BSA combines Native Sparse Attention with ball-tree neighborhoods to give transformers a global view of 3D point sets at lower compute cost.
Flashattention: Fast and memory-efficient exact attention with i/o-awareness
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
-
BSA: Ball Sparse Attention for Large-scale Geometries
BSA combines Native Sparse Attention with ball-tree neighborhoods to give transformers a global view of 3D point sets at lower compute cost.