AI Game Service MCTS Structure for Incomplete-Information Play
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The Monte-Carlo tree search (MCTS) algorithm is ineffective for static games due to incomplete information and dynamic rule disruptions, leading to irregular tree structures and preventing normal search in games like Mahjong, where participants cannot see each other's tiles or actions.
Innovation Solution
Incorporating logic distribution and gate nodes into the Monte-Carlo tree structure to enable unidirectional information flow and backpropagation, allowing the MCTS algorithm to be applied to static games by ensuring execution actions and benefits are transferred and propagated in a controlled manner.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If MCTS algorithm is applied to static games with incomplete information, then the algorithm can handle dynamic games effectively, but the tree structure becomes irregular and bidirectional information flow prevents normal search operations
Solution Approach 1:
The patent segments the Monte-Carlo tree into distinct node types (logic distribution nodes and logic gate nodes) to handle different aspects of static game processing. Logic distribution nodes manage candidate action distribution while logic gate nodes handle information aggregation, creating a structured framework that enables normal search operations in static games with incomplete information
Solution Approach 2:
The patent introduces logic distribution nodes and logic gate nodes as intermediary elements between participant nodes. These intermediary nodes facilitate unidirectional information flow by acting as mediators that receive actions from one participant and transmit them to appropriate logic gates, preventing bidirectional information flow issues while maintaining adaptability to static games
2Ease of operation
If logic distribution nodes and logic gate nodes are added to the Monte-Carlo tree structure, then unidirectional information flow is achieved enabling normal search, but the device complexity increases
Solution Approach 1:
The patent designs logic distribution nodes and logic gate nodes with multi-functional capabilities. Logic distribution nodes simultaneously handle candidate action generation, distribution to appropriate logic gates, and state management. Logic gate nodes aggregate information from multiple sources and determine next steps. This universality reduces the need for separate specialized components, mitigating the increase in device complexity while enabling normal search operations
Data Source
AI summary
This application discloses a game service execution method performed at a computer device for a game service. The method includes: constructing, in an ith round of game in the game service including a static game, a Monte-Carlo tree corresponding to the ith round of game, where the Monte-Carlo tree includes: an initial node, at least two participant nodes, at least two logic distribution nodes, and at least two logic gate nodes that correspond to the static game; simulating a game result on each path in the Monte-Carlo tree; backpropagating benefits of the game results corresponding to multiple paths in the Monte-Carlo tree to a highest-layer AI participant node of the AI participant in the Monte-Carlo tree; and selecting an execution action of the highest-layer AI participant node on a path that has the highest benefit, and outputting the execution action as an AI execution action of the AI participant.