引导式测验 · 中等 等级

AlphaGo 和 AlphaZero 测验

回顾 AlphaGo 的专家对弈和自我对弈训练、AlphaZero 基于规则的学习和搜索，以及 MuZero 的规划学习模型。

相关引导路径阿尔法戈和阿尔法零
问题 1 的 8

为什么围棋游戏被认为对计算机来说特别难？

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. 为什么围棋游戏被认为对计算机来说特别难？
    • 它有大量的可能位置
    • 它的可能动作很少
    • 它只在纸上播放
    • 它没有规则
  2. 2016 年 AlphaGo 4-1 击败谁？
    • 加里·卡斯帕罗夫
    • 李世石
    • 马格努斯·卡尔森
    • 随机的业余爱好者
  3. Which three games did AlphaZero’s paper evaluate?
    • Go, chess, and shogi
    • Poker, bridge, and backgammon
    • Atari, StarCraft, and Dota
    • Only Go and chess
  4. AlphaZero 与其神经网络配对的搜索方法是什么？
    • 电话簿的二分查找
    • 线性回归
    • 冒泡排序
    • 蒙特卡罗树搜索
  5. AlphaZero 的神经网络预测哪两件事来指导其搜索？
    • 政策（好的举措）和价值（谁获胜）
    • 天气和时间
    • 对手的姓名和年龄
    • 颜色和声音
  6. What did MuZero add to the AlphaZero-style planning approach?
    • A requirement for human game records
    • A learned model of reward and environment dynamics used for planning without being given the rules
    • A replacement for all tree search
    • A fixed database of every possible board position
  7. What information did AlphaZero need in its board-game experiments besides self-play?
    • A database of professional games
    • A complete endgame table for every position
    • The rules of the games
    • Human rankings for every possible move
  8. How did the original AlphaGo system combine supervised and reinforcement learning?
    • It trained a policy from expert games and then improved through self-play
    • It used only random moves and never trained on human games
    • It learned the game rules from images without playing
    • It relied only on a fixed database of winning games