Breaking the Communication-Privacy-Accuracy Tradeoff with $f$-Differential Privacy

Authors: Richeng Jin, Zhonggen Su, caijun zhong, Zhaoyang Zhang, Tony Quek, Huaiyu Dai

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we examine the performance of the proposed ternary compressor in the case of distributed mean estimation. We follow the set-up of [9] and generate N = 1000 user vectors with dimension d = 250, i.e., x1, ..., x N R250. Each local vector has bounded l2 and l norms, i.e., ||xi||2 C = 1 and ||xi|| c = 1. Fig. 4 compares the proposed ternary stochastic compressor with SQKR and the Gaussian mechanism.
Researcher Affiliation Academia Richeng Jin1 Zhonggen Su1 Caijun Zhong1 Zhaoyang Zhang1 Tony Q.S. Quek2 Huaiyu Dai3 1Zhejiang University 2Singapore University of Technology and Design 3North Carolina State University
Pseudocode Yes Algorithm 1 Binomial Noise [6] ... Algorithm 2 Binomial Mechanism [9] ... Algorithm 3 Poisson Binomial Mechanism
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets No The paper states that data was generated for the experiments ('generate N = 1000 user vectors'), but does not provide access information (link, DOI, citation) for this generated dataset to make it publicly available or reproducible.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning into training, validation, or test sets.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes In this section, we examine the performance of the proposed ternary compressor in the case of distributed mean estimation. We follow the set-up of [9] and generate N = 1000 user vectors with dimension d = 250, i.e., x1, ..., x N R250. Each local vector has bounded l2 and l norms, i.e., ||xi||2 C = 1 and ||xi|| c = 1.