Integer Subspace Differential Privacy
Authors: Prathamesh Dharangutte, Jie Gao, Ruobin Gong, Fang-Yi Yu
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the efficacy of our proposal with applications to a synthetic problem with intersecting invariants, a sensitive contingency table with known margins, and the 2010 Census county-level demonstration data with mandated fixed state population totals. |
| Researcher Affiliation | Academia | Prathamesh Dharangutte1, Jie Gao1, Ruobin Gong2, Fang-Yi Yu3 1Department of Computer Science, Rutgers University 2Department of Statistics, Rutgers University 3Department of Computer Science, George Mason University |
| Pseudocode | Yes | Algorithm 1 in Appendix D of this paper s full version (Dharangutte et al. 2022) presents a Gibbs-within Metropolis sampler that produces a sequences of dependent draws z(l) 0 <= l <= nsim from the target distribution q epsilon in (5) known only up to a normalizing constant. We use an additive jumping distribution whose element-wise construction is described in (11). The algorithm incurs a transition kernel that dictates how the chain moves from an existing state to the next one: z(l) ~ K(z(l-1)). |
| Open Source Code | No | The paper refers to an extended version on ArXiv for additional details and algorithms, but does not provide a direct link to any open-source code repository for the methodology described. |
| Open Datasets | Yes | The Federal Committee on Statistical Methodology published a fictitious dataset concerning delinquent children in the form of a 4 x 4 contingency table, tabulated across four counties by education level of household head (Table 4 in Federal Committee on Statistical Methodology 2005, reproduced in Table 2 of Appendix E). The confidential values are the 2010 Census Summary Files (CSF), curated by IPUMS NHGIS and are publicly available (Van Riper, Kugler, and Schroeder 2020). |
| Dataset Splits | No | The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning into training, validation, or test sets. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details, such as library or solver names with version numbers, needed to replicate the experiment. |
| Experiment Setup | Yes | To ensure adequate dispersion of the target distribution, we set epsilon = 0.25, a value on the smaller end within the range of meaningful privacy protection (e.g. Dwork 2011). The pre-jump proposal distributions eta_j are double geometric distributions, with parameter a = exp(-1) for the l1-norm target and a = exp(-1.5) for the l2-norm target. |