Semantic geometry emerges transiently early in next-token prediction training before collapsing to Neural Collapse symmetry in synthetic settings with latent semantic factors.
arXiv preprint arXiv:2110.02796 , year=
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 4verdicts
UNVERDICTED 4roles
background 1polarities
background 1representative citing papers
Supervised classification reaches neural collapse by design via normalized prototype losses on the hypersphere, outperforming CE and SCL on ImageNet-1K and other benchmarks with faster convergence and better transfer.
Wide neural networks with cross-entropy loss remain in the lazy training regime under parameter-space regularization or non-degenerate targets, allowing explicit NTK-based solution characterization and uncertainty analysis.
CPT creates cluster-invariant spaces from pre-trained VLM semantics and applies neural collapse losses to boost long-tail performance and unseen-class generalization in prompt tuning.
citing papers explorer
-
Structure Before Collapse: Transient semantic geometry in next-token prediction
Semantic geometry emerges transiently early in next-token prediction training before collapsing to Neural Collapse symmetry in synthetic settings with latent semantic factors.
-
Neural Collapse by Design: Learning Class Prototypes on the Hypersphere
Supervised classification reaches neural collapse by design via normalized prototype losses on the hypersphere, outperforming CE and SCL on ImageNet-1K and other benchmarks with faster convergence and better transfer.
-
The Neural Tangent Kernel for Classification
Wide neural networks with cross-entropy loss remain in the lazy training regime under parameter-space regularization or non-degenerate targets, allowing explicit NTK-based solution characterization and uncertainty analysis.
-
Cluster-Aware Neural Collapse Prompt Tuning for Long-Tailed Generalization of Vision-Language Models
CPT creates cluster-invariant spaces from pre-trained VLM semantics and applies neural collapse losses to boost long-tail performance and unseen-class generalization in prompt tuning.