BRAIN-LIKE MULTI-AGENT DECISION-MAKING ALGORITHM RESEARCH AND ITS APPLICATION

The multi-agent decision-making suffers from the problems of non-interpretability of models and long training time. Therefore, this paper represented the application scenario based on the knowledge base, established causal links between knowledge, and used a self-growing graph structure to fix new knowledge learned after few-shot-learning. A multi-agent decision-making model was designed in the Doudizhu scenario to achieve a behavioral performance brain-like level of intelligence with explainability. Compared with the state-of-the-art landlord model of today, it can achieve a 40% win rate when playing against each other as a landlord with 1 728 times higher time efficiency.