Martin Abadi, Andy Chu, Ian Goodfellow, H. Brendan McMahan, Ilya Mironov, Kunal Talwar, & Li Zhang (2016)
Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, 308-318.
DOI: https://doi.org/10.1145/2976749.2978318
Abstract. Introduces DP-SGD, a differentially private stochastic gradient descent algorithm that clips per-sample gradients and adds calibrated Gaussian noise, enabling training of deep networks with formal (ε, δ)-differential privacy guarantees.
Tags: privacy differential-privacy dp-sgd