Mean Estimation in the Add-Remove Model of Differential Privacy

Authors: Alex Kulesza, Ananda Theertha Suresh, Yuyan Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental In Figure 3 we plot the empirical performance of the algorithms discussed in Sections 3 and 4.1 on synthetic datasets and explore how the performance changes with parameters such as the privacy budget ε and the true mean µ. The underlying datasets are generated i.i.d. with varying µ in the range [ℓ= 0, u = 1]. All datasets have 10, 000 points, and mean squared error is computed over 100, 000 runs of each algorithm.
Researcher Affiliation Industry 1Google Research, NYC.
Pseudocode Yes Algorithm 1 Independent noise addition. Input: Multiset D [l, u], ε > 0. ... Algorithm 2 Shifted noise addition. Input: Multiset D [l, u], ε > 0. ... Algorithm 3 Transformed noise addition. Input: Multiset D [l, u], ε > 0.
Open Source Code No No explicit statement or link indicating the provision of open-source code for the methodology described in this paper.
Open Datasets No The underlying datasets are generated i.i.d. with varying µ in the range [ℓ= 0, u = 1].
Dataset Splits No No specific dataset split information (percentages, sample counts, or citations to predefined splits) for training, validation, or testing is provided.
Hardware Specification No No specific hardware details (e.g., exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running experiments are provided.
Software Dependencies No No specific ancillary software details (e.g., library or solver names with version numbers) are provided.
Experiment Setup Yes The underlying datasets are generated i.i.d. with varying µ in the range [ℓ= 0, u = 1]. All datasets have 10, 000 points, and mean squared error is computed over 100, 000 runs of each algorithm.