ADM-201 dump PMP dumps pdf SSCP exam materials CBAP exam sample questions

破解DeepMind的游戏人工智能 – 译学馆
未登录,请登录后再发表信息
最新评论 (0)
播放视频

破解DeepMind的游戏人工智能

Breaking DeepMind's Game AI System | Two Minute Papers

亲爱的学霸们 大家好 这里是由Károly Zsolnai-Fehér带来的两分钟论文
Dear Fellow Scholars, this is Two Minute Papers with Károly Zsolnai-Fehér.
不久之前谷歌DeepMind提出了一种新的机器学习算法
Not so long ago, Google DeepMind introduced a novel learning algorithm that was able to
它能够在许多雅达利游戏中达到超越人类的水平
reach superhuman levels in playing many Atari games.
这是人工智能研究中的一个重要的里程碑
It was a spectacular milestone in AI research.
有趣的是 虽然这些学习算法正在以惊人的速度不断改进
Interestingly, while these learning algorithms are being improved at a staggering pace, there
然而在它的一个平行分支里 研究人员正尝试通过略微改变提交给它们的信息
is a parallel subfield where researchers endeavor to break these learning systems by slightly
从而对这些学习系统加以破解
changing the information they are presented with.
如果你愿意 可以对图像或视频进行欺诈性的篡改
Fraudulent tampering with images or video feeds, if you will.
设想一个被设计用作图像内容识别的系统
Imagine a system that is designed to identify what is seen in an image.
在之前的节目里 我们用了个有趣的例子讨论过一个对抗算法
In an earlier episode, we discussed an adversarial algorithm, where in an amusing example, they
他们为这个图像增加了一点点几乎无法察觉的噪声 使得深层神经网络
added a tiny bit of barely perceptible noise to this image, to make the deep neural network
把一辆公共汽车误认为一只鸵鸟
misidentify a bus for an ostrich.
机器学习研究者喜欢称这些为“邪恶的伪造图像对抗样本”
Machine learning researchers like to call these evil forged images adversarial samples.
现在这一次 OpenAI发表了一个超级有趣的作品
And now, this time around, OpenAI published a super fun piece of work to fool these game
它能通过改变一些他们输入的视觉信息来愚弄这些游戏学习算法
learning algorithms by changing some of their input visual information.
正如你将看到的一样 它如此有效 并且只需使用一点点的信息
As you will see in a moment, it is so effective that by only using a tiny bit of information,
它就可以将一个强大的学习算法变成一个胡言乱语的白痴
it can turn a powerful learning algorithm into a blabbering idiot.
第一种方法在大部分的视频输入中增加了微小的噪点
The first method adds a tiny bit of noise to a large portion of the video input, where
其中的差异几乎不可察觉 但它能迫使学习算法做出
the difference is barely perceptible, but it forces the learning algorithm to choose
一个本不该如此的选择
a different action that it would have chosen otherwise.
另一个方法则使用了一个不同的修改方式 它有着更小的印记
In the other one, a different modification was used, that has a smaller footprint, but
但效果却更加明显
is more visible.
例如 在乒乓球游戏中添加一个微小的假球
For instance, in pong, adding a tiny fake ball to the game to coerce the learner into
从而迫使原本打算上升的学习者选择下降
going down when it was originally planning to go up.
该算法能够学习几乎任何其它游戏的特定知识
The algorithm is able to learn game-specific knowledge for almost any other game to fool
来愚弄玩家
the player.
尽管结果有很大的差异 但我喜欢这两种噪音类型中优雅的数学表达式
Despite the huge difference in the results, I loved the elegant mathematical formulation
因为尽管事实上它们做的事情毫不相同
of the two noise types, because despite the fact that they do something radically different,
但它们的数学表达式却非常相似 数学家喜欢说 我们正在解决的是相同的问题
their mathematical formulation is quite similar, mathematicians like to say that we’re solving
尽管需要针对不同的目标规范做出优化
the same problem, while optimizing for different target norms.
除了DeepMind的Deep Q-Learning之外 其它两种高质量的学习算法
Beyond DeepMind’s Deep Q-Learning, two other high-quality learning algorithms are also
也同样会被这种技术所欺骗
fooled by this technique.
在白盒方案中 我们可以访问算法的内部运作
In the white box formulation, we have access to the inner workings of the algorithm.
但有趣的是 还有一个黑盒方案被提出来
But interestingly, a black box formulation is also proposed, where we know much less
在这个方案里 我们对目标系统的了解少得多 但是我们了解游戏本身 并且我们能训练我们自己的系统
about the target system, but we know the game itself, and we train our own system and look
并寻找其中的弱点
for weaknesses in that.
当我们发现弱点时 我们便能使用这些知识破解其它系统
When we’ve found these weak points, we use this knowledge to break other systems.
我只能想象 当作者们在开发这些技术时
I can only imagine how much fun there was to be had for the authors when they were developing
他们会有多大的乐趣
these techniques.
看到这个创造更强学习算法的军备竞赛令人超级兴奋
Super excited to see how this arms race of creating more powerful learning algorithms,
而作为回应 更强的对抗技术也会打破它们的发展
and in response, more powerful adversarial techniques to break them develops.
在将来 我觉得学习算法的健壮性 或换句话说
In the future, I feel that the robustness of a learning algorithm, or in other words,
它对于对抗性攻击的适应力 将会成为一个与其强大性能
its resilience against adversarial attacks will be just as important of a design factor
同等重要的设计因素
as how powerful it is.
作者的网站上有大量已发布的视频 一定要去看一看
There are a ton of videos published on the authors’ website, make sure to have a look.
而且 如果您希望支持该系列作品 请务必查看我们的Patreon页面
And also, if you wish to support the series, make sure to have a look at our Patreon page.
我们非常感谢您的贡献 这绝对有助于保持该系列运行
We kindly thank you for your contribution, it definitely helps keeping the series running.
感谢您的关注和慷慨支持 我们下次再见!
Thanks for watching and for your generous support, and I’ll see you next time!

发表评论

译制信息
视频概述

随着学习算法的不断改善,人工智能正变得越来越强大。但是,你知道人工智能也可以轻易地被欺骗吗?对于DeepMind这样的人工智能,研究人员也找出了破解之道,轻松地愚弄了DeepMind的游戏人工智能。

听录译者

收集自网络

翻译译者

B11101001

审核员

知易行难

视频来源

https://www.youtube.com/watch?v=QFu0vZgMcqk

相关推荐