{"id":"arxiv/2611.06021","url":"https://arxiv.org/abs/2611.06021","tags":["cs.LG","math.OC"],"score":0,"title":"Attraction-Ranking Mismatch in Out-of-Distribution Detection: Theory and Evidence from Hamiltonian Feature Dynamics","domain":"cs.LG","source":"arxiv","authors":["Zilin Wang","Lianghai Xiao"],"reviews":0,"subject":"cs.LG","summary":"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…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-11-04","reactions":{}}