Locally Private Hypothesis Testing
Authors: Or Sheffet
ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiment is aimed at determining whether P(θ) can serve as a test statistic and assessing its sample complexity. Setting and Default Values. We set a true ground distribution on T possible types, p. We then pick a distribution q which is α-far from p using the counter example of Paninski (2008): we pair the types and randomly move 2α T probability mess between each pair of matched types. We then generate n samples according to q, and apply the non-symmetric ϵ-differentially private mechanism of (Bassily et al., 2017). Finally, we aggregate the suitable vectors to obtain our estimator θ and compute P(θ). |
| Researcher Affiliation | Academia | Or Sheffet 1 1Dept. of Computing Science, University of Alberta.. Correspondence to: Or Sheffet <osheffet@ualberta.ca>. |
| Pseudocode | No | No structured pseudocode or algorithm blocks are present. Procedures are described in numbered steps within the text, such as in Section 3.1 for Independence Testing. |
| Open Source Code | No | The paper does not provide any specific links to source code repositories nor explicitly state that the code for their methodology is publicly available. |
| Open Datasets | No | The paper describes generating synthetic data for its experiments ('We set a true ground distribution on T possible types, p. We then pick a distribution q which is α-far from p...'). It does not use or provide access information for any established public datasets. |
| Dataset Splits | No | The paper does not provide specific details on dataset splits (training, validation, test) for reproducibility. It describes generating samples from distributions for experiments. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU or CPU models, or cloud computing instance specifications used for running the experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers, such as programming languages, libraries, or specialized solvers. |
| Experiment Setup | Yes | Setting and Default Values. We set a true ground distribution on T possible types, p. We then pick a distribution q which is α-far from p... We have set the default values T = 10, p = u T (uniform on [T]), α = 0.2, n = 1000, ϵ = 0.25 and therefore η = 1 2 eϵ 1 eϵ+1, and t = 10000. |