Differentially Private Graph Learning via Sensitivity-Bounded Personalized PageRank
Authors: Alessandro Epasto, Vahab Mirrokni, Bryan Perozzi, Anton Tsitsulin, Peilin Zhong
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To complement our theoretical analysis, we also empirically verify the practical performances of our algorithms. In this section, we complement our theoretical analysis with an experimental study of our algorithms in terms of the accuracy of the PPR ranking and node classification. |
| Researcher Affiliation | Industry | Alessandro Epasto Google Research aepasto@google.com Vahab Mirrokni Google Research mirrokni@google.com Bryan Perozzi Google Research bperozzi@google.com Anton Tsitsulin Google Research tsitsulin@google.com Peilin Zhong Google Research peilinz@google.com |
| Pseudocode | Yes | Algorithm 1 PUSHFLOW(G, s, α, ξ), Algorithm 2 PUSHFLOWCAP(G, s, α, ξ, σ, type), Algorithm 3 DPPUSHFLOWCAP(G, s, α, ξ, σ, ε, type), Algorithm 4 INSTANTEMBEDDING(p, k) |
| Open Source Code | No | The paper does not provide any statement or link indicating the release of source code for the described methodology. |
| Open Datasets | Yes | We experiment on 2 publicly available datasets available from [14, 28]. Table 1 reports basic statistics about these datasets. POS is a word co-occurrence network built from Wikipedia. Blog Catalog a social network of bloggers from the blogcatalog website. |
| Dataset Splits | No | The paper mentions evaluating performance and metrics, but does not specify exact train/validation/test dataset splits or cross-validation setup. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., library names, framework versions, or exact solver versions) used in the experiments. |
| Experiment Setup | Yes | Hyperparameter settings. Unless otherwise specified, we ran all experiments that use our PUSHFLOWCAP algorithm (or the non-private PUSHFLOW algorithm [3]) to obtain PPR rankings consistent to the setting commonly used in the literature of α = 0.15.4 Moreover, we set ξ so that the number of iterations R = 100 in all algorithms. We use embedding dimensionality k = 256. We observe that our algorithm s utility is not strongly affected by the sensitivity parameter σ. For simplicity, we always set σ = 10 6, as this parameter generalized across different datasets tested. |