Privacy Amplification by Mixing and Diffusion Mechanisms

Authors: Borja Balle, Gilles Barthe, Marco Gaboardi, Joseph Geumlek

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical In this paper we initiate a systematic study of privacy amplification by stochastic post-processing. Specifically, given a DP mechanism M producing (probabilistic) outputs in X and a Markov operator K defining a stochastic transition between X and Y, we are interested in measuring the privacy of the post-processed mechanism K M producing outputs in Y.
Researcher Affiliation Academia Borja Balle Gilles Barthe MPI for Security and Privacy IMDEA Software Institute Marco Gaboardi Boston University Joseph Geumlek University of California, San Diego
Pseudocode Yes Algorithm 1: Noisy Projected Stochastic Gradient Descent. Input: Dataset D = (z1, . . . , zn), loss function ℓ: K D R, learning rate η, noise parameter σ, initial distribution ξ0 P(K) Sample x0 ξ0 for i [n] do xi ΠK (xi 1 η( xℓ(xi 1, zi) + Z)) with Z N(0, σ2I) return xn
Open Source Code No The paper does not provide any statement about releasing source code or links to a code repository for the described methodology.
Open Datasets No The paper is a theoretical work focusing on mathematical analysis and proofs of privacy amplification mechanisms. It does not involve empirical training on datasets.
Dataset Splits No The paper is theoretical and does not conduct empirical experiments with dataset splits for training, validation, or testing.
Hardware Specification No The paper is theoretical and does not describe any empirical experiments that would require specific hardware specifications.
Software Dependencies No The paper is theoretical and focuses on mathematical analysis; thus, it does not describe any software implementations or list software dependencies with version numbers.
Experiment Setup No The paper is theoretical and does not describe an experimental setup, hyperparameters, or training configurations.