Computational Intelligence and Computer Search

几个世纪以来，国际象棋和围棋等游戏一直是人类展示智力力量的竞技场。 如今，这类游戏已经成为机器展示其计算智能的竞技场。计算智能的一个关键方面是搜索。 这个项目的目标是找到改进的搜索算法。 一个典型的问题是在两个玩家的游戏中找到下一步，比如国际象棋或围棋。在这里，算法必须搜索游戏可能继续的大空间，并选择一个将被证明是强大的移动。 从历史上看，为国际象棋或围棋等游戏构建强大的人工玩家一直是人工智能领域的主要挑战。 最近，这一领域取得了巨大进展。大约10年前，一种新的搜索算法--蒙特卡罗树搜索（MCTS）--将经典的树搜索技术与随机移动采样相结合，彻底改变了计算机围棋。我计划继续研究那些既能解释MCTS成功又能导致其改进的基本问题。最近，另一种算法--深度学习--出现了，它预示着另一场这样的革命。我也打算在这方面努力。

For centuries, games such as chess and Go have served as an arena in which humans showed their intellectual strength. Today, such games serve as an arena in which machines show their computational intelligence. A key aspect of computational intelligence is search. The goal of this project is to find improved search algorithms. A typical problem is to find the next move in a two-player game such as chess or Go. Here, the algorithm must search a large space of possible continuations of the game and select a move which will turn out to be strong. Historically, the problem of building strong artificial players for games such as chess or Go has been a main challenge in the field of artificial intelligence. Recently, there has been tremendous progress in this field. Around 10 years ago, a new search algorithm --- Monte Carlo Tree Search (MCTS) --- that combines classical tree-search techniques with random move sampling revolutionized computer Go. I plan to continue work on fundamental problems that both explain the success of MCTS and lead to improvements for it. More recently, another algorithm --- Deep Learning --- has emerged that promises another such revolution. I plan to work on this as well.