Instance-Optimal Private Density Estimation in the Wasserstein Distance

Authors: Vitaly Feldman, Audra McMillan, Satchit Sivakumar, Kunal Talwar

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

Reproducibility Variable Result LLM Response
Research Type Theoretical We design and analyze instance-optimal algorithms for this problem that can adapt to easy instances. ... In order to show that the instance optimality definition is achievable, we give both algorithmic upper bounds and matching, up to logarithmic factors, theoretical lower bounds.
Researcher Affiliation Collaboration Vitaly Feldman Apple Audra Mc Millan Apple Satchit Sivakumar Boston University Kunal Talwar Apple
Pseudocode Yes Algorithm 1 Priv Density Est Tree
Open Source Code No We also explain the method and hyperparameters we use, but don t release the codehowever, the complete descriptions of the algorithms are publicly available in previous work.
Open Datasets Yes The distribution: We have taken a distribution on [0, 999], which is concentrated on two points 430 and 440, with p430 = 1/3 and p440 = 2/3.
Dataset Splits No The paper uses a single set of 1600 samples from a defined distribution for estimation and does not specify explicit training, validation, or test splits.
Hardware Specification No The paper is theoretical and does not specify any hardware used for the described experiment.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers.
Experiment Setup Yes In this particular case, we use k = 10 quantiles with ε = 1. ... Here we consider this algorithm with ε = , so that no noise is added. We have run it here with K = 40 buckets.