Exactly Minimax-Optimal Locally Differentially Private Sampling

Authors: Hyun-Young Park, Shahab Asoodeh, Si-Hyeon Lee

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our numerical experiments demonstrate the superiority of our mechanisms over baselines, in terms of theoretical utilities for finite data space and of empirical utilities for continuous data space.
Researcher Affiliation Academia Hyun-Young Park School of Electrical Engineering KAIST phy811@kaist.ac.kr Shahab Asoodeh Department of Computing and Software Mc Master University asoodeh@mcmaster.ca Si-Hyeon Lee School of Electrical Engineering KAIST sihyeon@kaist.ac.kr
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes All codes for experiments and figures are attached as a supplementary material, and can be found at the online repository1. The instructions to reproduce the results in the paper are in Appendix H. 1https://github.com/phy811/Optimal-LDP-Sampling.
Open Datasets No The paper describes generating synthetic data for its experiments: "We randomly construct N Gaussian mixture distributions P1, P2, ..., PN ∈ P, where each Pj is generated independently according to some rules specified in Appendix F.2." It does not use or provide access to a pre-existing publicly available dataset for training.
Dataset Splits No The paper does not explicitly mention the use of a validation dataset split in its experimental setup.
Hardware Specification Yes All experiments are performed on our simulation PC with the following specifications: OS: Ubuntu 22.04.1 CPU: Intel(R) Core(TM) i9-9900X Memory: 64GB
Software Dependencies No The paper mentions using Python and notes that the baseline used Julia, but it does not specify version numbers for Python, Julia, or any specific libraries (e.g., PyTorch, NumPy) used in the implementation within the main text or appendices. It mentions an 'environment.yaml' file in supplementary material, but the specific versions are not listed in the paper itself.
Experiment Setup Yes In the experiment in the paper, we use k0 = 2 and δ1 = δ2 = 10-5. For the baseline mechanism, we use MBDE with the same hyperparameter setup as [35, Section 5].