Self-Organized Conformal Prediction: Reducing Regional Coverage Gaps with Unsupervised Group Discovery

2026-06-28

AI Summary

Traditional conformal prediction often suffers from insufficient local coverage for safety-critical subpopulations in heterogeneous regions due to its reliance on global average calibration. This work proposes Self-Organizing Conformal Prediction (SOCP), which leverages unsupervised self-organizing maps (SOMs) to uncover local structures in the input space. SOCP constructs localized calibration buffers around the best-matching unit of a query point and its neighborhood, thereby achieving region-adaptive coverage guarantees without requiring supervised partitioning or model retraining. Theoretical analysis provides finite-sample validity guarantees for fixed groups and approximate validity for central units. Evaluated across eight benchmark datasets, SOCP significantly reduces the weighted regional coverage gap in seven of them—by an average of 7.1%—while increasing prediction set size by only 6.2%, with negligible computational overhead.

Abstract

Conformal prediction guarantees marginal coverage, but pooled calibration averages over heterogeneous regions and can mask regional undercoverage in safety-critical subgroups. We introduce Self-Organized Conformal Prediction (SOCP), a calibration scheme that discovers input-space groups with a Self-Organizing Map (SOM) and, at test time, draws a local calibration buffer from the query's best-matching unit (BMU) cell or a fixed grid neighborhood. The same retrieval rule applies to regression and classification tasks across tabular features and image embeddings, leaving the predictor and nonconformity score untouched. SOCP gives exact validity for BMU-cell retrieval and fixed retrieved-set validity for neighborhood buffers; central-cell validity for neighborhood retrieval holds up to a Kolmogorov-Smirnov (KS) bias term. A split-routed extension recovers fixed retrieved-set validity conditional on the routing split. On eight regression and classification benchmarks, SO-SCP reduces the weighted regional coverage gap on $7/8$ datasets (mean paired change $-7.1\%$) for a mean prediction-set size increase of $6.2\%$, with negligible overhead on the largest six datasets; SO-CQR yields smaller gains, since quantile regression already absorbs much of the heterogeneity. By learning groups directly from the input geometry, SOCP provides group-local calibration with exact fixed-group guarantees and approximate central-cell guarantees, without supervised partitions or predictor retraining.

Problem

• Conformal Prediction
• Regional Coverage Gap
• Group Calibration
• Heterogeneous Regions
• Safety-Critical Subgroups

Innovation

• Self-Organized Conformal Prediction
• Self-Organizing Map
• Local Calibration
• Regional Coverage Gap
• Unsupervised Group Discovery