Graphical-TS: An Interactive AI Pipeline for Multivariate Time Series with Ground-truth Graphical Modeling
Keywords: time series, causal discovery, benchmarking, interface
Abstract: We present \texttt{Graphical-TS}, an interactive simulation framework for multivariate time series (MTS) incorporating spatiotemporal causal graphical models. The system offers extensive customizability, enabling users to define and modify causal dynamics with uncertainty in spatiotemporal relationships and functional mappings. \texttt{Graphical-TS} integrates expert knowledge, supports MTS simulation, and allows for the input of real-world MTS data, facilitating a dynamic interplay between data-driven learning and domain expertise. The system iteratively enhances causal relationships and simulated data by simulating MTS data based on specified causal graphs, performing causal discovery from real or simulated MTS, and enabling the integration and refinement of expert knowledge with learned causality. This approach progressively improves the quality of causal models and the data they generate, supporting tasks such as time series forecasting, imputation, prediction, and robustness testing via scenario-driven distribution shifts. We compared state-of-the-art causal discovery methods on datasets generated by \texttt{Graphical-TS}. The empirical results demonstrate the platform’s consistent performance compared to existing methods while offering versatility under distinct scenarios. This enables users to explore datasets more thoroughly and drive improvements in causal discovery research. With an intuitive user interface that connects domain experts and algorithm developers, \texttt{Graphical-TS} empowers users to manipulate causal relationships, embedding domain knowledge into machine learning workflows. Originally developed to study physiological dynamics in patients, the system has broad applicability across various fields, offering a versatile platform for generating MTS datasets with known dynamics, validating causal discovery algorithms, and advancing research in time series analysis.
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13763
Loading