From Individual Calibration to Reliable Classifiers: ALD Parameterization with mPAIC Guarantees

Deming Sheng, Ricardo Henao
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:111559-111602, 2026.

Abstract

Modern neural classifiers can achieve remarkable predictive performance, yet often suffer from miscalibration. In this paper, we introduce a unified calibration framework applicable to arbitrary distribution-based classifiers. The proposed calibration objective guarantees a monotone Probably Approximately Individually Calibrated (mPAIC) predictor, which theoretically implies the properties of a Probably Approximately Calibrated Classifier (PACC) with explicit error bounds. To enable stable and effective optimization, we further devise a Decoupled Dual-Stream Optimization (DDSO) strategy with gradient detachment to reconcile discriminative representation learning and continuous calibration. Notably, our framework bridges calibration paradigms, supporting flexible deployment either as an end-to-end pre-calibration objective or as a lightweight post-calibration adapter. Extensive experiments across nine real-world datasets demonstrate that our approach consistently outperforms strong baselines, achieving superior performance on both accuracy and multi-level calibration.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-sheng26a, title = {From Individual Calibration to Reliable Classifiers: {ALD} Parameterization with m{PAIC} Guarantees}, author = {Sheng, Deming and Henao, Ricardo}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {111559--111602}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/sheng26a/sheng26a.pdf}, url = {https://proceedings.mlr.press/v306/sheng26a.html}, abstract = {Modern neural classifiers can achieve remarkable predictive performance, yet often suffer from miscalibration. In this paper, we introduce a unified calibration framework applicable to arbitrary distribution-based classifiers. The proposed calibration objective guarantees a monotone Probably Approximately Individually Calibrated (mPAIC) predictor, which theoretically implies the properties of a Probably Approximately Calibrated Classifier (PACC) with explicit error bounds. To enable stable and effective optimization, we further devise a Decoupled Dual-Stream Optimization (DDSO) strategy with gradient detachment to reconcile discriminative representation learning and continuous calibration. Notably, our framework bridges calibration paradigms, supporting flexible deployment either as an end-to-end pre-calibration objective or as a lightweight post-calibration adapter. Extensive experiments across nine real-world datasets demonstrate that our approach consistently outperforms strong baselines, achieving superior performance on both accuracy and multi-level calibration.} }
Endnote
%0 Conference Paper %T From Individual Calibration to Reliable Classifiers: ALD Parameterization with mPAIC Guarantees %A Deming Sheng %A Ricardo Henao %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-sheng26a %I PMLR %P 111559--111602 %U https://proceedings.mlr.press/v306/sheng26a.html %V 306 %X Modern neural classifiers can achieve remarkable predictive performance, yet often suffer from miscalibration. In this paper, we introduce a unified calibration framework applicable to arbitrary distribution-based classifiers. The proposed calibration objective guarantees a monotone Probably Approximately Individually Calibrated (mPAIC) predictor, which theoretically implies the properties of a Probably Approximately Calibrated Classifier (PACC) with explicit error bounds. To enable stable and effective optimization, we further devise a Decoupled Dual-Stream Optimization (DDSO) strategy with gradient detachment to reconcile discriminative representation learning and continuous calibration. Notably, our framework bridges calibration paradigms, supporting flexible deployment either as an end-to-end pre-calibration objective or as a lightweight post-calibration adapter. Extensive experiments across nine real-world datasets demonstrate that our approach consistently outperforms strong baselines, achieving superior performance on both accuracy and multi-level calibration.
APA
Sheng, D. & Henao, R.. (2026). From Individual Calibration to Reliable Classifiers: ALD Parameterization with mPAIC Guarantees. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:111559-111602 Available from https://proceedings.mlr.press/v306/sheng26a.html.

Related Material