Differentially Private Graph Diffusion with Applications in Personalized PageRanks

Authors: Rongzhe Wei, Eli Chien, Pan Li

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on real-world network data demonstrate the superiority of our method under stringent privacy conditions. In this section, we present empirical evaluations to support our theoretical findings. Specifically, we apply the widely-used PPR algorithm (Sec. 3.4) to real-world graphs.
Researcher Affiliation Academia Rongzhe Wei Georgia Institute of Technology rongzhe.wei@gatech.edu Eli Chien Georgia Institute of Technology ichien6@gatech.edu Pan Li Georgia Institute of Technology panli@gatech.edu
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks with explicit labels like 'Algorithm' or 'Pseudocode'.
Open Source Code Yes We have uploaded our code in supplementary file. We have uploaded our code in the supplementary file with instructions (README.md) to reproduce the results.
Open Datasets Yes We conduct experiments on the following datasets: Blog Catalog [57], a social network of bloggers with 10,312 nodes and 333,983 edges; Flickr [57], a photo-sharing social network with 80,513 nodes and 5,899,882 edges; and The Marker [58], an online social network with 69,400 nodes and 1,600,000 edges.
Dataset Splits No The paper does not provide specific training, validation, or test dataset split information (percentages, counts, or specific predefined splits).
Hardware Specification Yes Experiments were performed on a server with two AMD EPYC 7763 64-Core Processors, 2TB DRAM, six NVIDIA RTX A6000 GPUs (each with 48GB of memory).
Software Dependencies No The paper does not explicitly list software dependencies with specific version numbers in its main text or appendices.
Experiment Setup Yes PPR with parameter β = 0.8. Considering that both our approach and DP-PUSHFLOWCAP employ a thresholding parameter η to balance the privacy-utility trade-off, we select η from a set of seven values spanning orders of magnitude from 10-10 to 10-4... For each experiment, we randomly choose an initial seed for diffusion and execute PPR for 100 iterations following [20].