17 - 5 - Online Learning (13 min).srt

1
00:00:00,109 --> 00:00:02,030
In this video, I'd like to talk
在这个视频中 我将会

2
00:00:02,030 --> 00:00:03,738
about a new large-scale
讨论一种新的大规模的

3
00:00:03,738 --> 00:00:05,369
machine learning setting called
机器学习机制 叫做

4
00:00:05,369 --> 00:00:07,073
the online learning setting.
在线学习机制

5
00:00:07,442 --> 00:00:08,731
The online learning setting
在线学习机制

6
00:00:08,731 --> 00:00:10,659
allows us to model problems
让我们可以模型化问题

7
00:00:10,659 --> 00:00:12,074
where we have a continuous flood
在拥有连续一波数据

8
00:00:12,074 --> 00:00:14,064
or a continuous stream of data
或连续的数据流涌进来

9
00:00:14,064 --> 00:00:15,906
coming in and we would like
而我们又需要

10
00:00:15,906 --> 00:00:17,839
an algorithm to learn from that.
一个算法来从中学习的时候来模型化问题

11
00:00:18,762 --> 00:00:20,759
Today, many of the largest
今天 许多大型网站

12
00:00:20,759 --> 00:00:22,245
websites, or many of the largest
或者许多大型网络公司

13
00:00:22,245 --> 00:00:24,335
website companies use different
使用不同版本的

14
00:00:24,335 --> 00:00:25,901
versions of online learning
在线学习机制算法

15
00:00:25,901 --> 00:00:28,102
algorithms to learn from
从大批的涌入

16
00:00:28,117 --> 00:00:29,468
the flood of users that keep
又离开网站的用户身上

17
00:00:29,468 --> 00:00:31,370
on coming to, back to the website.
进行学习

18
00:00:31,370 --> 00:00:32,943
Specifically, if you have
特别要提及的是 如果你有

19
00:00:32,943 --> 00:00:34,992
a continuous stream of data
一个由连续的用户流引发的

20
00:00:34,992 --> 00:00:36,371
generated by a continuous
连续的数据流

21
00:00:36,371 --> 00:00:37,703
stream of users coming to
进入

22
00:00:37,703 --> 00:00:39,413
your website, what you can
你的网站

23
00:00:39,413 --> 00:00:40,844
do is sometimes use an
你能做的是使用一个

24
00:00:40,844 --> 00:00:42,632
online learning algorithm to learn
在线学习机制 从数据流中学习

25
00:00:42,632 --> 00:00:44,492
user preferences from the
用户的偏好

26
00:00:44,492 --> 00:00:46,324
stream of data and use that
然后使用这些信息

27
00:00:46,324 --> 00:00:47,470
to optimize some of the
来优化一些

28
00:00:47,470 --> 00:00:49,632
decisions on your website.
关于网站的决策

29
00:00:52,063 --> 00:00:54,506
Suppose you run a shipping service,
假定你有一个提供运输服务的公司

30
00:00:54,506 --> 00:00:56,163
so, you know, users come and ask
所以你知道 用户们来向你询问

31
00:00:56,163 --> 00:00:57,307
you to help ship their package from
把包裹从A地

32
00:00:57,307 --> 00:01:01,533
location A to location B and suppose
运到B地的服务

33
00:01:01,533 --> 00:01:02,717
you run a website, where users
同时假定你有一个网站

34
00:01:02,717 --> 00:01:04,110
repeatedly come and they
让用户们可多次登陆

35
00:01:04,110 --> 00:01:05,689
tell you where they want
然后他们告诉你

36
00:01:05,689 --> 00:01:07,291
to send the package from, and
他们想从哪里寄出包裹 以及

37
00:01:07,291 --> 00:01:08,523
where they want to send it to
包裹要寄到哪里去

38
00:01:08,523 --> 00:01:10,947
(so the origin and destination) and
也就是出发地与目的地

39
00:01:10,947 --> 00:01:12,748
your website offers to ship the package
然后你的网站开出运输包裹的

40
00:01:12,748 --> 00:01:14,515
for some asking price,
的服务价格

41
00:01:14,515 --> 00:01:16,092
so I'll ship your package for $50,
比如 我会收取$50来运输你的包裹

42
00:01:16,092 --> 00:01:17,926
I'll ship it for $20.
我会收取$20之类的

43
00:01:17,926 --> 00:01:19,343
And based on the price
然后根据

44
00:01:19,343 --> 00:01:20,922
that you offer to the users,
你开给用户的这个价格

45
00:01:20,922 --> 00:01:23,522
the users sometimes chose to use a shipping service;
用户有时会接受这个运输服务

46
00:01:23,522 --> 00:01:25,891
that's a positive example and
那么这就是个正样本

47
00:01:25,891 --> 00:01:28,168
sometimes they go away and
有时他们会走掉

48
00:01:28,168 --> 00:01:29,722
they do not choose to
然后他们拒绝

49
00:01:29,722 --> 00:01:31,719
purchase your shipping service.
购买你的运输服务

50
00:01:31,719 --> 00:01:34,552
So let's say that we want
所以 让我们假定我们想要一个

51
00:01:34,552 --> 00:01:36,386
a learning algorithm to help us
学习算法来帮助我们

52
00:01:36,386 --> 00:01:38,499
to optimize what is the asking
优化我们想给用户

53
00:01:38,499 --> 00:01:41,680
price that we want to offer to our users.
开出的价格

54
00:01:41,680 --> 00:01:43,724
And specifically, let's say we
而且特别的是 我们假定

55
00:01:43,724 --> 00:01:44,908
come up with some sort of features
我们找到了一些

56
00:01:44,908 --> 00:01:46,510
that capture properties of the users.
获取用户特点的方法

57
00:01:46,510 --> 00:01:49,376
If we know anything about the demographics,
如果我们知道一些用户的统计信息

58
00:01:49,376 --> 00:01:50,875
they capture, you know, the origin and
它们会获取 比如 包裹的起始地

59
00:01:50,875 --> 00:01:54,405
destination of the package, where they want to ship the package.
以及目的地 他们想把包裹运到哪里去

60
00:01:54,405 --> 00:01:55,635
And what is the price
以及我们提供给他们的

61
00:01:55,635 --> 00:01:57,911
that we offer to them for shipping the package.
运送包裹的价格

62
00:01:57,911 --> 00:01:59,931
and what we want to do
我们想要做的就是

63
00:01:59,931 --> 00:02:00,883
is learn what is the
学习

64
00:02:00,883 --> 00:02:02,439
probability that they will
在给出的价格下他们将会

65
00:02:02,439 --> 00:02:03,762
elect to ship the
选择

66
00:02:03,762 --> 00:02:05,457
package, using our
运输包裹的几率

67
00:02:05,457 --> 00:02:07,315
shipping service given these features, and
在已知用户特点的前提下 并且

68
00:02:07,315 --> 00:02:10,197
again just as a reminder these
我要再次指出

69
00:02:10,197 --> 00:02:14,121
features X also captures the price that we're asking for.
他们也同时获取了我们开出的价格

70
00:02:14,121 --> 00:02:15,790
And so if we could
所以如果我们可以

71
00:02:15,790 --> 00:02:17,486
estimate the chance that they'll
估计出用户选择

72
00:02:17,486 --> 00:02:19,629
agree to use our service
使用我们的服务时

73
00:02:19,629 --> 00:02:20,962
for any given price, then we
我们所开出的价格 那么我们

74
00:02:20,962 --> 00:02:21,967
can try to pick
可以试着去选择

75
00:02:21,967 --> 00:02:23,183
a price so that they
一个优化的价格 因而在这个价格下

76
00:02:23,183 --> 00:02:25,125
have a pretty high probability of
用户会有很大的可能性

77
00:02:25,125 --> 00:02:27,841
choosing our website while simultaneously
选择我们的网站

78
00:02:27,841 --> 00:02:29,188
hopefully offering us a
而且同时很有可能会提供给我们

79
00:02:29,188 --> 00:02:31,371
fair return, offering us
一个合适的回报 让我们

80
00:02:31,371 --> 00:02:34,293
a fair profit for shipping their package.
在提供运输服务时也能获得合适的利润

81
00:02:34,585 --> 00:02:36,489
So if we can learn this property
所以如果我们可以学习

82
00:02:36,489 --> 00:02:37,733
of y equals 1 given
y 等于 1 时的条件

83
00:02:37,733 --> 00:02:38,632
any price and given the other
在任何给定价格以及其他给定的

84
00:02:38,632 --> 00:02:39,660
features we could really
条件下y等于1的特征

85
00:02:39,660 --> 00:02:41,657
use this to choose appropriate
我们就真的可以利用这一些信息

86
00:02:41,657 --> 00:02:44,072
prices as new users come to us.
在新用户来的时候选择合适的价格

87
00:02:44,072 --> 00:02:45,907
So in order to model
所以为了

88
00:02:45,907 --> 00:02:47,277
the probability of y equals 1,
获得 y 等于 1 的概率的模型

89
00:02:47,277 --> 00:02:48,972
what we can do is use
我们能做的就是

90
00:02:48,972 --> 00:02:51,781
logistic regression or neural
用逻辑回归或者神经网络

91
00:02:51,781 --> 00:02:53,756
network or some other algorithm like that.
或者其他一些类似的算法

92
00:02:53,756 --> 00:02:55,889
But let's start with logistic regression.
但现在我们先来考虑逻辑回归

93
00:02:57,658 --> 00:02:59,583
Now if you have a
现在假定你有一个

94
00:02:59,583 --> 00:03:01,835
website that just runs continuously,
连续运行的网站

95
00:03:01,835 --> 00:03:05,342
here's what an online learning algorithm would do.
以下就是在线学习算法要做的

96
00:03:05,342 --> 00:03:07,478
I'm gonna write repeat forever.
我要写下"一直重复"

97
00:03:07,478 --> 00:03:09,730
This just means that our website
这只是代表着我们的网站

98
00:03:09,730 --> 00:03:11,170
is going to, you know, keep on
将会一直继续

99
00:03:11,170 --> 00:03:12,911
staying up.
保持在线学习

100
00:03:12,911 --> 00:03:14,351
What happens on the website is
这个网站将要发生的是

101
00:03:14,351 --> 00:03:16,465
occasionally a user
一个用户

102
00:03:16,465 --> 00:03:17,950
will come and for
偶然访问

103
00:03:17,950 --> 00:03:19,576
the user that comes we'll get
然后我们将会得到

104
00:03:19,576 --> 00:03:25,380
some x,y pair corresponding to
与其对应的一些(x,y)对

105
00:03:25,380 --> 00:03:29,096
a customer or to a user on the website.
这些(x,y)对是相对应于一个特定的客户或用户的

106
00:03:29,096 --> 00:03:30,884
So the features x are, you
所以特征 x 是指

107
00:03:30,884 --> 00:03:32,811
know, the origin and destination specified
客户所指定的起始地与目的地

108
00:03:32,811 --> 00:03:34,111
by this user and the price
以及

109
00:03:34,111 --> 00:03:35,358
that we happened to offer to
我们这一次提供

110
00:03:35,358 --> 00:03:37,292
them this time around, and
给客户的价格

111
00:03:37,292 --> 00:03:38,430
y is either one or
而y则取1或0

112
00:03:38,430 --> 00:03:40,148
zero depending one whether or
y值取决于

113
00:03:40,148 --> 00:03:41,518
not they chose to
客户是否选择了

114
00:03:41,518 --> 00:03:43,980
use our shipping service.
使用我们的运输服务

115
00:03:43,980 --> 00:03:45,419
Now once we get this {x,y}
现在我们一旦获得了这个{x,y}数据对

116
00:03:45,419 --> 00:03:46,813
pair, what an online
在线学习算法

117
00:03:46,813 --> 00:03:48,391
learning algorithm does is then
要做的就是

118
00:03:48,391 --> 00:03:50,690
update the parameters theta
更新参数θ

119
00:03:50,690 --> 00:03:54,011
using just this example
利用刚得到的(x,y)数据对来更新θ

120
00:03:54,011 --> 00:03:57,726
x,y, and in particular
具体来说

121
00:03:57,726 --> 00:03:59,839
we would update my parameters theta
我们将这样更新我们的参数θ

122
00:03:59,839 --> 00:04:01,842
as Theta j get updated as Theta j
θj 将会被更新为

123
00:04:01,842 --> 00:04:06,619
minus the learning rate alpha times
θj 减去学习率 α 乘以

124
00:04:06,619 --> 00:04:11,356
my usual gradient descent
梯度下降

125
00:04:11,356 --> 00:04:13,399
rule for logistic regression.
来做逻辑回归

126
00:04:13,399 --> 00:04:14,491
So we do this for j
然后我们对j等于0到n

127
00:04:14,491 --> 00:04:15,652
equals zero up to n,
重复这个步骤

128
00:04:15,652 --> 00:04:19,088
and that's my close curly brace.
这是我的另一边花括号

129
00:04:19,088 --> 00:04:21,218
So, for other learning algorithms
所以对于其他的学习算法

130
00:04:21,218 --> 00:04:22,873
instead of writing X-Y, right, I
不是写(x,y)对 对吧

131
00:04:22,873 --> 00:04:24,011
was writing things like Xi,
我之前写的是

132
00:04:24,011 --> 00:04:26,495
Yi but
(x(i),y(i)) 一样的数据对

133
00:04:26,495 --> 00:04:27,842
in this online learning setting
但在这个在线学习机制中

134
00:04:27,842 --> 00:04:29,723
where actually discarding the notion
我们实际上丢弃了

135
00:04:29,723 --> 00:04:31,464
of there being a fixed training
获取一个固定的数据集这样的概念

136
00:04:31,464 --> 00:04:32,904
set instead we have an algorithm.
取而代之的是 我们拥有一个算法

137
00:04:32,904 --> 00:04:34,924
Now what happens as we get
现在 当我们

138
00:04:34,924 --> 00:04:37,014
an example and then we
获取一个样本 然后我们

139
00:04:37,014 --> 00:04:38,825
learn using that example like
利用那个样本获取信息学习

140
00:04:38,825 --> 00:04:41,031
so and then we throw that example away.
然后我们丢弃这个样本

141
00:04:41,031 --> 00:04:43,098
We discard that example and we
我们丢弃那个样本 而且我们

142
00:04:43,098 --> 00:04:45,141
never use it again and
永远不会再使用它

143
00:04:45,141 --> 00:04:47,161
so that's why we just look at one example at a time.
这就是为什么我们在一个时间点只会处理一个样本的原因

144
00:04:47,161 --> 00:04:48,879
We learn from that example.
我们从样本中学习

145
00:04:48,879 --> 00:04:50,412
We discard it.
我们再丢弃它

146
00:04:50,412 --> 00:04:51,527
Which is why, you know, we're
这也就是为什么

147
00:04:51,527 --> 00:04:52,943
also doing away with this
我们放弃了一种拥有

148
00:04:52,943 --> 00:04:54,615
notion of there being this
我们放弃了一种拥有

149
00:04:54,615 --> 00:04:58,191
sort of fixed training set indexed by i.
固定的 由 i 来作参数的数据集的表示方法

150
00:04:58,191 --> 00:04:59,328
And, if you really run
而且 如果你真的运行

151
00:04:59,328 --> 00:05:01,488
a major website where you
一个大型网站

152
00:05:01,488 --> 00:05:03,624
really have a continuous stream
在这个网站里你有一个连续的

153
00:05:03,624 --> 00:05:05,737
of users coming, then this
用户流登陆网站 那么

154
00:05:05,737 --> 00:05:07,525
sort of online learning algorithm
这种在线学习算法

155
00:05:07,525 --> 00:05:10,358
is actually a pretty reasonable algorithm.
是一种非常合理的算法

156
00:05:10,358 --> 00:05:12,076
Because of data is essentially
因为数据本质上是自由的

157
00:05:12,076 --> 00:05:13,330
free if you have so
如果你有如此多的数据

158
00:05:13,330 --> 00:05:14,979
much data, that data
而数据

159
00:05:14,979 --> 00:05:17,022
is essentially unlimited then there
本质上是无限的 那么

160
00:05:17,022 --> 00:05:17,997
is really may be no
或许就真的没必要

161
00:05:17,997 --> 00:05:18,949
need to look at a
重复处理

162
00:05:18,949 --> 00:05:21,527
training example more than once.
一个样本

163
00:05:21,527 --> 00:05:22,432
Of course if we had only
当然 如果我们只有

164
00:05:22,432 --> 00:05:24,220
a small number of users then
少量的用户

165
00:05:24,220 --> 00:05:26,333
rather than using an online learning
那么我们就不选择像这样的在线学习算法

166
00:05:26,333 --> 00:05:27,912
algorithm like this, you might
你可能最好是要

167
00:05:27,912 --> 00:05:29,421
be better off saving away all
保存好所有的

168
00:05:29,421 --> 00:05:30,884
your data in a fixed training
数据 保存在一个固定的

169
00:05:30,884 --> 00:05:34,042
set and then running some algorithm over that training set.
数据集里 然后对这个数据集使用某种算法

170
00:05:34,042 --> 00:05:35,018
But if you really have a continuous
但是 如果你确实有一个连续的

171
00:05:35,018 --> 00:05:36,341
stream of data, then an
数据流 那么一个

172
00:05:36,341 --> 00:05:39,881
online learning algorithm can be very effective.
在线学习机制会非常的有效

173
00:05:39,881 --> 00:05:41,171
I should mention also that one
我也必须要提到一个

174
00:05:41,171 --> 00:05:43,015
interesting effect of this sort
这种在线学习算法

175
00:05:43,015 --> 00:05:44,073
of online learning algorithm is
会带来的有趣的效果 那就是

176
00:05:44,073 --> 00:05:49,391
that it can adapt to changing user preferences.
它可以对正在变化的用户偏好进行调适

177
00:05:51,006 --> 00:05:54,592
And in particular, if over
而且特别的 如果

178
00:05:54,592 --> 00:05:55,776
time because of changes in
随着时间变化 因为

179
00:05:55,776 --> 00:05:58,377
the economy maybe users
大的经济环境发生变化 用户们可能会

180
00:05:58,377 --> 00:05:59,957
start to become more price
开始变得对价格更敏感

181
00:05:59,957 --> 00:06:01,395
sensitive and willing to pay,
然后愿意支付

182
00:06:01,395 --> 00:06:03,717
you know, less willing to pay high prices.
你知道的 不那么愿意支付高的费用

183
00:06:03,717 --> 00:06:06,527
Or if they become less price sensitive and they're willing to pay higher prices.
也有可能他们变得对价格不那么敏感 然后他们愿意支付更高的价格

184
00:06:06,527 --> 00:06:08,292
Or if different things
又或者各种因素

185
00:06:08,292 --> 00:06:10,451
become more important to users,
变得对用户的影响更大了

186
00:06:10,451 --> 00:06:11,496
if you start to have new
如果你开始拥有

187
00:06:11,496 --> 00:06:12,587
types of users coming to your website.
某一种新的类型的用户涌入你的网站

188
00:06:12,587 --> 00:06:14,933
This sort of online learning algorithm
这样的在线学习算法

189
00:06:14,933 --> 00:06:17,278
can also adapt to changing
也可以根据变化着的

190
00:06:17,278 --> 00:06:18,950
user preferences and kind
用户偏好进行调适

191
00:06:18,950 --> 00:06:20,157
of keep track of what your
而且从某种程度上可以跟进

192
00:06:20,157 --> 00:06:21,991
changing population of users
变化着的用户群体所愿意

193
00:06:21,991 --> 00:06:24,685
may be willing to pay for.
支付的价格

194
00:06:24,685 --> 00:06:26,171
And it does that because if
而且 在线学习算法有这样的作用是因为

195
00:06:26,171 --> 00:06:28,168
your pool of users changes,
如果你的用户群变化了

196
00:06:28,168 --> 00:06:29,793
then these updates to your
那么参数θ的变化与更新

197
00:06:29,793 --> 00:06:31,953
parameters theta will just slowly adapt
会逐渐调适到

198
00:06:31,953 --> 00:06:33,555
your parameters to whatever your
你最新的用户群所应该体现出来的

199
00:06:33,555 --> 00:06:36,599
latest pool of users looks like.
参数

200
00:06:36,599 --> 00:06:37,781
Here's another example of a
这里有另一个

201
00:06:37,781 --> 00:06:40,753
sort of application to which you might apply online learning.
你可能会想要使用在线学习的例子

202
00:06:40,753 --> 00:06:43,472
this is an application in product
这是一个对于产品搜索的应用

203
00:06:43,472 --> 00:06:44,701
search in which we want to
在这个应用中 我们想要

204
00:06:44,701 --> 00:06:46,117
apply learning algorithm to learn
使用一种学习机制来学习如何

205
00:06:46,117 --> 00:06:48,973
to give good search listings to a user.
反馈给用户好的搜索列表

206
00:06:48,973 --> 00:06:51,156
Let's say you run an online
举个例子说 你有一个在线

207
00:06:51,156 --> 00:06:53,083
store that sells phones - that
卖电话的商铺

208
00:06:53,083 --> 00:06:55,312
sells mobile phones or sells cell phones.
一个卖移动电话或者手机的商铺

209
00:06:55,312 --> 00:06:56,682
And you have a user interface
而且你有一个用户界面

210
00:06:56,682 --> 00:06:58,284
where a user can come to
可以让用户登陆你的网站

211
00:06:58,284 --> 00:06:59,445
your website and type in the
并且键入一个

212
00:06:59,445 --> 00:07:02,626
query like "Android phone 1080p camera".
搜索条目 例如“安卓 手机 1080p 摄像头”

213
00:07:02,626 --> 00:07:03,509
So 1080p is a type
那么1080p 是指一个

214
00:07:03,509 --> 00:07:04,623
of a specification for a
对应于摄像头的

215
00:07:04,623 --> 00:07:05,808
video camera that you might
手机参数 这个参数可以出现在

216
00:07:05,808 --> 00:07:08,710
have on a phone, a cell phone, a mobile phone.
一部电话中 一个移动电话 或者一个手机中

217
00:07:08,710 --> 00:07:12,100
Suppose, suppose we have a hundred phones in our store.
假定 假定我们的商铺中有一百部电话

218
00:07:12,100 --> 00:07:13,354
And because of the way our
而且出于我们的网站设计

219
00:07:13,354 --> 00:07:15,321
website is laid out, when
当一个用户

220
00:07:15,321 --> 00:07:16,558
a user types in a query,
键入一个命令

221
00:07:16,558 --> 00:07:18,277
if it was a search query, we
如果这是一个搜索命令

222
00:07:18,277 --> 00:07:19,601
would like to find a
我们会想要找到一个

223
00:07:19,601 --> 00:07:20,900
choice of ten different phones to
合适的十部不同手机的列表

224
00:07:20,900 --> 00:07:22,921
show what to offer to the user.
来提供给用户

225
00:07:22,921 --> 00:07:24,987
What we'd like to do is have
我们想要做的是

226
00:07:24,987 --> 00:07:26,566
a learning algorithm help us figure
拥有一个在线学习机制来帮助我们

227
00:07:26,566 --> 00:07:28,447
out what are the ten phones
找到在这100部手机中

228
00:07:28,447 --> 00:07:29,771
out of the 100 we
哪十部手机

229
00:07:29,771 --> 00:07:31,791
should return the user in response to
是我们真正应该反馈给用户的

230
00:07:31,791 --> 00:07:34,531
a user-search query like the one here.
而且这个返回的列表是对类似这样的用户搜索条目最佳的回应

231
00:07:34,531 --> 00:07:36,695
Here's how we can go about the problem.
接下来要说的是一种解决问题的思路

232
00:07:37,218 --> 00:07:39,291
For each phone and given
对于每一个手机以及一个给定的

233
00:07:39,291 --> 00:07:41,311
a specific user query; we
用户搜索命令 我们

234
00:07:41,311 --> 00:07:44,120
can construct a feature vector
可以构建一个

235
00:07:44,120 --> 00:07:45,676
X. So the feature
特征矢量x 那么这个特征矢量x

236
00:07:45,676 --> 00:07:47,650
vector X might capture different properties of the phone.
可能会抓取手机的各种特点

237
00:07:47,650 --> 00:07:49,972
It might capture things like,
它可能会抓取类似于

238
00:07:49,972 --> 00:07:53,107
how similar the user search query is in the phones.
用户搜索命令与这部电话的类似程度有多高这样的信息

239
00:07:53,107 --> 00:07:54,059
We capture things like how many
我们获取类似于

240
00:07:54,059 --> 00:07:55,475
words in the user search
这个用户搜索命令中有多少个词

241
00:07:55,475 --> 00:07:56,172
query match the name of
可以与这部手机的名字相匹配

242
00:07:56,172 --> 00:07:57,356
the phone, how many words
或者这个搜索命令中有多少词

243
00:07:57,356 --> 00:08:01,303
in the user search query match the description of the phone and so on.
与这部手机的描述相匹配

244
00:08:01,303 --> 00:08:02,789
So the features x capture
所以特征矢量x获取

245
00:08:02,789 --> 00:08:03,672
properties of the phone and
手机的特点而且

246
00:08:03,672 --> 00:08:05,251
it captures things about how
它会获取

247
00:08:05,251 --> 00:08:06,412
similar or how well
这部手机与搜索命令

248
00:08:06,412 --> 00:08:10,591
the phone matches the user query along different dimensions.
的结果在各个方面的匹配程度

249
00:08:10,591 --> 00:08:11,868
What we like to do is
我们想要做的就是

250
00:08:11,868 --> 00:08:14,330
estimate the probability that a
估测一个概率 这个概率是指用户

251
00:08:14,330 --> 00:08:15,816
user will click on the
将会点进

252
00:08:15,816 --> 00:08:17,673
link for a specific phone,
某一个特定的手机的链接

253
00:08:17,673 --> 00:08:18,881
because we want to show
因为我们想要给用户展示

254
00:08:18,881 --> 00:08:20,065
the user phones that they
他们

255
00:08:20,065 --> 00:08:21,481
are likely to want to
想要买的手机

256
00:08:21,481 --> 00:08:22,921
buy, want to show the user
我们想要给用户提供

257
00:08:22,921 --> 00:08:24,082
phones that they have high
那些他们很可能

258
00:08:24,082 --> 00:08:27,240
probability of clicking on in the web browser.
在浏览器中点进去查看的手机

259
00:08:27,240 --> 00:08:29,562
So I'm going to define y equals
所以我将定义y等于1时

260
00:08:29,562 --> 00:08:30,676
one if the user clicks on
是指用户点击了

261
00:08:30,676 --> 00:08:31,930
the link for a phone and
手机的链接

262
00:08:31,930 --> 00:08:34,136
y equals zero otherwise and
而y等于0是指用户没有点击链接

263
00:08:34,136 --> 00:08:35,454
what I would like to do is
然后我们想要做的就是

264
00:08:35,454 --> 00:08:36,992
learn the probability the user
学习到用户

265
00:08:36,992 --> 00:08:38,246
will click on a specific
将会点击某一个背给出的特定的手机的概率

266
00:08:38,246 --> 00:08:39,802
phone given, you know,
你知道的

267
00:08:39,802 --> 00:08:41,693
the features x, which capture properties
特征X 获取了手机的特点

268
00:08:41,693 --> 00:08:43,819
of the phone and how well the query matches the phone.
以及搜索条目与手机的匹配程度

269
00:08:43,819 --> 00:08:45,700
To give this problem a name
如果要给这个问题命一个名

270
00:08:45,700 --> 00:08:47,720
in the language of
用一种运行这类网站的人们

271
00:08:47,720 --> 00:08:49,130
people that run websites like
所使用的语言来命名

272
00:08:49,130 --> 00:08:51,249
this, the problem of learning this is
这类学习问题

273
00:08:51,249 --> 00:08:53,223
actually called the problem of
这类问题其实被称作

274
00:08:53,223 --> 00:08:57,296
learning the predicted click-through rate, the predicted CTR.
学习预测的点击率 预估点击率CTR

275
00:08:57,296 --> 00:08:58,796
It just means learning the probability
它仅仅代表这学习

276
00:08:58,796 --> 00:09:00,491
that the user will click on
用户将点击某一个

277
00:09:00,491 --> 00:09:01,698
the specific link that you
特定的 你提供给他们的链接的概率

278
00:09:01,698 --> 00:09:03,022
offer them, so CTR is
所以CTR是

279
00:09:03,022 --> 00:09:06,528
an abbreviation for click through rate.
点击率(Click Through Rate)的简称

280
00:09:06,528 --> 00:09:07,550
And if you can estimate the
然后 如果你能够估计

281
00:09:07,550 --> 00:09:09,245
predicted click-through rate for any
任意一个特定手机的点击率

282
00:09:09,245 --> 00:09:10,847
particular phone, what we
我们可以做的就是

283
00:09:10,847 --> 00:09:12,171
can do is use this to
利用这个来

284
00:09:12,171 --> 00:09:13,819
show the user the ten phones
给用户展示十个

285
00:09:13,819 --> 00:09:15,770
that are most likely to click on,
他们最有可能点击的手机

286
00:09:15,770 --> 00:09:17,441
because out of the hundred phones,
因为从这一百个手机中

287
00:09:17,441 --> 00:09:20,553
we can compute this for
我们可以计算出

288
00:09:20,553 --> 00:09:21,737
each of the 100 phones and
100部手机中 每一部手机的可能的点击率

289
00:09:21,737 --> 00:09:22,759
just select the 10 phones
而且我们选择10部

290
00:09:22,759 --> 00:09:25,754
that the user is most likely to click on,
用户最有可能点击的手机

291
00:09:25,754 --> 00:09:26,892
and this will be a pretty reasonable
那么这就是一个非常合理的

292
00:09:26,892 --> 00:09:29,818
way to decide what ten results to show to the user.
来决定要展示给用户的十个搜索结果的方法

293
00:09:29,818 --> 00:09:32,186
Just to be clear, suppose that
更明确地说 假定

294
00:09:32,186 --> 00:09:33,440
every time a user does
每次用户

295
00:09:33,440 --> 00:09:35,576
a search, we return ten results
进行一次搜索 我们回馈给用户十个结果

296
00:09:35,576 --> 00:09:37,225
what that will do is it
在线学习算法会做的是

297
00:09:37,225 --> 00:09:38,990
will actually give us ten
它会真正地提供给我们十个

298
00:09:38,990 --> 00:09:40,870
x,y pairs, this actually
(x,y) 数据对 这就真的

299
00:09:40,870 --> 00:09:43,332
gives us ten training examples every
给了我们十个数据样本

300
00:09:43,332 --> 00:09:44,640
time a user comes to
每当一个用户来到

301
00:09:44,640 --> 00:09:46,257
our website because, because for
我们网站时就给了我们十个样本

302
00:09:46,257 --> 00:09:47,535
the ten phone that we chose
因为对于这十部我们选择

303
00:09:47,535 --> 00:09:48,881
to show the user, for each
要展示给用户的手机 对于

304
00:09:48,881 --> 00:09:49,896
of those 10 phones we get
这10部手机中的每一个 我们会得到

305
00:09:49,896 --> 00:09:51,389
a feature vector X, and
一个特征矢量x 而且

306
00:09:51,389 --> 00:09:52,737
for each of those 10 phones we
对于这10部手机中的任何一个手机

307
00:09:52,737 --> 00:09:54,563
show the user we will also
我们还会得到

308
00:09:54,563 --> 00:09:56,172
get a value for y, we
y的取值

309
00:09:56,172 --> 00:09:57,542
will also observe the value
我们也会观察这些取值

310
00:09:57,542 --> 00:09:59,517
of y, depending on whether
这些取值是根据

311
00:09:59,517 --> 00:10:00,925
or not we clicked on that
用户有没有点击

312
00:10:00,925 --> 00:10:02,465
url or not and
那个网页链接来决定的

313
00:10:02,465 --> 00:10:03,696
so, one way to run a
这样 运行此类网站的

314
00:10:03,696 --> 00:10:04,903
website like this would be to
一种方法就是

315
00:10:04,903 --> 00:10:06,830
continuously show the user,
连续给用户展示

316
00:10:06,830 --> 00:10:08,363
you know, your ten best guesses for
你的十个最佳猜测

317
00:10:08,363 --> 00:10:09,895
what other phones they might like
这十个推荐是指用户可能会喜欢的其他的手机

318
00:10:09,895 --> 00:10:11,428
and so, each time a user
那么 每次一个用户访问

319
00:10:11,428 --> 00:10:12,728
comes you would get ten
你将会得到十个

320
00:10:12,728 --> 00:10:14,493
examples, ten x,y pairs,
样本 十个(x,y) 数据对

321
00:10:14,493 --> 00:10:16,304
and then use an online
然后利用一个在线学习

322
00:10:16,304 --> 00:10:17,953
learning algorithm to update the
算法来更新你的参数

323
00:10:17,953 --> 00:10:20,182
parameters using essentially 10
更新过程中会对这十个样本利用10步

324
00:10:20,182 --> 00:10:21,691
steps of gradient descent on these
梯度下降法

325
00:10:21,691 --> 00:10:23,386
10 examples, and then
然后

326
00:10:23,386 --> 00:10:25,081
you can throw the data away, and
你可以丢弃你的数据了

327
00:10:25,081 --> 00:10:26,590
if you really have a continuous
如果你真的拥有一个连续的

328
00:10:26,590 --> 00:10:27,891
stream of users coming to
用户流进入

329
00:10:27,891 --> 00:10:29,354
your website, this would be
你的网站 这将会是

330
00:10:29,354 --> 00:10:31,095
a pretty reasonable way to learn
一个非常合理的学习方法

331
00:10:31,095 --> 00:10:32,395
parameters for your algorithm
来学习你的算法中的参数

332
00:10:32,395 --> 00:10:33,835
so as to show the ten phones
从而来给用户展示

333
00:10:33,835 --> 00:10:35,669
to your users that may
十部他们

334
00:10:35,669 --> 00:10:39,013
be most promising and the most likely to click on.
最有可能点击查看的手机

335
00:10:39,013 --> 00:10:40,151
So, this is a product search
所以 这是一个产品搜索问题

336
00:10:40,151 --> 00:10:41,498
problem or learning to rank
或者说是一个学习将手机排序

337
00:10:41,498 --> 00:10:44,214
phones, learning to search for phones example.
的问题 学习搜索手机的样例

338
00:10:44,214 --> 00:10:46,422
So, I'll quickly mention a few others.
接着 我会快速地提及一些其他的例子

339
00:10:46,422 --> 00:10:47,372
One is, if you have
其中一个例子是 如果你有

340
00:10:47,372 --> 00:10:48,231
a website and you're trying to
一个网站 你在尝试着

341
00:10:48,231 --> 00:10:49,439
decide, you know, what special
来决定 你知道的 你要给用户

342
00:10:49,439 --> 00:10:50,321
offer to show the user,
展示什么样的特别优惠

343
00:10:50,321 --> 00:10:53,154
this is very similar to phones,
这与手机那个例子非常类似

344
00:10:53,154 --> 00:10:54,710
or if you have a
或者你有一个

345
00:10:54,710 --> 00:10:58,216
website and you show different users different news articles.
网站 然后你想给不同的用户展示不同的新闻文章

346
00:10:58,216 --> 00:10:59,911
So, if you're a news aggregator
那么 如果你是一个新闻抓取网站

347
00:10:59,911 --> 00:11:01,374
website, then you can
那么你又可以

348
00:11:01,374 --> 00:11:02,303
again use a similar system to
使用一个类似的系统

349
00:11:02,303 --> 00:11:03,882
select, to show to
来选择 来展示给用户

350
00:11:03,882 --> 00:11:05,554
the user, you know, what
他们最有可能感兴趣的

351
00:11:05,554 --> 00:11:06,877
are the news articles that they
他们最有可能感兴趣的

352
00:11:06,877 --> 00:11:08,154
are most likely to be interested
新闻文章

353
00:11:08,154 --> 00:11:11,103
in and what are the news articles that they are most likely to click on.
以及那些他们最有可能点击的新闻文章

354
00:11:11,103 --> 00:11:13,495
Closely related to special offers, will we profit from recommendations.
与特别优惠所密切相关的是 我们将会从这些推荐中获利

355
00:11:13,495 --> 00:11:15,097
And in fact, if you have
而且实际上 如果你有

356
00:11:15,097 --> 00:11:17,953
a collaborative filtering system, you
一个协作过滤系统

357
00:11:17,953 --> 00:11:20,693
can even imagine a collaborative filtering
你可以想象到 一个协作过滤系统

358
00:11:20,693 --> 00:11:22,643
system giving you additional
可以给你更多的

359
00:11:22,643 --> 00:11:23,897
features to feed into a
特征 这些特征可以整合到

360
00:11:23,897 --> 00:11:25,732
logistic regression classifier to try
逻辑回归的分类器 从而可以尝试着

361
00:11:25,732 --> 00:11:28,100
to predict the click through
预测对于你可能推荐给用户的

362
00:11:28,100 --> 00:11:29,981
rate for different products that you might recommend to a user.
不同产品的点击率

363
00:11:29,981 --> 00:11:32,280
Of course, I should say that
当然 我需要说明的是

364
00:11:32,280 --> 00:11:34,207
any of these problems could also
这些问题中的任何一个都可以

365
00:11:34,207 --> 00:11:35,600
have been formulated as a
被归类到

366
00:11:35,600 --> 00:11:39,873
standard machine learning problem, where you have a fixed training set.
标准的 拥有一个固定的样本集的机器学习问题中

367
00:11:39,873 --> 00:11:40,894
Maybe, you can run your
或许 你可以运行一个

368
00:11:40,894 --> 00:11:41,823
website for a few days and
你自己的网站 尝试运行几天

369
00:11:41,823 --> 00:11:43,727
then save away a training set,
然后保存一个数据集

370
00:11:43,727 --> 00:11:44,842
a fixed training set, and run
一个固定的数据集 然后对其运行

371
00:11:44,842 --> 00:11:45,771
a learning algorithm on that.
一个学习算法

372
00:11:45,771 --> 00:11:48,696
But these are the actual
但是这些是实际的

373
00:11:48,696 --> 00:11:49,950
sorts of problems, where you do
问题 在这些问题里

374
00:11:49,950 --> 00:11:51,901
see large companies get so
你会看到大公司会获取

375
00:11:51,901 --> 00:11:53,712
much data, that there's really
如此多的数据 所以真的没有必要

376
00:11:53,712 --> 00:11:55,221
maybe no need to save away
来保存一个

377
00:11:55,221 --> 00:11:56,963
a fixed training set, but instead
固定的数据集 取而代之的是

378
00:11:56,963 --> 00:11:59,563
you can use an online learning algorithm to just learn continuously.
你可以使用一个在线学习算法来连续的学习

379
00:11:59,563 --> 00:12:04,091
from the data that users are generating on your website.
从这些用户不断产生的数据中来学习

380
00:12:05,183 --> 00:12:07,249
So, that was the online
所以 这就是在线学习机制

381
00:12:07,249 --> 00:12:08,990
learning setting and as we
然后就像我们所看到的

382
00:12:08,990 --> 00:12:10,616
saw, the algorithm that we apply to
我们所使用的这个算法

383
00:12:10,616 --> 00:12:12,357
it is really very similar
与随机梯度下降算法

384
00:12:12,357 --> 00:12:13,867
to this schotastic gradient descent
非常类似

385
00:12:13,867 --> 00:12:15,330
algorithm, only instead of
唯一的区别的是 我们不会

386
00:12:15,330 --> 00:12:16,871
scanning through a fixed
使用一个固定的数据集

387
00:12:16,871 --> 00:12:18,000
training set, we're instead getting
我们会做的是获取

388
00:12:18,000 --> 00:12:19,974
one example from a user,
一个用户样本

389
00:12:19,974 --> 00:12:21,290
learning from that example, then
从那个样本中学习 然后

390
00:12:21,290 --> 00:12:22,644
discarding it and moving on.
丢弃那个样本并继续下去

391
00:12:22,644 --> 00:12:25,593
And if you have a continuous
而且如果你对某一种应用有一个连续的

392
00:12:25,593 --> 00:12:26,777
stream of data for some application,
数据流

393
00:12:26,777 --> 00:12:28,356
this sort of algorithm may be
这样的算法可能会

394
00:12:28,356 --> 00:12:31,816
well worth considering for your application.
非常值得考虑

395
00:12:31,816 --> 00:12:33,952
And of course, one advantage of
当然 在线学习的一个优点

396
00:12:33,952 --> 00:12:36,128
online learning is also that
就是

397
00:12:36,128 --> 00:12:37,458
if you have a changing pool
如果你有一个变化的

398
00:12:37,458 --> 00:12:38,967
of users, or if the
用户群 又或者

399
00:12:38,967 --> 00:12:40,082
things you're trying to predict are
你在尝试预测的事情

400
00:12:40,082 --> 00:12:42,032
slowly changing like your user
在缓慢变化 就像你的用户的

401
00:12:42,032 --> 00:12:43,751
taste is slowly changing, the online
品味在缓慢变化 这个在线学习

402
00:12:43,751 --> 00:12:45,492
learning algorithm can slowly
算法可以慢慢地

403
00:12:45,492 --> 00:12:47,211
adapt your learned hypothesis to
调试你所学习到的假设

404
00:12:47,211 --> 00:12:49,161
whatever the latest sets of
将其调节更新到最新的

405
00:12:49,161 --> 00:12:51,161
user behaviors are like as well.
用户行为 (字幕整理：中国海洋大学 黄海广,haiguang2000@qq.com )