垮娉ヤ锛浠虹扮ㄥ界浜浜ワ

ㄥ绉扳浜虹被烘ч涓搴у♀存棰锛绁绉楝奸 Master娆＄韬寮寰姹婀涓琛ㄨラ

ㄥ绛藉ュ存缃绔涓绾垮Master ★瑙绁绁锛瑙楝兼楝硷浜虹被存澶у缁垫锛濂浣浠ュ靛荤筹婧涓1.3ヨ交╁界涓浜烘寤舵锛璁╀腑藉存绗浜哄娲涓澶1.4ヨ存ｈ钩锛ㄧ浜猴璁╄钩绉颁涓衡存涓甯娲炬ョ寮瀵笺

60杩涓涔 挎ㄧ姣Deep Mind CEO Demis Hassabisㄦ逛濡锛

浠杩寰板垮娉存锛杩澶╂浠ㄧ缁瀵瑰骞冲杩琛浜涓浜姝ｅ蹇瀵瑰锛负浜妫楠浠扮垮娉存浠棰浠瑕璋㈡涓浠 Magister(P) Master(P) 璐峰ㄥ存缃浠ュ存缃瀵瑰妫锛涔瑕璋㈡瑙浜猴扮垮娉存ㄥ灞涓瀵х帮浠腑浠ュ扮锛浠ュ瀵瑰缁戒护浠甯告ㄣ

濡姝ょ肩绋搴瀹稿板涔风锛灏烘繁浜涓涓锛澶ф轰涓轰涓哄ㄥ浠ュ板姝ゅ虹ュ

垮娉存锛AlphaGo锛娆惧存浜哄伐鸿界搴朵富瑕宸ヤ娣卞害瀛杩澶ф充娇寮濮澶浜ｅ夸浠ユ缁璧惰涓ュ澶х涓ㄥ娣卞害瀛杩涓涓氨寰浜虹被锛浣跺杩涔澶уぇ浠烘ョ绋搴璇寸瑰氨灏辫浜虹被澶цㄥ璇╀娉ㄥ剧涓枫

杩涓搴富瑕4涓ㄥ锛

1. 璧版缃缁锛Policy Network锛锛缁瀹褰灞棰娴/蜂涓姝ョ璧版

2. 蹇璧板锛Fast rollout锛锛1涓凤浣ㄩ褰虹茶蛋妫璐ㄩ′欢涓锛搴姣1蹇1000

3. 浼板肩缁锛Value Network锛锛缁瀹褰灞浼拌借杩

4. 瑰＄绱Monte Carlo Tree Search锛MCTS锛锛浠ヤ杩涓涓ㄥ杩璧锋ワ褰㈡涓涓寸绯荤

ㄦ甯镐汉浠ョ瑙ｇ璇ヨ锛灏辨浣瑁兼涓璇磋锛跺′贡璧颁姝ュぇ姒浣浠涔搴锛ラ浣濂璺浠ュ灏辨寰涓浜

垮娉哥稿濡缇ゆュご涓灏苟涓瑰娆汉宸ユ鸿界瀹朵浠

村峰ㄦDemis Hassabis锛锛浜哄伐鸿戒涓瀹讹DeepMindTechnologies濮浜恒4宀寮濮涓介璞℃锛8宀缂绋锛13宀寰介璞℃澶у绉板17宀杩ュ妗ュぇ瀛昏璁＄虹瀛涓ㄥぇ瀛锛浠寮濮瀛存2005骞磋ヤ鸡ぇ瀛㈡昏绁缁瀛澹╁ぇ涓娴烽┈浣浣涓虹绌跺璞°涓ゅ勾浠璇5浣涓烘捣椹浼よｄュ浜猴ㄧ虫ユ朵浼复纰锛骞跺椤圭绌跺ラ绉瀛蹇骞村害绐村2011骞村DeepMind Technologies锛浠モ瑙ｅ虫鸿解涓哄缁

澶у峰腑灏 锛David Silver锛锛妗ュぇ瀛绠虹瀛澹纭澹垮ぇ垮浼澶у璁＄虹瀛澹浼ぇ瀛㈣甯锛Google DeepMind绌跺

榛澹帮Aja Huang锛锛婀句氦澶у璁＄虹瀛澹婀惧澶у璁＄虹瀛澹澹垮ぇ垮浼澶у璁＄虹瀛澹Google DeepMind绌跺

や杩颁汉涔澶锛垮娉存璁捐㈤稿浜哄杩DeepMindTechnologies濮浜烘芥┞疯硷Shane Legg锛绌娉路硷Mustafa Suleyman锛绛

界惰浜浜烘病涓浣ㄥ存寰杩楠浜虹缁╋浣浠瀛╁浜寰涓存绁璇锛灏轰姣浣╂

涓涓垮娉存瀵瑰浠ュ浜锛浜虹被涓汉宸ユ鸿藉卞绱㈠存涓澶у冲寮浠ㄧ浜姝ｅ娴璇锛甯告浠骞磋芥轰涓存缁缁浠ュ涓瀹跺卞绱㈢绉存锛ㄧ镐姘姘涓卞棰浼存濂ュ浠灏藉揩甯朵

涓ラ璇昏灏浼村ㄨ娈佃瑙颁ワラ杩杩姝ｅ娴璇锛濂芥ㄥご灏轰瑰浠瀛瀛浠浠ヨ刀蹇涔杈浠缁锛╂ョ烘璐ラ垮娉寸哄ㄤ汉

哄界瀛鹃″瀛瀛濮xuanxiaodi