NaLaFormer restores query-norm sensitivity in linear attention with a norm-aware power feature map and a cosine direction similarity that keeps attention scores non-negative.
Cluster-wise spatio-temporal masking for efficient video-language pretraining.CoRR, abs/2603.22953,
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
-
Norm$\times$Direction: Restoring the Missing Query Norm in Vision Linear Attention
NaLaFormer restores query-norm sensitivity in linear attention with a norm-aware power feature map and a cosine direction similarity that keeps attention scores non-negative.