Georgia Institute of Technology
Accelerating Monte Carlo Tree Search for discrete action grand strategy games
Abstract
dc:description.abstractIn grand strategy games like Axis & Allies, combat and post combat unit movement generate scale combinatorially, making exhaustive enumeration infeasible. In Python-based implementation of the player, the dominant costs are move generation, breadth-first search over maps, and repeated state cloning. This reduces the iterations for effective search in the time budget. Sequential Monte Carlo Tree Search addresses the move generation problem by decomposing the action space per territory, but the forward model functions remain dominant computational bottlenecks. This thesis investigates how hybrid execution models and selective action pruning can be combined to accelerate Monte Carlo Tree Search. The model achieves a 77% win rate across 100 games in the four-player environment. Neural policy guidance is further explored as a direction for improving strategic effectiveness.
Degree
thesis:*- Name thesis:degree_name
- Electrical and Computer Engineering, MS
- Department thesis:degree_department
- Electrical and Computer Engineering
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Nayak, Sanjana Ganesh
- Advisor dc:contributor.advisor
-
- Hale, Matthew
- Committee members dc:contributor.committeemember
-
- Tucker, Maegan
- Coogan, Samuel
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81614