当前位置:首页 > 操作系统 > MacOs

VC-Dimension and Rademacher Complexity-based bounds

VC-Dimension和Rademacher complexity是机器学习中常提到的度量复杂的的概念,一直远观而没有亵玩,今天对这个概念进行学习记录。

VC-Dimension

全称为Vapnik-Chervonenkis dimension,从wiki上搞来一段定义

In Vapnik–Chervonenkis theory, the Vapnik–Chervonenkis (VC) dimension is a measure of the capacity (complexity, expressive power, richness, or flexibility) of a space of functions that can be learned by a statistical classification algorithm. It is defined as the cardinality of the largest set of points that the algorithm can shatter. It was originally defined by Vladimir Vapnik and Alexey Chervonenkis.[1]

即:在VC 理论中,VC维度是一种度量由统计分类算法得到的函数空间的容量(复杂度,表达能力,丰富性或灵活性)的方法。它被定义为算法可以破坏的最大点集的基数。

集合族的VC维

VC-Dimension and Rademacher Complexity-based bounds - 文章图片为集合族,VC-Dimension and Rademacher Complexity-based bounds - 文章图片是一个集合。它们的交被定义为

VC-Dimension and Rademacher Complexity-based bounds - 文章图片
我们说集合VC-Dimension and Rademacher Complexity-based bounds - 文章图片VC-Dimension and Rademacher Complexity-based bounds - 文章图片shatter,如果VC-Dimension and Rademacher Complexity-based bounds - 文章图片包含VC-Dimension and Rademacher Complexity-based bounds - 文章图片的所有子集,即:

VC-Dimension and Rademacher Complexity-based bounds - 文章图片

VC-Dimension and Rademacher Complexity-based bounds - 文章图片的VC维VC-Dimension and Rademacher Complexity-based bounds - 文章图片是被破坏的最大基数,如果任意大的子集能被破坏,那么VC维是VC-Dimension and Rademacher Complexity-based bounds - 文章图片

分类模型的VC维

一个分类模型VC-Dimension and Rademacher Complexity-based bounds - 文章图片有着参数向量VC-Dimension and Rademacher Complexity-based bounds - 文章图片被称为shatter一个集合的数据点VC-Dimension and Rademacher Complexity-based bounds - 文章图片如果对这些数据点所有赋予的标签,都存在VC-Dimension and Rademacher Complexity-based bounds - 文章图片使得VC-Dimension and Rademacher Complexity-based bounds - 文章图片在评估这个集合的数据点的时候无错误。
模型VC-Dimension and Rademacher Complexity-based bounds - 文章图片的VC维是能被VC-Dimension and Rademacher Complexity-based bounds - 文章图片shatter的最大点数。

例子

例如一条直线在平面分类,那么最多二分类的点数为3,所以VC维为3.

Rademacher complexity

In computational learning theory (machine learning and theory of computation), Rademacher complexity, named after Hans Rademacher, measures richness of a class of real-valued functions with respect to a probability distribution.

在计算学习理论中(机器学习和计算理论),Rademacher comlexity由Hans Rademacher命名,度量一类基于概率分布实值函数的丰富性。

集合的Rademacher comlexity

给定集合VC-Dimension and Rademacher Complexity-based bounds - 文章图片VC-Dimension and Rademacher Complexity-based bounds - 文章图片的Rademacher complexity定义如下:

VC-Dimension and Rademacher Complexity-based bounds - 文章图片

其中VC-Dimension and Rademacher Complexity-based bounds - 文章图片是Rademacher分布得到的独立随机变量。

VC-Dimension and Rademacher Complexity-based bounds - 文章图片 forVC-Dimension and Rademacher Complexity-based bounds - 文章图片andVC-Dimension and Rademacher Complexity-based bounds - 文章图片

许多作者在取上界前取了和的绝对值,但是如果VC-Dimension and Rademacher Complexity-based bounds - 文章图片是对称的,那么这就没区别。

函数类的Rademacher comlexity

给定样本VC-Dimension and Rademacher Complexity-based bounds - 文章图片,类别VC-Dimension and Rademacher Complexity-based bounds - 文章图片定义在空间VC-Dimension and Rademacher Complexity-based bounds - 文章图片VC-Dimension and Rademacher Complexity-based bounds - 文章图片的实验Rademacher

complexity对给定的VC-Dimension and Rademacher Complexity-based bounds - 文章图片定义为:

VC-Dimension and Rademacher Complexity-based bounds - 文章图片

这也可以使用之前的定义来写:

VC-Dimension and Rademacher Complexity-based bounds - 文章图片

其中表示函数复合,即:

VC-Dimension and Rademacher Complexity-based bounds - 文章图片
VC-Dimension and Rademacher Complexity-based bounds - 文章图片VC-Dimension and Rademacher Complexity-based bounds - 文章图片的概率分布。函数类VC-Dimension and Rademacher Complexity-based bounds - 文章图片的Rademacher complexity基于VC-Dimension and Rademacher Complexity-based bounds - 文章图片且样本大小为VC-Dimension and Rademacher Complexity-based bounds - 文章图片为:
VC-Dimension and Rademacher Complexity-based bounds - 文章图片
其中上述期望都是从VC-Dimension and Rademacher Complexity-based bounds - 文章图片中采样独立同分布的样本。

 

 

 

承接Matlab、Python和C++的编程,机器学习、计算机视觉的理论实现及辅导,本科和硕士的均可,咸鱼交易,专业回答请走知乎,详谈请联系QQ号757160542,非诚勿扰。

 

 

 

 


【说明】本文章由站长整理发布,文章内容不代表本站观点,如文中有侵权行为,请与本站客服联系(QQ:254677821)!

相关教程推荐

其他课程推荐