Muffliato: Peer-to-Peer Privacy Amplification for Decentralized Optimization and Averaging

Authors: Edwige Cyffers, Mathieu Even, Aurélien Bellet, Laurent Massoulié

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we illustrate our privacy gains with experiments on synthetic and real-world datasets. ... We demonstrate the usefulness of our approach and analysis through experiments on synthetic and real-world datasets and network graphs.
Researcher Affiliation Academia 1Univ. Lille, Inria, CNRS, Centrale Lille, UMR 9189 CRISt AL, F-59000 Lille 2Inria Paris Département d informatique de l ENS, PSL Research University, Paris, France
Pseudocode Yes Algorithm 1: MUFFLIATO ... Algorithm 2: RANDOMIZED MUFFLIATO ... Algorithm 3: MUFFLIATO-SGD and MUFFLIATO-GD
Open Source Code Yes The code used to obtain these results is available at https://github.com/totilas/muffliato.
Open Datasets Yes We use a binarized version of UCI Housing dataset.3 [footnote: 3https://www.openml.org/d/823] ... We consider the graphs of the Facebook ego dataset [41]
Dataset Splits Yes We split the dataset uniformly at random into a training set (80%) and a test set
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies or their version numbers (e.g., Python, PyTorch, specific libraries).
Experiment Setup Yes Input: initial points θ0 v RD, number of iterations T, step sizes ν > 0, noise variance σ2, gossip matrices (Wt)t 0, local functions ϕv, number of communication rounds K ... After each gradient step of Muffliato-GD, we draw at random an Erdös-Rényi graph of same parameter q to perform the gossiping step and run the theoretical number of steps required for convergence.