Rep-MTL regularizes the shared representation space of multi-task networks by adding an entropy penalty on task saliency and a contrastive alignment between tasks, reporting competitive gains on NYUv2, Cityscapes, Office-31, and Office-Home.
A simple framework for contrastive learning of visual representations
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
background 1representative citing papers
citing papers explorer
-
Rep-MTL: Unleashing the Power of Representation-level Task Saliency for Multi-Task Learning
Rep-MTL regularizes the shared representation space of multi-task networks by adding an entropy penalty on task saliency and a contrastive alignment between tasks, reporting competitive gains on NYUv2, Cityscapes, Office-31, and Office-Home.