![[2511.21537] Context-Specific Causal Graph Discovery with Unobserved Contexts: Non-Stationarity, Regimes and Spatio-Temporal Patterns](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2511.21537] Context-Specific Causal Graph Discovery with Unobserved Contexts: Non-Stationarity, Regimes and Spatio-Temporal Patterns
Summary
This paper presents a framework for context-specific causal graph discovery that addresses non-stationarity and spatio-temporal patterns, enhancing causal reasoning in complex systems.
Why It Matters
Understanding causal relationships in data with varying contexts is crucial for accurate modeling, especially in fields like climate science. This research provides a modular approach to tackle challenges in causal inference, making it relevant for researchers and practitioners dealing with dynamic systems.
Key Takeaways
- Introduces a framework for causal graph discovery that accounts for unobserved contexts.
- Addresses challenges of non-stationarity and statistical convergence in causal inference.
- Modifies existing constraint-based methods to improve independence testing.
- Facilitates easier analysis by breaking down complex problems into manageable subproblems.
- Demonstrates applicability through numerical experiments and existing causal discovery methods.
Computer Science > Machine Learning arXiv:2511.21537 (cs) [Submitted on 26 Nov 2025 (v1), last revised 13 Feb 2026 (this version, v2)] Title:Context-Specific Causal Graph Discovery with Unobserved Contexts: Non-Stationarity, Regimes and Spatio-Temporal Patterns Authors:Martin Rabel, Jakob Runge View a PDF of the paper titled Context-Specific Causal Graph Discovery with Unobserved Contexts: Non-Stationarity, Regimes and Spatio-Temporal Patterns, by Martin Rabel and 1 other authors View PDF Abstract:Real-world problems, for example in climate applications, often require causal reasoning on spatially gridded time series data or data with comparable structure. While the underlying system is often believed to behave similarly at different Points in space and time, those variations that do exist are relevant twofold: They often encode important information in and of themselves. And they may negatively affect the stability and validity of results if not accounted for. We study the information encoded in changes of the causal graph, with stability in mind. Two core challenges arise, related to the complexity of encoding system-states and to statistical convergence properties in the presence of imperfectly recoverable non-stationary structure. We provide a framework realizing principles conceptually suitable to overcome these challenges - an interpretation supported by numerical experiments. Primarily, we modify constraint-based causal discovery approaches on the level of independe...
