@article{Chen_Chang_Naiel_Zelek_2023, title={Causal Discovery from Sparse Time-Series Data Using Echo State Network}, volume={8}, url={https://openjournals.uwaterloo.ca/index.php/vsl/article/view/5373}, DOI={10.15353/jcvis.v8i1.5373}, abstractNote={<table style="height: 348px;" width="846"> <tbody> <tr> <td width="711">Causal discovery between collections of time-series data can help<br>diagnose causes of symptoms and hopefully prevent faults before<br>they occur. However, reliable causal discovery can be very chal-<br>lenging, especially when the data acquisition rate varies (i.e., non-<br>uniform data sampling), or in the presence of missing data points<br>(e.g., sparse data sampling). To address these issues, we propose<br>a new system comprised of two parts, the first part fills in missing data<br>with a Gaussian Process Regression, and the second part lever-<br>ages an Echo State Network, which is a type of reservoir computer<br>(i.e., used for chaotic system modelling) for Causal discovery.<br>We evaluate the performance of our proposed system against<br>three other off-the-shelf causal discovery algorithms, namely, struc-<br>tural expectation maximization, sub-sampled linear auto-regression<br>absolute coefficients, and multivariate Granger Causality with vector<br>auto-regressive using the Tennessee Eastman chemical dataset;<br>we report on their corresponding Matthews Correlation Coefficient<br>(MCC) and Receiver Operating Characteristic curves (ROC) and<br>show that the proposed system outperforms existing algorithms,<br>demonstrating the viability of our approach to discover causal re-<br>lationships in a complex system with missing entries.</td> </tr> </tbody> </table>}, number={1}, journal={Journal of Computational Vision and Imaging Systems}, author={Chen, Haonan and Chang, Bo Yuan and Naiel, Mohamed and Zelek, John}, year={2023}, month={May}, pages={42–44} }