Conformal Policy Control

Drew Prinster, Clara Fannjiang, Ji Won Park, Kyunghyun Cho, Anqi Liu, Suchi Saria, Samuel Stanton

July, 2026 Spotlight

@inproceedings{prinster2026conformal,
  title = {Conformal Policy Control},
  author = {Prinster, Drew and Fannjiang, Clara and Park, Ji Won and Cho, Kyunghyun and Liu, Anqi and Saria, Suchi and Stanton, Samuel Don},
  booktitle = {Forty-third International Conference on Machine Learning},
  year = {2026},
  url = {https://arxiv.org/abs/2603.02196}
}

Citation

Prinster, D., Fannjiang, C., Park, J. W., Cho, K., Liu, A., Saria, S., & Stanton, S. (2026). Conformal Policy Control. International Conference on Machine Learning (ICML). arXiv preprint arXiv:2603.02196.

Informal Summary

Proposes using any safe reference policy as a probabilistic regulator for an optimized but untested policy, using conformal calibration to determine how aggressively the new policy may act while respecting a declared risk tolerance. Unlike conservative optimization approaches, this yields finite-sample guarantees for non-monotonic bounded loss functions without requiring correct model specification or hyperparameter tuning, with experiments spanning natural language and biomolecular applications.