Abstract:Time series generation (TSG) synthesizes realistic sequences and has achieved remarkable success. Among TSG, conditional models generate sequences given observed covariates, however, such models learn observational correlations without considering unobserved confounding. In this work, we propose a causal perspective on conditional TSG and introduce causal time series generation as a new TSG task family, formalized within Pearl's causal ladder, extending beyond observational generation to include interventional and counterfactual settings. To instantiate these tasks, we develop CaTSG, a unified diffusion-based framework with backdoor-adjusted guidance that causally steers sampling toward desired interventions and individual counterfactuals while preserving observational fidelity. Specifically, our method derives causal score functions via backdoor adjustment and the abduction-action-prediction procedure, thus enabling principled support for all three levels of TSG. Extensive experiments on both synthetic and real-world datasets show that CaTSG achieves superior fidelity and also supporting interventional and counterfactual generation that existing baselines cannot handle. Overall, we propose the causal TSG family and instantiate it with CaTSG, providing an initial proof-of-concept and opening a promising direction toward more reliable simulation under interventions and counterfactual generation.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2509.20846 [cs.LG] |
| (or arXiv:2509.20846v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2509.20846 arXiv-issued DOI via DataCite |
Submission history
From: Yutong Xia [view email]
[v1]
Thu, 25 Sep 2025 07:34:46 UTC (3,145 KB)
[v2]
Mon, 19 Jan 2026 07:53:59 UTC (3,153 KB)
[v3]
Thu, 14 May 2026 08:13:14 UTC (3,153 KB)