of 10-20% compared to backprop-based private training across various architectures and datasets.
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.