人工智能：一种现代方法ch05 Adversarial Search - 2016.pdf

博弈搜索（对抗搜索） 1 博弈： 如何应对巨大状态空间？ 博弈的形式化。 2 Minimax算法 3 Alpha-Beta算法 1 Review  Breadth-first f(n)=depth(n)  Depth-first f(n)=-depth(n)  Depth limited  Iterative deepening  Uniform-cost f(n)=g(n)  Bidirectional  Greedy best-first f(n)=h(n)  A* f(n)=g(n)+h(n)  Hill climbing：steepest ascent /stochastic /first-choice /random restart  Simulated annealing (allowing bad moves)  Local beam (keeping k states)  Genetic algorithm (combining two states to generate successors) 3 寻找满意的后继 博弈 4 Game*  Adversarial(对抗) search problems  often known as games(博弈).  Game theory  For centuries  humans have used games like Chess or Go to exert their intelligence.  Recently  there has been great success in building game programs that challenge human supremacy. 5 Game  Today’s topic about games  Two-player, turn-taking  fully observable, deterministic  zero-sum(零和, 1 + (-1) = 0)  time-constrained 6 Games are a form of multi-agent environment.  multi-agent environment  What do other agents do and how do they affect our success?  Cooperative vs. competitive multi-agent environments.  Competitive multi-agent environments give rise to adversarial search a.k.a. games