AI RESEARCH
SMA-DP: Spectral Memory-Aware Differential Privacy for Deep Learning
arXiv CS.LG
•
ArXi:2605.20450v1 Announce Type: new Differentially private stochastic gradient descent (DP-SGD) enables private deep learning through per-example clipping and calibrated Gaussian noise, but its high-variance updates can reduce utility on challenging datasets. We propose \textbf{SMA-DP-SGD}, a \textbf{Spectral Memory-Aware Differentially Private Stochastic Gradient Descent} method that augments DP-SGD with a fractional memory branch built only from previously privatized noisy releases.