Data augmentations in contrastive learning are proved to be point estimates of positive-incentive noise, enabling a new learnable π-noise generator framework for augmentations.
Learning transferable visual models from natural language supervision
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Data Augmentation of Contrastive Learning is Estimating Positive-incentive Noise
Data augmentations in contrastive learning are proved to be point estimates of positive-incentive noise, enabling a new learnable π-noise generator framework for augmentations.