100:00:02,840 --> 00:00:07,040If yo

1
00:00:02,840 --> 00:00:07,040
If you’ve been ignoring neural nets cuz you think they’re too hard to understand

2
00:00:07,040 --> 00:00:11,040
or you think you don’t need them…boy do I have a treat for you!

3
00:00:11,040 --> 00:00:16,080
In this video you’ll learn about neural nets without any of the math or code –

4
00:00:16,080 --> 00:00:19,200
just an intro to what they are and how they work.

5
00:00:19,200 --> 00:00:23,280
My hope is that you’ll get an idea for why they’re such an important tool.

6
00:00:23,280 --> 00:00:24,560
Let’s get started.

7
00:00:24,560 --> 00:00:25,920


8
00:00:25,920 --> 00:00:30,360
The first thing you need to know is that deep learning is about neural networks.

9
00:00:30,360 --> 00:00:34,360
The structure of a neural network is like any other kind of network;

10
00:00:34,360 --> 00:00:38,160
there is an interconnected web of nodes, which are called neurons,

11
00:00:38,160 --> 00:00:40,560
and the edges that join them together.

12
00:00:40,560 --> 00:00:44,500
A neural network's main function is to receive a set of inputs,

13
00:00:44,500 --> 00:00:47,580
perform progressively complex calculations,

14
00:00:47,580 --> 00:00:50,380
and then use the output to solve a problem.

15
00:00:50,380 --> 00:00:53,800
Neural networks are used for lots of different applications,

16
00:00:53,800 --> 00:00:57,140
but in this series we will focus on classification.

17
00:00:57,140 --> 00:01:01,520
If you wanna learn about neural nets in a bit more detail, including the math,

18
00:01:01,520 --> 00:01:08,360
my two favourite resources are Michael Nielsen's book, and Andrew Ng's class.

19
00:01:08,360 --> 00:01:15,100
Before we talk more about neural networks, I’m gonna give you a quick overview of the problem of classification.

20
00:01:15,100 --> 00:01:19,000
Classification is the process of categorizing a group of objects,

21
00:01:19,000 --> 00:01:23,160
while only using some basic data features that describe them.

22
00:01:23,160 --> 00:01:26,040
There are lots of classifiers available today -

23
00:01:26,040 --> 00:01:32,340
like Logistic Regression, Support Vector Machines, Naive Bayes, and of course, neural networks.

24
00:01:32,340 --> 00:01:38,240
The firing of a classifier, or activation as its commonly called, produces a score.

25
00:01:38,240 --> 00:01:42,400
For example, say you needed to predict if a patient is sick or healthy,

26
00:01:42,400 --> 00:01:46,720
and all you have are their height, weight, and body temperature.

27
00:01:46,720 --> 00:01:53,240
The classifier would receive this data about the patient, process it, and fire out a confidence score.

28
00:01:53,240 --> 00:02:01,900
A high score would mean a high confidence that the patient is sick, and a low score would suggest that they are healthy.

29
00:02:01,900 --> 00:02:05,960
Neural nets are used for classification tasks where an object can fall

30
00:02:05,960 --> 00:02:09,080
into one of at least two different categories.

31
00:02:09,080 --> 00:02:11,820
Unlike other networks like a social network,

32
00:02:11,820 --> 00:02:15,120
a neural network is highly structured and comes in layers.

33
00:02:15,120 --> 00:02:17,300
The first layer is the input layer,

34
00:02:17,300 --> 00:02:19,320
the final layer is the output layer,

35
00:02:19,320 --> 00:02:23,040
and all layers in between are referred to as hidden layers.

36
00:02:23,040 --> 00:02:28,560
A neural net can be viewed as the result of spinning classifiers together in a layered web.

37
00:02:28,560 --> 00:02:33,920
This is because each node in the hidden and output layers has its own classifier.

38
00:02:33,920 --> 00:02:35,760
Take that node for example -

39
00:02:35,760 --> 00:02:39,440
it gets its inputs from the input layer, and activates.

40
00:02:39,440 --> 00:02:45,040
Its score is then passed on as input to the next hidden layer for further activation.

41
00:02:45,040 --> 00:02:45,940
So,

42
00:02:45,940 --> 00:02:50,100
let’s see how this plays out end to end across the entire network.

43
00:02:50,100 --> 00:02:53,020
A set of inputs is passed to the first hidden layer,

44
00:02:53,020 --> 00:02:56,720
the activations from that layer are passed to the next layer and so on,

45
00:02:56,720 --> 00:02:58,960
until you reach the output layer,

46
00:02:58,960 --> 00:03:04,480
where the results of the classification are determined by the scores at each node.

47
00:03:04,480 --> 00:03:07,120
This happens for each set of inputs.

48
00:03:07,120 --> 00:03:09,920
Here's another one...

49
00:03:09,920 --> 00:03:11,460
like so.

50
00:03:11,460 --> 00:03:16,500
This series of events starting from the input where each activation is sent to the next layer,

51
00:03:16,500 --> 00:03:18,840
and then the next, a

結果 (中文) 1:

100:00:02，840--> 00:00:07，040如果你忽略的神经网因为你觉得它们太难理解200:00:07，040--> 00:00:11，040或者你认为你不需要他们...男孩做我有治疗给你 ！300:00:11，040--> 00:00:16，080在这个视频中，您将了解到没有任何的数学或代码 — — 神经网络400:00:16，080--> 00:00:19，200只是它们是什么以及它们是如何工作简介。500:00:19，200--> 00:00:23，280我是希望你会明白了，为他们为什么这样一种重要的工具。600:00:23，280--> 00:00:24，560我们开始吧。700:00:24，560--> 00:00:25，920 800:00:25，920--> 00:00:30，360你需要知道的第一件事是深入学习神经网络。900:00:30，360--> 00:00:34，360神经网络的结构就像任何其他类型的网络;1000:00:34，360--> 00:00:38，160那里是相互的交织的节点上，这被称为神经元，1100:00:38，160--> 00:00:40，560和他们在一起的边缘。1200:00:40，560--> 00:00:44，500一种神经网络的主要功能是接收一组输入，1300:00:44，500--> 00:00:47，580执行逐步复杂的计算，1400:00:47，580--> 00:00:50，380然后使用输出来解决问题。1500:00:50，380--> 00:00:53，800神经网络用于很多不同的应用程序，1600:00:53，800--> 00:00:57，140但在本系列中我们将集中在分类。1700:00:57 140--> 00:01:01，520如果你想要了解神经网的更多细节，包括数学，1800:01:01，520--> 00:01:08，360我最喜欢的两个资源是迈克尔 · 尼尔森书和 Andrew Ng 类。1900:01:08，360--> 00:01:15，100我们讨论更多关于神经网络之前，我要给你的分类问题的快速概览。2000:01:15，100--> 00:01:19，000分类是对一组对象，进行分类的过程2100:01:19，000--> 00:01:23，160同时只使用描述他们一些基本的数据特性。2200:01:23，160--> 00:01:26，040现在有了许多的量词可用-2300:01:26，040--> 00:01:32，340如 Logistic 回归分析，支持向量机，朴素贝叶斯和当然，神经网络。2400:01:32，340--> 00:01:38，240射击的分类器或作为其俗称，激活产生一个分数。2500:01:38，240--> 00:01:42，400例如，假设您需要预测是否病人是生病或是健康、2600:01:42，400--> 00:01:46，720和所有你的身高，体重和身体的温度。2700:01:46，720--> 00:01:53，240分类器将收到关于病人的资料、 处理，并火出置信度得分。2800:01:53，240--> 00:02:01，900很高的分数将意味着病人有病，和很低的分数会建议他们健康高置信度。2900:02:01，900--> 00:02:05，960神经网络用于分类任务对象可以落在哪里3000:02:05，960--> 00:02:09，080至少两个不同的类别之一。3100:02:09，080--> 00:02:11，820与其他网络像一个社交网络，不同3200:02:11，820--> 00:02:15，120神经网络高度结构化和图层。3300:02:15，120--> 00:02:17，300第一层是输入的层，3400:02:17，300--> 00:02:19，320最后一层是输出层，3500:02:19，320--> 00:02:23，040和之间的所有图层被称为隐藏的图层。3600:02:23，040--> 00:02:28，560一个神经网络可以看作纺分类器的结果一起在分层的 web。3700:02:28，560--> 00:02:33，920这是因为在隐藏和输出的图层中的每个节点有其自己的分类器。3800:02:33，920--> 00:02:35，760以该节点为例 — —3900:02:35，760--> 00:02:39，440从输入层，获取其输入，并激活。4000:02:39，440--> 00:02:45，040其得分是然后作为输入传递给下一个隐藏层进一步激活。4100:02:45，040--> 00:02:45，940所以，4200:02:45，940--> 00:02:50，100让我们看看如何这扮演了端到端在整个网络。4300:02:50，100--> 00:02:53，020一组输入传递给第一隐藏层，4400:02:53，020--> 00:02:56，720从该图层激活被传递到下一层，等等4500:02:56，720--> 00:02:58，960直到你到达输出层，4600:02:58，960--> 00:03:04，480在分类的结果由在每个节点的分数决定。4700:03:04，480--> 00:03:07，120每个输入集的出现这种情况。4800:03:07，120--> 00:03:09，920这里是另一个...4900:03:09，920--> 00:03:11，460就像这样。5000:03:11，460--> 00:03:16，500这一系列的事件从输入开始到下一层，发送每个激活的位置5100:03:16，500--> 00:03:18，840然后下一步，

結果 (中文) 3:

102840 00：00：00：00 - >：07040如果你认为你已经ignoring神经网队太硬，因为他们了解207040 00：00：00：00 - >：11040或者你认为你不...boy need them for You have a给你治疗。311040 00：00：00：00 - >：16080在这个视频你learn about without any of the数学神经网队或队列。416080 00：00：00：00 - >：19200只是什么是安了他们和他们的工作技能。500：00：00：00 - > 19200：23280我希望你知道is that get an an important for为什么他们这样的工具。623280 00：00：00：00 - >：24560让我们开始。724560 00：00：00：00 - >：259208：：00 00 00 00 25920—>：：30360第一件事，你知道the need to is that is about深学习神经网络。930360 00：00：00：00 - >：34360神经网络结构of a kind of network is like any other；1034360 00：00：00：00 - >：38160there is an interconnected Web of which are called nodes，神经元，1138160 00：00：00：00 - >：40560join them，and the边在一起。1240560 00：00：00：00 - >：44500在main函数的神经network' to receive is of a集输入，1344500 00：00：00：00 - >：47580progressively做复杂的计算。1447580 00：00：00：00 - >：50380然后使用to the输出问题的解决。1550380 00：00：00：00 - >：800神经网络是used for很多of different applications，16：：00 00 00 00 800—>：：57140但在这个系列我们将重点分类。1700：00：00 - > 01 57140：：01520如果你想learn about detail位神经网队在黑莓，including the Math，1801520：：00 01 01 00 - >：：08360我的两nielsen'图书资源的宠儿是迈克尔和安德鲁，ng' S级。1908360：：00 01 01 00 - >：：15 100在我们谈话关于神经网络的黑莓，你快给我裙子overview of the of分类问题。20：：15 100 00 01 01 00 - >：：19000categorizing is the process of Classification of a组对象，2100：00 - > 01：19000 23160：01：while using only that describe them some basic features的日期。2223160：：00 01 01 00 - >：：26040there are of available today - classifiers地段2326040：：00 01 01 00 - >：：32340Logistic回归支持向量机类，天真贝叶斯，神经网络，和大学课程，。2432340：：00 01 01 00 - >：：38240the烧of a as its commonly分类器，或所谓的活化，在produces评分。2538240：：00 01 01 00 - >：：42 400for example，if You needed to say is predict病人生病或健康，26：：00 01 01 00 42 400—>：：46720你已经是他们的身高和体重，体，和温度。2746720：：00 01 01 00 - >：：53240分类器会receive this about the process the病人日，恩，与火了信心评分。2800：00 - > 01：53240发现自己：：02我在高高的信心评分均值that is the病人生病，蓝晶石和低分数，他们会是健康的。2902 00：00 - >：发现自己：：05960 02神经分类任务在网队are used for an object can…3005960 02 00：00 - >：：：09080 02into one of至少两个不同的分类。3109080 02 00：00 - >：：：11820 02像其他的社交网络于，3202 00：00 - >：：11820呢02在神经网络的结构和高度，是在层。33：：02 00 00 02呢——>：：不the first is the层输入层、3402 00：00 - >：：不：19320 02is the the最终层输出层，3502 00：00 - >：：19320：23040 02在所有的层和隐层之间are as to。3623040 02 00：00 - >：：：28560 02神经网络在can be viewed as the result of a Web层classifiers纺在一起。3728560 02 00：00 - >：：：33920 02这是因为茶叶中的每个节点输出层和隐有自己的分类器。3833920 02 00：00 - >：：：35760 02take that for example -节点3935760 02 00：00 - >：：：39440 02它从输入其输入层，和activates。4039440 02 00：00 - >：：：45040 02它是由当时的输入值passed as to the next for further隐层激活。4145040 02 00：00 - >：：：45940 02我知道，42：：45940 02 00 00 02—>：：50让我们看到了本次how the全部在网络端到端。4300：00 - > 50 02：：：53020 02在集合的输入passed of is to the第一隐层，4453020 02 00：00 - >：：：56720 02激活层passed from that are the next to the层在线和我知道，4556720 02 00：00 - >：：：58960 02直到你达到在输出层。4658960 02 00：00 - >：：：04480 03where the results of the Classification by the scores at each是矛盾的节点。4704480 03 00：00 - >：：：07120 03for each of this能投入集。48：：07120 03 00 00 03—>：：你知道，对吧这是另一个…4903 00：00 - >：：你知道，对吧11460 03：我知道。5011460 03 00：00 - >：：：16500 03Starting from the events of this series is sent to each激活输入在下一层，5116500 03 00：00 - >：：：18840 03然后我在the next，