Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
We propose an adaptive conformal prediction framework for regression tasks with Grassmann-valued outputs, constructing prediction regions as metric balls on the manifold using basis-invariant geometric distances. By scaling non-conformity scores with a learned prediction difficulty estimator, the method achieves locally adaptive, input-dependent regions that preserve finite-sample marginal coverage guarantees. Evaluation on synthetic and real-world inverse design problems demonstrates that geometry-aware scores and local adaptation improve coverage uniformity and reduce prediction region size in heteroscedastic settings.
Geometry-aware conformal prediction on the Grassmannian achieves locally adaptive, manifold-respecting prediction regions that significantly reduce size in heteroscedastic settings while guaranteeing coverage.
We introduce a conformal prediction framework for regression with Grassmann-valued responses. The proposed method constructs prediction regions in the form of metric balls directly on the Grassmann manifold, using basis-invariant geometric distances as non-conformity measures and scaling the corresponding non-conformity scores through a learned prediction difficulty estimator. This yields locally adaptive, input-dependent prediction regions that respect the underlying manifold geometry, while retaining the finite-sample marginal coverage guarantee of split conformal prediction under exchangeability. We evaluate several Grassmannian distance metrics and compare their prediction-region efficiency on synthetic benchmarks and an airfoil inverse shape-design problem. The results highlight the advantages of geometry-aware non-conformity scores, while local adaptation improves empirical coverage uniformity across prediction difficulty levels and, in strongly heteroscedastic settings, can substantially reduce prediction-region size while maintaining the target coverage.