Finite-Sample Conformal Coverage Recovery via Fusion under Degraded Local Guarantees in Occupancy Map Estimation
arXiv:2607.14906
2026
Regularization
1 ideas extracted · analyzed Aug 30, 2026
What the math gives to ML
The paper offers a constructive distributed uncertainty-fusion mechanism: locally unreliable or degraded conformal predictors exchange scalar e-values, and a receiver combines them under a neighborhood miscoverage budget to recover finite-sample coverage. The transferable asset is validity-preserving e-value fusion with uncertainty attenuation and abstention when evidence is insufficient. A neural-network implementation can attach conformal e-value heads to independently calibrated models, fuse their predictions over an ensemble or communication graph, and return a set-valued prediction instead of a forced point prediction. The key falsifiable signatures are the Markov threshold 1/alpha, empirical coverage at least 1 minus alpha, and graph-density-dependent reduction in abstention.
Ideas from this paper
△ Mechanism confirmed, baseline not beaten
2026
Equip each neural-network expert or robot with a locally calibrated e-value for every candidate label, then fuse neighboring e-values using uncertainty-attenuated convex weights. At inference time, retain all labels whose fused e-value does not cross the finite-sample rejection threshold, so the model abstains instead of making an unsupported point prediction. This transfers the paper's coverage-recovery mechanism to ensembles, federated models, and graph neural networks.
Useful7/10
Difficulty5/10
Novelty5/10