Unverified50% confidenceFactExact time
EMA (Efficient Multi-scale Attention) is a lightweight attention mechanism that enhances joint channel-spatial representation through cross-dimensional multi-scale feature interaction, with lower computational overhead than standard self-attention.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCommon attention mechanism modules for model improvement include the SE attention mechanism, CBAM channel and spatial attention, and ECA efficient channel attention.69% similarVerifiedMulti-Head Latent Attention (MLA) is an attention mechanism proposed by DeepSeek-V2.65% similarUnverifiedSE (Squeeze-and-Excitation) and CBAM (Convolutional Block Attention Module) are common attention modules in the field of computer vision.61% similarUnverified滑动窗口注意力（SWA）由 Mistral AI 广泛采用，将注意力复杂度从 O(N²) 降至 O(N×W)61% similarUnverified新内核为滑动窗口注意力（SWA）层提供双输入支持，并引入 attention sinks 机制61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/926047API
curl https://kongchang.com/api/v1/knowledge/claims/926047MCP
get_claim(id=926047)