Leave-One-Out Stable Conformal Prediction

This paper proposes LOO-StabCP, a conformal prediction method that uses leave-one-out algorithmic stability to accelerate full conformal prediction while preserving statistical validity. The work connects uncertainty quantification, stability theory, and scalable prediction.

Recommended citation: Lee, K. and Zhang, Y. (2025). "Leave-One-Out Stable Conformal Prediction." The Thirteenth International Conference on Learning Representations (ICLR).
Download Paper