Introduces an information-theoretic formalization of the binding problem and a probing method to quantify binding information in deep learning model representations, tested on ViTs across challenging datasets.
Are we done with object-centric learning?arXiv preprint arXiv:2504.07092
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
fields
cs.CV 2years
2026 2representative citing papers
Under category–scene correlation shift, segmentation models often preserve foreground extent but swap confusable class identities; Flip, FG-Corr/Flip/Miss, and entropy flip-risk make that failure measurable and monitorable.
citing papers explorer
-
Formalizing the Binding Problem
Introduces an information-theoretic formalization of the binding problem and a probing method to quantify binding information in deep learning model representations, tested on ViTs across challenging datasets.
-
Right Regions, Wrong Labels: Semantic Label Flips in Segmentation under Correlation Shift
Under category–scene correlation shift, segmentation models often preserve foreground extent but swap confusable class identities; Flip, FG-Corr/Flip/Miss, and entropy flip-risk make that failure measurable and monitorable.