Restarted Bayesian Online Change-point Detector achieves Optimal Detection Delay

Authors: Reda Alami, Odalric Maillard, Raphael Feraud

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on synthetic and realworld data show that this proposal outperforms the state-of-art change-point detection strategy
Researcher Affiliation Collaboration 1Total S.A. 2INRIA-SCOOL Team 3Orange Labs.
Pseudocode Yes Algorithm 1 BOCPD (Fearnhead & Liu, 2007); Algorithm 2 R-BOCPD
Open Source Code Yes Software and simulation code is available at https://github.com/Ralami1859/Restarted-BOCPD.
Open Datasets Yes These data have been studied in the context of change-point detection by (Fearnhead & Clifford, 2003) and has become a benchmark data set for uni-variate changepoint detection.
Dataset Splits No No explicit details on train/validation/test splits, percentages, or cross-validation were provided for the datasets used.
Hardware Specification No No specific hardware (e.g., CPU, GPU models, or cloud instance types) used for experiments was mentioned.
Software Dependencies No No specific software dependencies with version numbers (e.g., library or solver names with versions) were provided.
Experiment Setup Yes In all the experiment, we choose ηr,s,t = 1 nr:t for RBOCPD and h = 3/1200 for BOCPD. The curves are averaged over 300 runs. (Their error bars are also plotted). The parameter δ (false alarm rate of Imp GLR) is set to 0.01.