I will quickly explain what the huge deep learning rage is all about.

我将快速解释“深度学习狂潮”到底是怎么回事


This graph depicts a neural network that we build and simulate on a computer.

这张示意图传达了我们人类在计算机上构建和模拟的神经网络。

It is a very crude approximation of the human brain.

神经网络和人脑有着一种非常粗糙的近似

The leftmost layer denotes inputs, which can be, for instance, the pixels of an input image.

最左边的层是输入层,例如可以输入图像的像素

The rightmost layer is the output, which can be, for instance, a decision,whether the image depicts a horse or not.

最右边的层是输出层,它可以输出一个判决,无论图像是否描绘了一匹马。

After we have given many inputs to the neural network,in its hidden layers, it will learn to figure out a way to recognize different classes of inputs, such as horses, people or school buses.

在我们为神经网络输入了很多数据之后,它的隐藏层中将学习出一种方法,识别不同类别的输出,例如马,人或校车。


What is really surprising is that it's quite faithful to the way the brain does represent obxts on a lower level.

真正令人感觉神奇的是,这与大脑皮层在较浅层次表述物体的方式上相当吻合。

It has a very similar edge detector.

都有着非常类似的边缘检测器


And, it also works for audio:Here you can find the difference between the neurons in the hearing system of a cat,versus a simulated neural network on the same audio signals.

另外,神经网络的输入输出层还适用于音频:从这张图你可以找出猫咪的听觉系统中神经元与模拟神经网络在相同音频信号处理上的差异

I mean, come on, this is amazing!

我是说,神经网络的运转,这也太神奇了吧!

What is the deep learning part of it all?

那到底什么是深度学习呢?

Well it means that our neural network has multiple hidden layers on top of each other.

很好理解,就是我们的神经网络彼此之间有多个隐藏层

The first layer for an image consists of edges,and as we go up, a combination of edges gives us obxt parts.

图像的第一层由边缘组成,当我们更深入一层时,边缘的组合会给我们生成出对象物体的局部

A combination of obxt parts yield obxts models, and so on.

再由对象物体的局部组合成对象模型,依此类推等等


This kind of hierarchy provides us very powerful capabilities.

神经网络这种层次结构为我们提供了非常强大的能力


For instance, in this traffic sign recognition contest,the second place was taken by humans,but what's more interesting, is that the first place was not taken by humans,
it was taken a by a neural network algorithm.

例如,在这个交通标志识别竞赛中,拿第二名的是人类,便更好玩的是,拿第一名不是人类,是神经网络算法。