TGO-II analysis of ViT-Small/16 training finds decreasing CKA and SVCCA, rising then stable intrinsic dimensionality, and persistent token covariance, indicating simultaneous specialization and manifold expansion without token decoupling.
Imagenet: A large-scale hierarchical image database
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
fields
cs.CV 2representative citing papers
Tree-aware loss functions, especially a Wasserstein compound loss, improve whole brain parcellation and sparse hyperspectral surgical segmentation over standard CE+Dice baselines.
citing papers explorer
-
Transformer Geometry Observatory TGO-II: Representational Similarity Observatory
TGO-II analysis of ViT-Small/16 training finds decreasing CKA and SVCCA, rising then stable intrinsic dimensionality, and persistent token covariance, indicating simultaneous specialization and manifold expansion without token decoupling.
-
Label tree semantic losses for rich multi-class medical image segmentation
Tree-aware loss functions, especially a Wasserstein compound loss, improve whole brain parcellation and sparse hyperspectral surgical segmentation over standard CE+Dice baselines.