UNIST ECE students, (Jiyeon Han, Kyowoon Lee and Anh Tong), and Prof. Jaesik Choi’s paper has been accepted at IJCAI 2019
2019.05.20
Authors: Jiyeon Han*, Kyowoon Lee* , Anh Tong, Jaesik Choi (corresponding author) (* contributed equally)
Title: Confirmatory Bayesian Online Change Point Detection in the Covariance Structure of Gaussian Processes
Conference: International Joint Conference on Artificial Intelligence (IJCAI 2019)
A research paper entitled “Confirmatory Bayesian Online Change Point Detection in the Covariance Structure of Gaussian Processes” has been accepted to IJCAI (International Joint Conference on Artificial Intelligence) 2019. The International Joint Conference on Artificial Intelligence (IJCAI) presents premier international gatherings of AI researchers and practitioners.
The paper proposes statistical hypothesis tests for detecting covariance structure changes in locally smooth time series modeled by Gaussian Processes (GPs) and proposes an online change point detection algorithm called Confirmatory Bayesian Online Change Point Detection (Confirmatory BOCPD). The research team provides theoretically justified thresholds for the tests and uses them to improve BOCPD by confirming statistically significant changes and non-changes. The proposed algorithm outperforms in predictions on both synthetic and real-world datasets in terms of regression error and likelihood.