《科学》（20181207出版）一周论文导读

Science, 7 DECEMBER 2018, VOL 362, ISSUE 6419
[本文来自：www.11jj.com]

《科学》2018年12月7日第6419期362卷 [转载出处：www.11jj.com]

人工智能  Artificial Intelligence

A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play

横扫围棋、国际象棋和将棋的AI程序

▲ 作者：David Silver, Thomas Hubert, et al

▲ 链接：

http://science.sciencemag.org/content/362/6419/1140

▲ 摘要：

数十年来，人工智能领域对棋类游戏的开发研究越来越深入。

传统的棋类游戏程序依托于先进的搜索技术、人为制定的评价规则以及对某些特定领域的修正。

而如今AlphaGoZero正通过不断的自我训练，在围棋游戏中取得了超人的表现，成为最强大的棋类人工智能。

在《科学》发表的最新研究中，来自Deepmind公司的研究人员将这种自我训练的方法推广至单个AlphaZero算法中，该算法能在许多博弈游戏中取得惊人表现。

从随机对弈开始，除基本游戏规则外，AlphaZero并没有先验知识，但已相继打败了围棋、国际象棋、..将棋的世界冠军程序。

这种通过自学掌握多种游戏规则的能力使得人工智能向通用游戏系统又迈进了一步。

▲ Abstract

The game of chess is the longest-studied domain in the history of artificial intelligence. The strongest programs are based on a combination of sophisticated search techniques, domain-specific adaptations, and handcrafted evaluation functions that have been refined by human experts over several decades. By contrast, the AlphaGo Zero program recently achieved superhuman performance in the game of Go by reinforcement learning from self-play. In this paper,we generalize this approach into a single AlphaZero algorithm that can achieve superhuman performance in many challenging games. Starting from random play and given no domain knowledge except the game rules, AlphaZero convincingly defeated a world champion program in the games of chess and shogi (Japanese chess), as well as Go.

化学 Chemistry

Room-temperature cycling of metal fluoride electrodes: Liquid electrolytes for high-energy fluoride ion cells

金属氟化物电极的室温循环：用于高能氟离子电池的液体电解质

▲ 作者：Victoria K. Davis, Christopher M. Bates, Kaoru Omichi, et al