Differentially Private Identity and Equivalence Testing of Discrete Distributions

Authors: Maryam Aliakbarpour, Ilias Diakonikolas, Ronitt Rubinfeld

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform an experimental evaluation of our algorithms on synthetic data. Our experiments illustrate that our private testers achieve small type I and type II errors with sample size sublinear in the domain size of the underlying distributions.
Researcher Affiliation Academia 1CSAIL, MIT, Cambridge, MA 02139, USA 2Department of Computer Science, USC, Los Angeles, CA 90089, USA 3TAU, Tel Aviv-Yafo, Israel.
Pseudocode Yes Algorithm 1 Private Uniformity Testing via Unique Elements: Private-Unique-Elements-Uniformity; Algorithm 2 Private uniformity tester based on the number of collisions: Private-Collisions-Uniformity; Algorithm 3 Private Equivalence Tester: Private Equivalence-Test
Open Source Code No The paper does not provide any explicit statements or links regarding the availability of its source code.
Open Datasets No The paper uses "synthetic data" which is mathematically defined within the paper but does not refer to a publicly available dataset with a specific link, DOI, repository, or formal citation.
Dataset Splits No The paper does not specify training, validation, or test dataset splits. It describes running algorithms multiple times to estimate error probabilities on generated samples.
Hardware Specification Yes All experiments were performed on a computer with a 1.6 GHz Intel(R) Core(TM) i5-4200U CPU and 3 GB of RAM.
Software Dependencies No The paper mentions implementing algorithms but does not provide specific version numbers for any software dependencies (e.g., programming languages, libraries, or frameworks).
Experiment Setup Yes We run our two algorithms using samples from q+ and q- with the following parameters: n = 800, 000, = 0.3, r = 300, and = 0.2. We vary the sample size staring from 50, 000 and up to 3 x 10^6, increasing it by 50, 000 at each step, and repeat the algorithm for r = 200 times to estimate the maximum of type I and type II errors.