A persistent-entropy regularization loss reduces anisotropy of BERT and RoBERTa embeddings during fine-tuning, but downstream accuracy gains are small and mixed across tasks.
Representation Degenera- tion Problem in Training Natural Language Generation Models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Shrink the longest: improving latent space isotropy with symplicial geometry
A persistent-entropy regularization loss reduces anisotropy of BERT and RoBERTa embeddings during fine-tuning, but downstream accuracy gains are small and mixed across tasks.