Divide et Calibra: Multiclass Local Calibration via Vector Quantization

2026-05-20

AI Summary

This work addresses the challenge of achieving both high accuracy and reliable calibration in multi-class models deployed in high-stakes scenarios. While existing global calibration methods assume uniform error distributions and local approaches often suffer from information loss due to dimensionality reduction, this paper proposes a vector quantization–based composite local calibration framework. By structuring the latent space into regions and learning region-specific calibration mappings, the method preserves representational fidelity while enabling heterogeneous calibration. A codeword-index–driven parameterization of Dirichlet concentration parameters facilitates cross-region parameter sharing without compromising regional specificity. Experiments across multiple benchmark datasets demonstrate that the proposed approach significantly improves local calibration performance while maintaining strong global calibration and predictive accuracy.

Abstract

Accurate and well-calibrated Machine Learning (ML) models are mandatory in high-stakes settings, yet effective multiclass calibration remains challenging: global approaches assume calibration errors are homogeneous across the latent space, while local methods often rely on latent-space dimensionality reduction, which leads to information loss. To address these issues, we propose a compositional approach to multiclass calibration, where region-specific calibration maps are constructed from shared codeword-dependent factors. We instantiate this idea via Vector Quantization (VQ), which induces a structured partition of the representation space, and an indexed parameterization of Dirichlet concentrations that enables parameter sharing across regions. Our approach learns heterogeneous calibration maps that generalize well even to sparse regions of the latent space. Experiments on benchmark datasets show significant improvements in local calibration while maintaining competitive global calibration and predictive performance.

Problem

• multiclass calibration
• local calibration
• calibration error
• latent space
• machine learning

Innovation

• Vector Quantization
• Multiclass Calibration
• Local Calibration
• Dirichlet Calibration
• Representation Space Partitioning