Citation

If you use CausalFM Toolkit in your research, please cite our paper:

BibTeX

@article{ma2025causalfm,
  title={Foundation Models for Causal Inference via Prior-Data Fitted Networks},
  author={Ma, Yuchen and Frauen, Dennis and Javurek, Emil and Feuerriegel, Stefan},
  journal={arXiv preprint arXiv:2506.10914},
  year={2025}
}

Paper Information

Title: Foundation Models for Causal Inference via Prior-Data Fitted Networks

Authors: Yuchen Ma, Dennis Frauen, Emil Javurek, Stefan Feuerriegel

Year: 2025

arXiv: https://arxiv.org/abs/2506.10914

Abstract:

We introduce CausalFM, a comprehensive framework for training foundation models for causal inference using Prior-Data Fitted Networks (PFNs). Unlike traditional approaches that require training on individual datasets, CausalFM learns from distributions of synthetic datasets, enabling zero-shot transfer to new datasets without fine-tuning. We develop foundation models for multiple causal inference settings, including standard CATE estimation, instrumental variables, and front-door adjustment. Our experimental results demonstrate that CausalFM achieves state-of-the-art performance across diverse causal inference tasks while providing calibrated uncertainty quantification through Gaussian Mixture Model prediction heads.

Acknowledgments

This work was supported by [funding information if applicable].

We thank the contributors and users of the CausalFM Toolkit for their valuable feedback and contributions.

Related Work

CausalFM builds upon several key ideas from the literature:

Prior-Data Fitted Networks (PFNs):

• Müller, S., Hollmann, N., Arango, S. P., Grabocka, J., & Hutter, F. (2022). Transformers can do bayesian inference. ICLR.

Causal Inference:

• Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of Educational Psychology.

• Pearl, J. (2009). Causality: Models, Reasoning, and Inference (2nd ed.). Cambridge University Press.

Heterogeneous Treatment Effects:

• Wager, S., & Athey, S. (2018). Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association.

• Künzel, S. R., Sekhon, J. S., Bickel, P. J., & Yu, B. (2019). Metalearners for estimating heterogeneous treatment effects using machine learning. Proceedings of the National Academy of Sciences.

License

CausalFM Toolkit is released under the Apache License 2.0. See the LICENSE file for more details.

Contact

For questions, issues, or contributions, please:

• Open an issue on GitHub: https://github.com/yccm/CausalFM-toolkit

• Contact the authors via email: [contact information]

Contributing

We welcome contributions to CausalFM Toolkit! Please see our contributing guidelines on GitHub for more information on how to contribute code, documentation, or bug reports.

Ways to Contribute:

• Report bugs and issues

• Suggest new features

• Improve documentation

• Submit pull requests with bug fixes or enhancements

• Share your use cases and applications

Community

Join our community to stay updated:

• GitHub: https://github.com/yccm/CausalFM-toolkit

• Documentation: https://causalfm.readthedocs.io

• Paper: https://arxiv.org/abs/2506.10914

We appreciate any feedback and contributions from the community!