TASTE: Task-Aware Out-of-Distribution Detection via Stein Operators

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kozyra, Michał, Reinert, Gesine
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914314185605120
author Kozyra, Michał
Reinert, Gesine
author_facet Kozyra, Michał
Reinert, Gesine
contents Out-of-distribution detection methods are often either data-centric, detecting deviations from the training input distribution irrespective of their effect on a trained model, or model-centric, relying on classifier outputs without explicit reference to data geometry. We propose TASTE (Task-Aware STEin operators): a task-aware framework based on so-called Stein operators, which allows us to link distribution shift to the input sensitivity of the model. We show that the resulting operator admits a clear geometric interpretation as a projection of distribution shift onto the sensitivity field of the model, yielding theoretical guarantees. Beyond detecting the presence of a shift, the same construction enables its localisation through a coordinate-wise decomposition, and for image data-provides interpretable per-pixel diagnostics. Experiments on controlled Gaussian shifts, MNIST under geometric perturbations, and CIFAR-10 perturbed benchmarks demonstrate that the proposed method aligns closely with task degradation while outperforming established baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2602_07640
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TASTE: Task-Aware Out-of-Distribution Detection via Stein Operators
Kozyra, Michał
Reinert, Gesine
Machine Learning
Out-of-distribution detection methods are often either data-centric, detecting deviations from the training input distribution irrespective of their effect on a trained model, or model-centric, relying on classifier outputs without explicit reference to data geometry. We propose TASTE (Task-Aware STEin operators): a task-aware framework based on so-called Stein operators, which allows us to link distribution shift to the input sensitivity of the model. We show that the resulting operator admits a clear geometric interpretation as a projection of distribution shift onto the sensitivity field of the model, yielding theoretical guarantees. Beyond detecting the presence of a shift, the same construction enables its localisation through a coordinate-wise decomposition, and for image data-provides interpretable per-pixel diagnostics. Experiments on controlled Gaussian shifts, MNIST under geometric perturbations, and CIFAR-10 perturbed benchmarks demonstrate that the proposed method aligns closely with task degradation while outperforming established baselines.
title TASTE: Task-Aware Out-of-Distribution Detection via Stein Operators
topic Machine Learning
url https://arxiv.org/abs/2602.07640