Contrastive learning evolves population measures on a fixed manifold into either a unique convex Gibbs equilibrium or a cross-coupled multimodal landscape containing a persistent negative symmetric divergence.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
The Geometric Mechanics of Contrastive Representation Learning: Alignment Potentials, Entropic Dispersion, and Cross-modal Divergence
Contrastive learning evolves population measures on a fixed manifold into either a unique convex Gibbs equilibrium or a cross-coupled multimodal landscape containing a persistent negative symmetric divergence.