Shifted Interpolation for Differential Privacy

arXiv:2403.00278v1 Announce Type: new
Abstract: Noisy gradient descent and its variants are the predominant algorithms for differentially private machine learning. It is a fundamental question to quantify their privacy leakage, yet tight characterizations remain open even in the foundational setting of convex losses. This paper improves over previous analyses by establishing (and refining) the "privacy amplification by iteration" phenomenon in the unifying framework of $f$-differential privacy--which tightly captures all aspects of the privacy loss and immediately implies tighter privacy accounting …

abstract algorithms arxiv cs.cr cs.lg differential differential privacy gradient losses machine machine learning math.oc math.st paper privacy question stat.ml stat.th type variants