Tuning-free Estimation and Inference of Cumulative Distribution Function under Local Differential Privacy

Authors: Yi Liu, Qirui Hu, Linglong Kong

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

Reproducibility Variable Result LLM Response
Research Type Experimental Through mathematical proofs and extensive numerical testing, we demonstrate that our method achieves uniform and L2 error bounds when estimating the entire CDF curve. Computationally, we demonstrate that our constrained isotonic estimator can be efficiently computed deterministically, eliminating the need for hyperparameters or random optimization.
Researcher Affiliation Academia 1Department of Mathematical and Statistical Sciences, University of Alberta, Edmonton, Canada 2Center for Statistical Science, Department of Industrial Engineering, Tsinghua University, Beijing, China.
Pseudocode Yes Algorithm 1 Constrained isotonic estimation
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets No The paper uses data generated from standard statistical distributions (Uniform, Truncated normal, Continuous Bernoulli) for its experiments, rather than pre-existing public datasets with specific access information. Thus, there is no public dataset for which access information would be provided.
Dataset Splits No The paper conducts experiments by generating data from specified distributions and varying sample sizes (n spans from 10^3 to 10^7, with a total of 10,000 replications). It does not explicitly define or refer to standard train/validation/test dataset splits in percentages or counts for reproduction.
Hardware Specification Yes For n 10^7, it took less than 1s to execute on a single core of an AMD Threadripper PRO 3995WX CPU.
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes For the Truncated normal distribution, the parameters are set as µ = 1/2 and σ2 = 1/4. In the case of the Continuous Bernoulli distribution, the parameter λ is selected to be 1/4... We consider the truthful response rate r = 0.25, 0.5, 0.9, which means the privacy budget is ϵ = log((1 + r)/(1 r)) corresponding to 0.51, 1.09, 2.94 respectively. The sample size ranges n spans from 10^3 to 10^7, with a total of 10,000 replications.