Multivariate prediction sets improve by focusing on residual tails
Tail-Aware Geometry Learning for Conformal Ellipsoids
Machine Learning
Summary
Making reliable predictions about many things at once is hard, especially when you want to quantify uncertainty without assumptions. The authors studied how shapes used for predictions can better handle extreme, unusual errors by learning from the severity of those errors instead of just their frequency. They show a way to create prediction sets shaped like ellipsoids that prioritize bigger errors when shaping their size and form while still guaranteeing how often they cover true results. Their experiments suggest this new method works better than older ways that ignore extreme errors' severity.
What this means in practice
- •For risk management teams: Generate multivariate uncertainty sets that better capture rare, severe deviations to improve risk assessment in finance or insurance portfolios.
- •For autonomous vehicle engineers: Create more reliable uncertainty bounds around sensor data by shaping prediction areas sensitive to extreme residual errors, enhancing safety decisions.
Authors
Xiang Zhang
Abstract
This paper studies multivariate conformal prediction (CP), a distribution-free uncertainty quantification framework with finite-sample coverage guarantees. The efficiency of multivariate prediction sets hinges critically on the residual geometry encoded by the nonconformity score, while existing minimum-volume methods rely on quantile thresholds that ignore tail residual severity and implicitly bind geometry learning to coverage level. We propose a tail-aware geometry learning framework for conformal ellipsoids that decouples tail sensitivity in geometry learning from the final coverage guarantee. Using a two-split design, we learn the metric matrix via volume minimization under a CVaR constraint on an estimation split, then apply standard conformal calibration on a held-out calibration split. The resulting problem is convex and admits a bounded-reweighting interpretation that prioritizes high-residual samples. Moreover, we theoretically characterize the trade-off between ellipsoidal volume and tail severity. Experimental results demonstrate the effectiveness of the proposed method.