Coordinate-wise adaptive clipping in DP-SGD, with per-coordinate scales derived from a noise-minimization problem, yields higher MNIST accuracy than standard L2 clipping at the same privacy budget.
Imagenet classification with deep convolutional neural networks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
AdaCliP: Adaptive Clipping for Private SGD
Coordinate-wise adaptive clipping in DP-SGD, with per-coordinate scales derived from a noise-minimization problem, yields higher MNIST accuracy than standard L2 clipping at the same privacy budget.