Google AlphaGo has changed, Ke Jie can win this year? -Man-machine war, AlphaGo, Ke Jie, go-IT information
Google's artificial intelligence system, Alpha Dog (AlphaGo) next opponent may prove to be the world number one chess player Ke Jie, China.
On June 4, at the 37th session of the State General Administration of sport on the world amateur go Championship press conference chess sport Administration Centre Party Secretary, international go Federation Affairs Officer, Yang Junan said if nothing else Ke Jie section will be carried out this year and the "Alpha Dog" go "the ultimate man-machine war."
However, this battle of the time and place yet to be decided.
On March 15 this year, Alpha Dog 4:1 to beat Korea player Li Shishi, followed by Korea go Institute for Alpha Dog issue honorary certificates.
Under the "Go Ratings" world go score ranking as of June 4, 2016, Ke Jie rankings, AlphaGo ranked second.
World Chess rankings.
AlphaGo software and hardware upgrade, become more powerful
Battle of AlphaGo and Li Shishi, artificial intelligence game between the first and the highest grade go player. Different in 1997, deep blue chess chess AlphaGo chess chess is more complex, each more likely to produce after the next step. But behind the AlphaGo algorithm has been shown to their advantage in the competition. By deep learning, it has been able to simulate human chess, with "intuitive", and gives the next best choice.
For the AlphaGo, and after winning the Li Shishi, did not stop the pace of learning, both in software and hardware upgrades. In May this year, Google announced it had set up a specially tailored for machine learning "TPU" (Tensor Processing Unit) of ASIC chips, it is specifically optimized for TensorFlow, and performance than the NVIDIA GPU, equivalent to Google based on Moore's law skips forward three steps.
According to the reports of the verge, Google's use of TPU to address data has been more than a year, and use it to optimize search results and Google Street view project. Meanwhile, AlphaGo and Li Shishi TPU processor challenges.
Computers, almost all of the applications are available through the dedicated hardware acceleration, but you have to let the operation worthwhile. For example, because people like to watch the video, most people will use a number of standard codecs, makes playback smoother and less battery drain. Now Google can "TPU" benefit of ASIC chips.
In addition, if AlphaGo really sure Ke Jie war, Google may use a more powerful version of the AlphaGo. AlphaGo in two versions, one is the "single version", the other is a "decentralized", also known as the network version. Both algorithms were the same, the difference is the hardware.
Stand-alone AlphaGo with 48 central processing unit (CPU) and 8 graphics processing unit (GPU). Decentralized version AlphaGo with 1202 CPU and 176 graphics processors.
In comparison, several go before the AlphaGo artificial intelligence programs, only 32 CPU CrazyStone and Zen only 8 CPU. Stand-alone AlphaGo once and 5 Kyu are 495 games, competitors are CrazyStone, and Zen, Pachi, Fuego and GnuGo,AlphaGo all scored a landslide victory. Even if it is 4, AlphaGo Crazystone and Zen win percentage reached 77% and 86%. As for the distributed version of the AlphaGo, if the stand-alone version to its own AlphaGo, the distributed version of the game winning is 77%. In addition, there are foreign media said AlphaGo team and Li Shishi on war, after a simple rest, there is no reduction in the study of go, but increasing the number of teams and resources. In other words, Ke Jie will face a more powerful "Alpha Dog".
In addition to the improvements expected outside of their ability, when AlphaGo duel with Li Shishi, criticized the rules of this trend might have changed the game. AlphaGo and Li Shishi vs 5, using the rules of Chinese chess competition (at the end of the game, player after player posted), Alpha Go the usual "training" of rules, and Li Shishi was more familiar with Korean rules. In accordance with the rules of the game, each with 2 hours of layout time 60 seconds the seconds and 3 times (player will be limited after the countdown time). Each game is expected to 4-5 hours.
Ke Jie said: beat Li Shishi afagou, but it can't beat me
Li Shishi AlphaGo during the war, Chinese Professional go player Ke Jie had introduced himself on weibo: me Ke Jie, was born in 97. Now is the first person world go, race I'm rolling tangentbifurcation 8:2 Li Shishi (he won two sets without any effect), world champion, has won three times (this age so unprecedented), national champion many times. Alpha Dog to win after the first game, Ke Jie Tweet said, "even beat Li Shishi afagou, but it can't beat me. ”
Eventually the game ends with 4:1, Ke Jie said: "come on! you are afagou or Alfa Laval cat! I Ke Jie Qi wave not seen? Made storms more violent it! x I installed how! "
Google is also looking after the win Li Shishi Ke Jie. DeepMind research engineer laliya·hadesaier in public on their personal Facebook account, "said AlphaGo may now have an official rank. (Thank you, losing you, Li Shishi, if win wouldn't be ranking algorithm). Now the integral of AlphaGo is-3533, ranking 4th in the world. Ke Jie, are you ready? ”
Finally, Ke Jie AlphaGo game and can make the trip, and Ke Jie can win the game, maybe we should pay more attention to Google artificial intelligence input and strategic vision.
AlphaGo development team after beating Li Shishi DeepMind Demis Hassabis, founder of AlphaGo said, "to some extent, we are only interested in those things that help us to research projects. DeepMind goal is not just to get the game winning, but also fun and inspiration. But from a personal standpoint, I like to play games, I've developed a computer game. But in a sense, they are test-bed, an attempt to write algorithms and testing platform. Eventually, we hope to be able to apply technology to solve real world problems. ”
谷歌AlphaGo已今非昔比，柯洁年内能否获胜？ - 人机大战,AlphaGo,柯洁,围棋 - IT资讯
谷歌人工智能系统阿尔法狗（AlphaGo）下一个对手或许会是目前世界排名第一的中国棋手柯洁。
6月4日，在第37届世界业余围棋锦标赛新闻发布会上国家体育总局棋牌运动管理中心党委书记、国际围棋联盟事务总长杨俊安透露，如果不出意外柯洁九段将在年内进行和“阿尔法狗”的围棋“终极人机大战”。
不过，这场对决的具体时间地点目前尚未确定。
今年3月15日，阿尔法狗最终以4:1的比分战胜韩国棋手李世石，随后韩国棋院为阿尔法狗颁发名誉九段证书。
根据“Go Ratings”世界围棋等级分排名，截至2016年6月4日，柯洁排名第一，AlphaGo排名第二。
世界围棋等级分排名。
AlphaGo软件、硬件升级，变得更加强大
AlphaGo与李世石之战，是人工智能首次与最高段位围棋选手之间的对弈。不同于1997年深蓝对弈的国际象棋，AlphaGo对弈的围棋更加复杂，每下一步后能产生的可能性也更多。但AlphaGo背后的算法在比赛中已经显示出自己的优势。通过深度学习，它已经能模拟人类下棋，拥有“直觉”，并能给出下一步的最佳策略选择。
对于AlphaGo来说，赢了李世石后，并没有停下学习的脚步，在软件和硬件上都进行了升级。就在今年5月份，谷歌宣布它已经建立了专为机器学习定制的“TPU”（Tensor Processing Unit）的ASIC芯片，它专门针对TensorFlow进行优化，性能要优于英伟达的GPU，相当于谷歌依据摩尔定律向前跳了三步。
根据《the verge》的报道，谷歌其实利用TPU来处理数据已有一年多时间，并利用其来优化了搜索结果和谷歌街景项目。同时，AlphaGo与李世石的挑战也用了TPU处理器。
对于计算机来说，几乎所有的应用程序都可以通过专用硬件加速，但你必须让这样的操作物有所值。例如，因为人们喜欢看视频，大多人会使用一些标准的编解码器，使得回放更流畅，电池消耗也更少一些。现在谷歌可以从“TPU”的ASIC芯片中受益。
此外，AlphaGo若真的确定与柯洁对战，谷歌可能会使用另一个更强大版本的AlphaGo。AlphaGo两个版本，一个是“单机版”，另一个是“分散式”，也可以称为联网版本。两者的演算法完全相同，差别在于所用的硬件。
单机版AlphaGo拥有48个中央处理器（CPU）和8个图形处理器（GPU）。而分散式版AlphaGo则拥有1202个中央处理器和176个图形处理器。
作为比较，在AlphaGo之前的几个围棋人工智能程序，CrazyStone只有32个CPU，而Zen只有8个CPU。单机版AlphaGo曾经与5种围棋程序进行了495场比赛，对手是CrazyStone、Zen、Pachi、Fuego和GnuGo，AlphaGo全部取得压倒性的胜利。即使是让4子，AlphaGo对Crazystone和Zen的胜率也达到了77%和86%。至于分散式版的AlphaGo，如果对上自家的单机版AlphaGo，则分散式版本的比赛胜率是77%。另外，有外媒称，AlphaGo团队在与李世石的对战后，经过简单的休整，不但没有减少对围棋的研究，反而扩大了团队人数和投入资源。换句话说，柯洁将会面对一个更强大的“阿尔法狗”。
除了自身能力的改进让外界期待外，此前AlphaGo与李世石对决时，被人诟病的比赛规则这次可能会改变比赛走向。AlphaGo与李世石大战5盘，使用的是中国围棋竞赛规则(比赛结束时，后走棋的棋手贴目)，也是Alpha Go平时“训练”使用的规则，而李世石则更熟悉韩式规则。按照比赛规则，双方各有2小时的布局时间及3次60秒的读秒(棋手将限时用完后的倒计时时间)。每场比赛预计需要4-5个小时。
柯洁扬言：阿法狗战胜了李世石，但它赢不了我
在李世石与AlphaGo大战期间，中国职业围棋九段选手柯洁曾在微博上介绍过自己：我叫做柯洁，97年出生。现在暂时是世界围棋第一人，正式比赛我是单盘8:2碾压李世石（他赢得这两盘没有任何作用），世界冠军获得过三次（此年龄如此成绩前无古人），大小国内赛事冠军无数次。在阿尔法狗赢下第一盘棋后，柯洁发微博称，“就算阿法狗战胜了李世石，但它赢不了我。”
最终比赛以4:1结束，柯洁扬言：“来吧!管你是阿法狗还是阿法猫!我柯洁在棋上什么大风大浪没见过？让风暴再来的猛烈点吧!这个X我就装了怎样!”
谷歌方面也在赢下李世石后就将目光投向了柯洁。DeepMind的研发工程师拉利亚·哈德塞尔在其个人Facebook账号上公开表示，“AlphaGo现在可以有正式排名了。（谢谢输给你，李世石，如果一直赢棋就不能被排名算法统计）。现在AlphaGo的积分是3533分，排名世界第4。柯洁，准备好了吗？”
最终，无论柯洁与AlphaGo的比赛能否成行，以及柯洁能否在比赛中获胜，我们或许更应该关注的是谷歌对于人工智能的投入和战略眼光。
AlphaGo的开发团队DeepMind的创始人Demis Hassabis在AlphaGo击败李世石后说：“在一定程度上，我们只对那些有助于我们研究计划的东西感兴趣。DeepMind的目标不仅仅是获得游戏胜利，还要从中获得乐趣和启发。但从个人角度来说，我喜欢玩游戏，我也曾开发过电脑游戏。可是从某种程度上说，它们又都是试验台，即尝试编写算法并对其进行测试的平台。最终，我们希望能将技术应用于解决现实世界的难题。”