A spectral-graph-derived weakly-supervised contrastive loss, using estimated semantic similarity as edge weights, improves noisy-label and partial-label learning.
The second way to construct S is the same as the method used in our main paper
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Weakly-Supervised Contrastive Learning for Imprecise Class Labels
A spectral-graph-derived weakly-supervised contrastive loss, using estimated semantic similarity as edge weights, improves noisy-label and partial-label learning.