Mobile Networks for Computer <i>Go</i>

The architecture of the neural networks used in deep reinforcement learning programs such as AlphaZero or Polygames has been shown to have great impact on performances resulting playing engines. For example, use residual gave a 600 ELO increase strength AlphaGo. This article proposes evaluate interest mobile for game Go using supervised well policy head and value different from heads. accuracy policy, mean squared error value, efficiency with number parameters, speed, trained are evaluated.