arXiv:2507.10425v2 Announce Type: replace 
Abstract: Conformal prediction is a distribution-free uncertainty quantification method that has gained popularity in the machine learning community due to its finite-sample guarantees and ease of use. Its most common variant, dubbed split conformal prediction, is also computationally efficient as it boils down to collecting statistics of the model predictions on some calibration data not yet seen by the model. Nonetheless, these guarantees only hold if the calibration and test data are exchangeable, a condition that is difficult to verify and often violated in practice due to so-called distribution shifts. The literature is rife with methods to mitigate the loss in coverage in this non-exchangeable setting, but these methods require some prior information on the type of distribution shift to be expected at test time. In this work, we study this problem via a new perspective, through the lens of optimal transport, and show that it is possible to estimate the loss in coverage and mitigate arbitrary distribution shifts, offering a principled and broadly applicable solution.

تقدم دراسة حديثة نهجًا جديدًا للتنبؤ المتوافق يستخدم النقل الأمثل لمعالجة تغييرات التوزيع في التعلم الآلي، خاصة عند التعامل مع بيانات غير مصنفة. تعزز هذه الطريقة موثوقية قياس عدم اليقين، وهو أمر حاسم لتطوير نماذج تعلم آلي قوية. مع استمرار تطور هذا المجال، فإن مثل هذه التقدمات مهمة لأنها تعد بتحسين أداء النماذج وقابليتها للتطبيق في السيناريوهات الواقعية.

Un estudio reciente presenta un enfoque novedoso para la predicción conforme que utiliza el transporte óptimo para abordar los cambios de distribución en el aprendizaje automático, especialmente al tratar con datos no etiquetados. Este método mejora la fiabilidad de la cuantificación de la incertidumbre, que es crucial para desarrollar modelos de aprendizaje automático robustos. A medida que el campo continúa evolucionando, tales avances son significativos ya que prometen mejorar el rendimiento de los modelos y su aplicabilidad en escenarios del mundo real.

Une étude récente présente une nouvelle approche de la prédiction conforme qui utilise le transport optimal pour traiter les changements de distribution en apprentissage automatique, en particulier lorsqu'il s'agit de données non étiquetées. Cette méthode améliore la fiabilité de la quantification de l'incertitude, ce qui est crucial pour le développement de modèles d'apprentissage automatique robustes. Alors que le domaine continue d'évoluer, de tels progrès sont significatifs car ils promettent d'améliorer les performances des modèles et leur applicabilité dans des scénarios réels.

A recent study introduces a novel approach to conformal prediction that utilizes optimal transport to address distribution shifts in machine learning, particularly when dealing with unlabeled data. This method enhances the reliability of uncertainty quantification, which is crucial for developing robust machine learning models. As the field continues to evolve, such advancements are significant as they promise to improve model performance and applicability in real-world scenarios.

Non-exchangeable Conformal Prediction with Optimal Transport: Tackling Distribution Shifts with Unlabeled Data

arXiv:2511.16845v1 Announce Type: new 
Abstract: Ordinal classification has been widely applied in many high-stakes applications, e.g., medical imaging and diagnosis, where reliable uncertainty quantification (UQ) is essential for decision making. Conformal prediction (CP) is a general UQ framework that provides statistically valid guarantees, which is especially useful in practice. However, prior ordinal CP methods mainly focus on heuristic algorithms or restrictively require the underlying model to predict a unimodal distribution over ordinal labels. Consequently, they provide limited insight into coverage-efficiency trade-offs, or a model-agnostic and distribution-free nature favored by CP methods. To this end, we fill this gap by propose an ordinal-CP method that is model-agnostic and provides instance-level optimal prediction intervals. Specifically, we formulate conformal ordinal classification as a minimum-length covering problem at the instance level. To solve this problem, we develop a sliding-window algorithm that is optimal on each calibration data, with only a linear time complexity in K, the number of label candidates. The local optimality per instance further also improves predictive efficiency in expectation. Moreover, we propose a length-regularized variant that shrinks prediction set size while preserving coverage. Experiments on four benchmark datasets from diverse domains are conducted to demonstrate the significantly improved predictive efficiency of the proposed methods over baselines (by 15% decrease on average over four datasets).

تم اقتراح طريقة جديدة للتصنيف الترتيبي تركز على التنبؤ المتوافق (CP) لتحسين قياس عدم اليقين في التطبيقات الحساسة مثل التصوير الطبي والتشخيص. تهدف هذه الطريقة المستقلة عن النموذج إلى توفير فترات تنبؤ مثالية على مستوى الحالة، مما يعالج القيود التي كانت تعاني منها طرق CP الترتيبية السابقة التي اعتمدت على خوارزميات تجريبية أو توزيعات أحادية الشكل.

Se ha propuesto un nuevo método para la clasificación ordinal, centrado en la predicción conforme (CP) para mejorar la cuantificación de la incertidumbre en aplicaciones críticas como la imagenología médica y el diagnóstico. Este enfoque, independiente del modelo, tiene como objetivo proporcionar intervalos de predicción óptimos a nivel de instancia, abordando las limitaciones de los métodos CP ordinales anteriores que dependían de algoritmos heurísticos o distribuciones unimodales.

Une nouvelle méthode de classification ordinale a été proposée, axée sur la prédiction conforme (CP) pour améliorer la quantification de l'incertitude dans des applications critiques telles que l'imagerie médicale et le diagnostic. Cette approche, indépendante du modèle, vise à fournir des intervalles de prédiction optimaux au niveau des instances, répondant aux limitations des méthodes CP ordinales précédentes qui reposaient sur des algorithmes heuristiques ou des distributions unimodales.

A new method for ordinal classification has been proposed, focusing on conformal prediction (CP) to enhance uncertainty quantification in high-stakes applications like medical imaging and diagnosis. This model-agnostic approach aims to provide optimal prediction intervals at the instance level, addressing limitations of previous ordinal CP methods that relied on heuristic algorithms or unimodal distributions.

Non-exchangeable Conformal Prediction with Optimal Transport: Tackling Distribution Shifts with Unlabeled Data

Was this article worth reading? Share it

Mapfit

Octofy

Keywords AI