The winning rate against people is 84%, and deepmind AI has reached the level of human experts in army chess for the first time

In recent days, , stay DeepMind In a recent paper of , The researchers put forward DeepNash, It is a human free demonstration 、 With no model （model-free） Learn in a way Stratego Self game agent .DeepNask Beat the previous SOTA AI agent , And in the most complex variant of the game Stratego Classic It has achieved the level of expert human players .

Address of thesis ： https://arxiv.org/pdf/2206.15378.pdf

DeepNash The core of is a kind of organization 、 Model free reinforcement learning algorithm , Researchers call it Regularized Nash Dynamics（R-NaD）.DeepNash take R-NaD Combined with a deep neural network architecture , And converge to Nash equilibrium , This means that it learned to compete under incentive , And be robust to competitors who try to take advantage of it .