Toward Training at ImageNet Scale with Differential Privacy
This work addresses the problem of scaling DP training to real-world, large-scale datasets like ImageNet for researchers and practitioners, though it is incremental in nature.
The paper tackles the challenge of training large neural networks with differential privacy (DP) while maintaining reasonable accuracy, using ImageNet classification as a test case. It achieves 47.9% accuracy with DP parameters ε=10, δ=10^{-6} for a ResNet-18 model, a significant improvement over naive DP training but still far from the 75% non-private baseline.
Differential privacy (DP) is the de facto standard for training machine learning (ML) models, including neural networks, while ensuring the privacy of individual examples in the training set. Despite a rich literature on how to train ML models with differential privacy, it remains extremely challenging to train real-life, large neural networks with both reasonable accuracy and privacy. We set out to investigate how to do this, using ImageNet image classification as a poster example of an ML task that is very challenging to resolve accurately with DP right now. This paper shares initial lessons from our effort, in the hope that it will inspire and inform other researchers to explore DP training at scale. We show approaches that help make DP training faster, as well as model types and settings of the training process that tend to work better in the DP setting. Combined, the methods we discuss let us train a Resnet-18 with DP to $47.9\%$ accuracy and privacy parameters $ε= 10, δ= 10^{-6}$. This is a significant improvement over "naive" DP training of ImageNet models, but a far cry from the $75\%$ accuracy that can be obtained by the same network without privacy. The model we use was pretrained on the Places365 data set as a starting point. We share our code at https://github.com/google-research/dp-imagenet, calling for others to build upon this new baseline to further improve DP at scale.