Differentially Private Fractional Frequency Moments Estimation with Polylogarithmic Space

Authors: Lun Wang, Iosif Pinelis, Dawn Song

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The evaluation shows that Fp sketch can achieve reasonable accuracy with differential privacy guarantee. The evaluation code is included in the supplementary material. 4 EVALUATION
Researcher Affiliation Academia Lun Wang UC Berkeley wanglun@berkeley.edu Iosif Pinelis Michigan Tech ipinelis@mtu.edu Dawn Song UC Berkeley dawnsong@cs.berkeley.edu
Pseudocode Yes Algorithm 1: Fp sketch. Algorithm 2: Fp sketch with sub-sampling.
Open Source Code Yes The evaluation code is included in the supplementary material.
Open Datasets Yes Real-world Data. We also evaluate Fp sketches using real-world application usage data (Ye et al., 2019) collected by Talking Data SDK.
Dataset Splits No The paper does not explicitly provide training/test/validation dataset splits, only mentions data sampling and multiple runs for statistical analysis.
Hardware Specification Yes All the experiments were run on a Ubuntu18.04 LTS server with 32 AMD Opteron(TM) Processor 6212 with 512GB RAM.
Software Dependencies No The paper mentions 'Ubuntu18.04 LTS server' but does not specify other software dependencies with version numbers.
Experiment Setup Yes For all the evaluation, the sketch size r is 50 as suggested in Li (2008). The sub-sampling rate in all the experiments is 0.02.