Locally private online change point detection

Authors: Tom Berrett, Yi Yu

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section we present the results of a numerical study of our locally private method s performance.
Researcher Affiliation Academia Thomas Berrett Department of Statistics University of Warwick Coventry, CV4 7AL, U.K. tom.berrett@warwick.ac.uk Yi Yu Department of Statistics University of Warwick Coventry, CV4 7AL, U.K. yi.yu.2@warwick.ac.uk
Pseudocode Yes Algorithm 1 Online change point detection via CUSUM statistics and Algorithm 2 Online change point detection via CUSUM statistics
Open Source Code Yes Full details of the implementation and simulation study can be found in the code available online. The supplementary material contains all the technical details and code of this paper.
Open Datasets No The paper uses synthetic data generated for the numerical study ('raw data (X1, Y1), . . . , (Xn, Yn) with n = 10000, = 5000, Xi Unif[0, 1] and Yi Unif[mi(x) 1/2, mi(x) + 1/2]'). This data is generated by the authors for their simulation and not a publicly available dataset with concrete access information.
Dataset Splits No The paper does not explicitly provide training/test/validation dataset splits. It describes generating synthetic data and using permutations to select thresholds, but not a fixed data split for reproduction.
Hardware Specification No The paper does not specify any hardware details (e.g., CPU, GPU models, or specific computing infrastructure) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers used for the experiments.
Experiment Setup Yes The choices M = 1 and h = 0.2 are used to privatise the data and calculate the test statistics. We permute this sample B = 1000 times to choose our thresholds, as follows: for a range of values of C we run the modified Algorithm 1 on each permutation of the privatised data with the choice t h2α2 C2 log t γh ; , otherwise and we choose the minimal value of C for which the overall false alarm probability is bounded above by γ = 0.1. With the thresholds chosen we ran the experiment over 1000 repetitions