Mechanisms Under Shifts: Interpretable Clustering With Self-Improving Heterogeneous Causal Graphs
IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026-08 · 48 (8) · 10083-10096
DOI: 10.1109/tpami.2026.3683072 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026-08. Verfasst von Wenrui Li und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 72 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Beijing Municipal Natural Science Foundation.
Abstract (Originalsprache)
Understanding causal heterogeneity is crucial for building robust and interpretable learning systems that operate reliably under environmental shifts. However, existing methods lack causal awareness, with insufficient modeling of heterogeneity, confounding, and observational constraints, leading to poor interpretability and difficulty distinguishing true causal heterogeneity from spurious associations. We propose an unsupervised framework, HCL (Interpretable Causal Mechanism-Aware Clustering with Self-Improving Adaptive Heterogeneous Causal Structure Learning), that jointly infers latent clusters and their associated causal structures from mixed-type observational data without requiring temporal ordering, environment labels, interventions or other prior knowledge. HCL relaxes the homogeneity and sufficiency assumptions by introducing an equivalent representation that encodes both structural heterogeneity and confounding. It further develops a bi-directional iterative strategy to alternately refine causal clustering and structure learning, along with a self-supervised regularization that balances cross-cluster universal and specific mechanisms under shifts. Together, these components enable convergence toward interpretable, heterogeneous causal patterns. Theoretically, we show identifiability of heterogeneous causal structures under mild conditions. Empirically, HCL achieves superior performance in both clustering and structure learning tasks, and recovers biologically meaningful mechanisms in real-world single-cell perturbation and clinical intervention data, demonstrating its utility for discovering interpretable, mechanism-level causal heterogeneity.
Förderung
- National Natural Science Foundation of China
- Beijing Municipal Natural Science Foundation
Zitieren
APA
Li, W., Zhang, Q., & Wang, X. (2026). Mechanisms Under Shifts: Interpretable Clustering With Self-Improving Heterogeneous Causal Graphs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 48(8), 10083-10096. https://doi.org/10.1109/tpami.2026.3683072
Vancouver
Li W, Zhang Q, Wang X. Mechanisms Under Shifts: Interpretable Clustering With Self-Improving Heterogeneous Causal Graphs. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2026;48(8):10083-10096. doi:10.1109/tpami.2026.3683072
BibTeX
@article{Li2026,
author = {Li, Wenrui and Zhang, Qinghao and Wang, Xiaowo},
title = {Mechanisms Under Shifts: Interpretable Clustering With Self-Improving Heterogeneous Causal Graphs},
journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
year = {2026},
volume = {48},
number = {8},
pages = {10083--10096},
doi = {10.1109/tpami.2026.3683072}
}
EUVIMED