Differentially Private Bayesian Neural Networks on Accuracy, Privacy and Reliability. (arXiv:2107.08461v2 [cs.LG] UPDATED)

Bayesian neural network (BNN) allows for uncertainty quantification in
prediction, offering an advantage over regular neural networks that has not
been explored in the differential privacy (DP) framework. We fill this
important gap by leveraging recent development in Bayesian deep learning and
privacy accounting to offer a more precise analysis of the trade-off between
privacy and accuracy in BNN. We propose three DP-BNNs that characterize the
weight uncertainty for the same network architecture in distinct ways, namely
DP-SGLD (via the …

accounting accuracy analysis deep learning development differential privacy framework gap important network networks neural network neural networks offer prediction privacy private quantification reliability trade uncertainty