Attraction-Ranking Mismatch in Out-of-Distribution Detection: Theory and Evidence from Hamiltonian Feature Dynamics
Abstract
Attracting features toward learned in-distribution (ID) regions is an intuitive strategy for out-of-distribution (OOD) detection, but does such attraction improve detection? We investigate this question using class-conditioned Hamiltonian dynamics as a controlled testbed and identify an attraction-ranking mismatch: improving individual class-affinity scores need not improve the relative ordering of ID and OOD samples. Our analysis separates misalignment between the driving field and the detection score from ranking failure that persists even when they are aligned. An exact AUROC decomposition and a local boundary analysis identify relative score changes at the ID-OOD ranking boundary as the source of first-order AUROC change under regularity conditions. Experiments demonstrate that a score-aligned rule can improve performance on a held-out source validation split yet consistently degrade it after transfer, even without a single winning-class change. A separate perturbation study finds that near-boundary relative score changes track the direction of small AUROC changes in the tested settings. Finally, centroid approximation and shared scales connect the zero-step score to a static centroid family that includes the existing CTM detector. Together, these results explain why class attraction is insufficient to justify feature refinement and motivate evaluating its effect on the ranking boundary.
Related
- OUT-OF-DISTRIBUTION AI-GENERATED IMAGE DETECTION WITH ZERO-PHASE DECONVOLUTION FOR RESIDUAL WHITENING
- Anytime-Valid Distribution Shift Detection via Predictive Rank Martingales
- High-precision fault inception time detection in active distribution networks using Euclidean distance and Gaussian filtering
- The Singularity Has a Distribution: Exact Timing Laws and Detection Limits for a Stochastic Intelligence Explosion
- Investigating the Distribution: Signal, Noise, and the Resolution Toolkit in Evidence‐Based Medicine
BibTeX
@misc{wang2026attractionranking,
title = {{Attraction-Ranking Mismatch in Out-of-Distribution Detection: Theory and Evidence from Hamiltonian Feature Dynamics}},
author = {Zilin Wang and Lianghai Xiao},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.06021},
eprint = {2611.06021},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.06021},
}
RIS
TY - UNPB
TI - Attraction-Ranking Mismatch in Out-of-Distribution Detection: Theory and Evidence from Hamiltonian Feature Dynamics
AU - Zilin Wang
AU - Lianghai Xiao
PY - 2026
DA - 2026/11/04
PB - arXiv
DO - 10.48550/arxiv.2611.06021
UR - https://arxiv.org/abs/2611.06021
AB - Attracting features toward learned in-distribution (ID) regions is an intuitive strategy for out-of-distribution (OOD) detection, but does such attraction improve detection? We investigate this question using class-conditioned Hamiltonian dynamics as a controlled testbed and identify an attraction-ranking mismatch: improving individual class-affinity scores need not improve the relative ordering of ID and OOD samples. Our analysis separates misalignment between the driving field and the detection score from ranking failure that persists even when they are aligned. An exact AUROC decomposition and a local boundary analysis identify relative score changes at the ID-OOD ranking boundary as the source of first-order AUROC change under regularity conditions. Experiments demonstrate that a score-aligned rule can improve performance on a held-out source validation split yet consistently degrade it after transfer, even without a single winning-class change. A separate perturbation study finds that near-boundary relative score changes track the direction of small AUROC changes in the tested settings. Finally, centroid approximation and shared scales connect the zero-step score to a static centroid family that includes the existing CTM detector. Together, these results explain why class attraction is insufficient to justify feature refinement and motivate evaluating its effect on the ranking boundary.
ER -