当前位置:网站首页>最大似然估计,散度,交叉熵
最大似然估计,散度,交叉熵
2022-07-03 05:45:00 【code bean】
如果这个信息,可以将之前非常不确定的事情,确定了,说明这个信息的信息量很大!

这张图是解释,数学家如何定义信息量的过程:

也就是说,阿根廷夺冠的信息量 = 阿根廷进决赛的信息量+阿根廷赢了决赛的信息量,但是x本身是概率,所以这里的1/8 是 x1*x2 的结果。及:
![]()
这种特性是log特有的。 然后底数大于1的时候,log单调递增,所以加上负号让其单独递减。
到此信息量的定义完成了。
接下看信息熵的定义:

求信息熵的过程就是求期望的过程:信息量乘以事件发生的概率,然后求和。

信息熵,描述的是“消息”的不确定性程度或者说是混乱程度。而这里的“消息”是可以看成一个概率模型的。而求概率模型的信息熵,就是求概率模型的期望啊!
如果想对比两个模型的分布差异,就需要一个概念叫做相对熵(KL散度)


这里有个证明,交叉熵一定大于熵:
所以如果要两个模型最接近,那就是求交叉熵的最小值。那深度学习干的事情就是让神经网络这个模型逼近人脑模型。所以神经网络和人脑模型的交叉熵,就可以作为损失函数!(交叉熵越小,两个模型越相近)
那如下函数,其实是,一个二分类的交叉熵的展开:

这里Y是人的判定,只有两种可能,1或者0,而yhead,是神经网络的输出值,是个概率。他们之间形成的交叉熵构成了二分类的损失函数。

参考:
边栏推荐
- Today, many CTOs were killed because they didn't achieve business
- chromedriver对应版本下载
- Azure file synchronization of altaro: the end of traditional file servers?
- Communication - how to be a good listener?
- Sophomore dilemma (resumption)
- [Shangshui Shuo series together] day 10
- Final review (Day7)
- 今天很多 CTO 都是被干掉的,因为他没有成就业务
- [teacher Zhao Yuqiang] MySQL high availability architecture: MHA
- 2022.DAY592
猜你喜欢

Notepad++ wrap by specified character

Apt update and apt upgrade commands - what is the difference?
![[set theory] relational closure (reflexive closure | symmetric closure | transitive closure)](/img/c8/2995c503e9dabae4e2cc704449e04f.jpg)
[set theory] relational closure (reflexive closure | symmetric closure | transitive closure)

中职网络子网划分例题解析

一起上水硕系列】Day 9
![[teacher Zhao Yuqiang] Alibaba cloud big data ACP certified Alibaba big data product system](/img/cc/5509b62756dddc6e5d4facbc6a7c5f.jpg)
[teacher Zhao Yuqiang] Alibaba cloud big data ACP certified Alibaba big data product system
![[trivia of two-dimensional array application] | [simple version] [detailed steps + code]](/img/84/98c1220d0f7bc3a948125ead6ff3d9.jpg)
[trivia of two-dimensional array application] | [simple version] [detailed steps + code]

大二困局(复盘)

Altaro set grandfather parent child (GFS) archiving
![[teacher Zhao Yuqiang] redis's slow query log](/img/a7/2140744ebad9f1dc0a609254cc618e.jpg)
[teacher Zhao Yuqiang] redis's slow query log
随机推荐
Progressive multi grasp detection using grasp path for rgbd images
Linux登录MySQL出现ERROR 1045 (28000): Access denied for user ‘root‘@‘localhost‘ (using password: YES)
Es 2022 officially released! What are the new features?
Source insight License Activation
chromedriver对应版本下载
Strategy pattern: encapsulate changes and respond flexibly to changes in requirements
Classification and discussion of plane grab detection methods based on learning
[set theory] relational closure (relational closure related theorem)
Apache+php+mysql environment construction is super detailed!!!
If function of MySQL
[teacher Zhao Yuqiang] Alibaba cloud big data ACP certified Alibaba big data product system
"C and pointer" - Chapter 13 function of function pointer 1 - callback function 1
redis 遇到 NOAUTH Authentication required
Introduction to redis using Lua script
ansible防火墙firewalld设置
【一起上水硕系列】Day 7 内容+Day8
Ansible firewall firewalld setting
2022.DAY592
Final review (Day2)
How to use source insight