The abstract proposes replacing layer normalization with DynTanh in IJEPA to preserve token energy and reports improved ImageNet and depth metrics, but the full text is a different paper.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Elucidating the Role of Feature Normalization in IJEPA
The abstract proposes replacing layer normalization with DynTanh in IJEPA to preserve token energy and reports improved ImageNet and depth metrics, but the full text is a different paper.