Bandit Quickest Changepoint Detection

Authors: Aditya Gopalan, Braghadeesh Lakshminarayanan, Venkatesh Saligrama

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

Reproducibility Variable Result LLM Response
Research Type Experimental We then perform a number of experiments on synthetic and real datasets demonstrating the effectiveness of our proposed method. ... We perform numerical experiments of ϵ-GCD on synthetic and real datasets and show that under variations of changepoints, anomalies, and action sets, we realize gains due to adaptive sensing. ... Our goal in this section is to illustrate various aspects of our theory through experiments on synthetic environments, and to explore the performance of the proposed ϵ-GCD algorithm on a setting based on a real-world dataset.
Researcher Affiliation Academia Aditya Gopalan Electrical Communication Engineering Indian Institute of Science, Bengaluru 560012 aditya@iisc.ac.in Braghadeesh Lakshminarayanan Electrical Communication Engineering Indian Institute of Science, Bengaluru 560012 braghadeesh94@gmail.com Venkatesh Saligrama Electrical and Computer Engineering Boston University, Boston, MA 02215 srv@bu.edu
Pseudocode Yes Algorithm 1 ϵ-GCD
Open Source Code No The paper does not provide a direct link to source code or explicitly state that the code for the methodology is being released.
Open Datasets Yes We experiment using the MIMII audio dataset [Pur+19b].
Dataset Splits No The paper describes how the MIMII audio dataset was processed (e.g., concatenating files), but does not specify explicit train/test/validation dataset splits with percentages or sample counts.
Hardware Specification Yes All experiments were performed on a laptop with Intel Core i5 CPU and 8GB of RAM, and take under an hour to execute.
Software Dependencies No The paper mentions training autoencoders using mel-spectrogram features and fitting Gaussians, but it does not specify any software libraries or packages with version numbers used for these tasks (e.g., TensorFlow, PyTorch, Scikit-learn, etc. with versions).
Experiment Setup Yes We choose σ2 = 0.5, β such that the false alarms are about 1% for the Oracle, and a forced exploration rate ϵ = 0.2. Our results are averaged over 5000 Monte-Carlo runs.