.png){kind=icon}
.png){kind=icon}
Return
LiSA: A linear graph transformer framework with sample aggregation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2026.115905.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• LiSA: novel attention via optimized kernels; gains without extra params or complexity. • Redesigned kernels match softmax; 16.4% faster training without accuracy loss. • Superior on 8 datasets vs 14 baselines; strong noise robustness; broad GNN applicability.
Keywords:
LiSA
attention mechanism
graph neural networks
kernel optimization
sample aggregation
Journal
K
IF:
7.6
Papers:
1.2W
Citations:
4.5W