SMA-DP-SGD augments DP-SGD with a spectral memory-aware fractional branch from prior privatized updates to improve accuracy on CIFAR and MNIST while preserving conditional differential privacy.
Private empirical risk minimization: Efficient algorithms and tight error bounds
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SMA-DP: Spectral Memory-Aware Differential Privacy for Deep Learning
SMA-DP-SGD augments DP-SGD with a spectral memory-aware fractional branch from prior privatized updates to improve accuracy on CIFAR and MNIST while preserving conditional differential privacy.