.png){kind=icon}
.png){kind=icon}
Return
Global Structure-Aware R-Tree: a spatial indexing mechanism using Deep Reinforcement Learning and Self-Play
.png){kind=icon}
PRE
AI
S
H
L
J
DOI：10.1016/j.envsoft.2026.107077.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• A deep reinforcement learning framework is proposed for incremental R-Tree construction. • A self-play mechanism is introduced to guide policy improvement through query efficiency comparison. • A subtree-structure-aware state representation captures hierarchical and distributional information. • The learned insertion policy improves global tree structure beyond local MBR-based heuristics. • GSAR-Tree reduces node accesses by 37.9%–1193.8% compared with R*-Tree.
Journal
E
IF:
4.6
Papers:
191
Citations:
0
Organization
Cited Papers