Injecting EDM-style noise into masked-token position embeddings and adding two auxiliary losses improves I-JEPA's linear-probing accuracy by about 1.5 points on ImageNet-1K.
Big self- supervised models are strong semi-supervised learners
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.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Improving Joint Embedding Predictive Architecture with Diffusion Noise
Injecting EDM-style noise into masked-token position embeddings and adding two auxiliary losses improves I-JEPA's linear-probing accuracy by about 1.5 points on ImageNet-1K.