In generalized contrastive learning with imbalanced classes, optimal representations collapse to class means whose angular geometry is determined by class proportions via convex optimization, and extreme imbalance causes all minority classes to collapse to one vector.
Advances in Neural Information Processing Systems , volume=
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
citation-role summary
method 1
citation-polarity summary
fields
cs.LG 3years
2026 3verdicts
UNVERDICTED 3roles
method 1polarities
use method 1representative citing papers
The paper proves statistical consistency of contrastive loss for retrieval via an AUC criterion and derives generalization bounds O(1/m + 1/sqrt(n)) supervised and O(1/sqrt(m) + 1/sqrt(n)) self-supervised that explain benefits of large negative sets.
Derives novel generalization error bounds for multimodal pairwise metric learning showing that fine-grained modality features reduce hypothesis space complexity via enhanced complementarity.
citing papers explorer
-
Optimal Representations for Generalized Contrastive Learning with Imbalanced Datasets
In generalized contrastive learning with imbalanced classes, optimal representations collapse to class means whose angular geometry is determined by class proportions via convex optimization, and extreme imbalance causes all minority classes to collapse to one vector.
-
Statistical Consistency and Generalization of Contrastive Representation Learning
The paper proves statistical consistency of contrastive loss for retrieval via an AUC criterion and derives generalization bounds O(1/m + 1/sqrt(n)) supervised and O(1/sqrt(m) + 1/sqrt(n)) self-supervised that explain benefits of large negative sets.
-
Quantifying Multimodal Capabilities: Formal Generalization Guarantees in Pairwise Metric Learning
Derives novel generalization error bounds for multimodal pairwise metric learning showing that fine-grained modality features reduce hypothesis space complexity via enhanced complementarity.