This paper introduces Deep Monte Carlo (DMC), a neural-enhanced algorithm designed for mastering Four-Player Military Chess, a complex imperfect-information game characterized by dynamic alliances and hierarchical combat mechanics. The proposed framework integrates three key components through a lightweight multi-branch neural network: a 17×17×8 tensor encoding spatial unit distributions, terrain features, and multi-scale threat maps; a 64-dimensional action vector representing move parameters and combat outcomes; and an 8-step historical seq...
