Test-Time Learning of Causal Structure from Interventional Data

Wei Chen, Rui Ding, Huang Bojun, Yang Zhang, Qiang Fu, Yuxuan Liang, Shi Han, Dongmei Zhang
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:13980-14018, 2026.

Abstract

Supervised Causal Learning has shown promise in causal discovery, yet it often struggles with generalization across diverse interventional settings, particularly when intervention targets are unknown. To address this, we propose TICL (Test-time Interventional Causal Learning), a novel method that synergizes Test-Time Training with Joint Causal Inference (JCI). Specifically, we design a self-augmentation strategy to generate instance-specific training data at test time, effectively avoiding distribution shifts. Furthermore, by integrating JCI, we developed a PC-inspired two-phase supervised learning scheme, which effectively leverages self-augmented data while ensuring theoretical identifiability. Extensive experiments on bnlearn benchmarks demonstrate TICL’s superiority in multiple aspects of causal discovery and intervention target detection.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-chen26t, title = {Test-Time Learning of Causal Structure from Interventional Data}, author = {Chen, Wei and Ding, Rui and Bojun, Huang and Zhang, Yang and Fu, Qiang and Liang, Yuxuan and Han, Shi and Zhang, Dongmei}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {13980--14018}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/chen26t/chen26t.pdf}, url = {https://proceedings.mlr.press/v306/chen26t.html}, abstract = {Supervised Causal Learning has shown promise in causal discovery, yet it often struggles with generalization across diverse interventional settings, particularly when intervention targets are unknown. To address this, we propose TICL (Test-time Interventional Causal Learning), a novel method that synergizes Test-Time Training with Joint Causal Inference (JCI). Specifically, we design a self-augmentation strategy to generate instance-specific training data at test time, effectively avoiding distribution shifts. Furthermore, by integrating JCI, we developed a PC-inspired two-phase supervised learning scheme, which effectively leverages self-augmented data while ensuring theoretical identifiability. Extensive experiments on bnlearn benchmarks demonstrate TICL’s superiority in multiple aspects of causal discovery and intervention target detection.} }
Endnote
%0 Conference Paper %T Test-Time Learning of Causal Structure from Interventional Data %A Wei Chen %A Rui Ding %A Huang Bojun %A Yang Zhang %A Qiang Fu %A Yuxuan Liang %A Shi Han %A Dongmei Zhang %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-chen26t %I PMLR %P 13980--14018 %U https://proceedings.mlr.press/v306/chen26t.html %V 306 %X Supervised Causal Learning has shown promise in causal discovery, yet it often struggles with generalization across diverse interventional settings, particularly when intervention targets are unknown. To address this, we propose TICL (Test-time Interventional Causal Learning), a novel method that synergizes Test-Time Training with Joint Causal Inference (JCI). Specifically, we design a self-augmentation strategy to generate instance-specific training data at test time, effectively avoiding distribution shifts. Furthermore, by integrating JCI, we developed a PC-inspired two-phase supervised learning scheme, which effectively leverages self-augmented data while ensuring theoretical identifiability. Extensive experiments on bnlearn benchmarks demonstrate TICL’s superiority in multiple aspects of causal discovery and intervention target detection.
APA
Chen, W., Ding, R., Bojun, H., Zhang, Y., Fu, Q., Liang, Y., Han, S. & Zhang, D.. (2026). Test-Time Learning of Causal Structure from Interventional Data. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:13980-14018 Available from https://proceedings.mlr.press/v306/chen26t.html.

Related Material