Hubness Change Point Detection

Authors: Ikumi Suzuki, Kazuo Hara, Eiji Murakami

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments with synthetic data show that the proposed method achieves accuracy comparable to or exceeding that of existing methods. Additionally, the proposed method achieves good accuracy with real-world data from hydraulic systems and gas sensors, along with excellent runtime performance.
Researcher Affiliation Collaboration Ikumi Suzuki1, Kazuo Hara1, Eiji Murakami2 1Yamagata University 2Azbil Kimmon Co.,Ltd., Keio University EMAIL, EMAIL, EMAIL
Pseudocode Yes Algorithm 1: Residual Normalization Algorithm 2: Proposed Hub-CPD Algorithm
Open Source Code Yes The Appendix and code are available at https://analogy.sakura.ne.jp/AAAI25/appendix.pdf and https://analogy.sakura.ne.jp/AAAI25/code and data.zip.
Open Datasets Yes real-world data from hydraulic systems and gas sensors (Helwig and Eliseo Pignanelli 2015)2 and the concentration changes of carbon monoxide detected by gas sensors (Burgu es, Jim enez-Soto, and Marco 2018).3 2https://archive.ics.uci.edu/dataset/447 3https://archive.ics.uci.edu/dataset/487
Dataset Splits No The parameters of the change detection methods were determined using validation data. The paper discusses data generation and evaluation points but does not specify explicit train/test/validation percentages, sample counts, or a detailed splitting methodology for reproducibility beyond this general statement.
Hardware Specification Yes The execution was performed on an Intel Xeon Gold 6134 CPU with 526GB RAM and an NVIDIA Quadro P5000 GPU with 16GB SGRAM. KL-CPD was run using GPU.
Software Dependencies No Ru LSIF was run using MATLAB s built-in multithreading. No specific version numbers for MATLAB or other key software components used in the proposed method's implementation are provided in the main text.
Experiment Setup Yes Algorithm 2: Proposed Hub-CPD Algorithm Input: D = {v1, . . . , vn}, D = {v 1, . . . , v n}, k, r. The paper explicitly lists 'k' and 'r' as input parameters for the proposed algorithm and discusses their impact in the results section, for example, 'the r = 1 and r = 2 variants' and 'The behavior with respect to k differs'.