Unified Framework for Causal Discovery and Long-term Forecasting in Non-stationary Environments
Keywords: causal discovery, long-term forecasting
Abstract: Non-stationary data is prevalent in various real-world domains such as climate science, economics, and neuroscience, presenting significant challenges for tasks like forecasting and causal discovery from observational data. Existing approaches often operate under the assumption that the data is stationary. In this work, we introduce a unified framework that combines long-term forecasting and causal discovery with non-linear relations in a non-stationary setting. Specifically, we assume that the nonlinear causal relations in the observed space can be transformed into linear relations in the latent space via projections. In addition, we model the non-stationarity in the system as arising from time-varying causal relations. The proposed model demonstrates that adopting a causal perspective for long-term forecasting not only addresses the limitations of each task but also makes the causal process identifiable, enhances interpretability, and provides more reliable predictions. Moreover, our approach reformulates causal discovery into a scalable, non-parametric deep learning problem. Through experiments on both synthetic and real-world datasets, we show that our framework outperforms baseline methods in both forecasting and causal discovery, underscoring the benefits of this integrated approach.
Primary Area: causal reasoning
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: 12796
Loading