没有合适的资源?快使用搜索试试~ 我知道了~
boosting 算法
资源推荐
资源详情
资源评论
Boosting
Manuela Vasconcelos
ECE Department, UCSD
Classification
a classification problem has two types of variables
• – vector of observations (features) in the world
• – state (class) of the world
e.g.
•
= (fever, blood pressu re)
• y = disease, no disease
, related by (unknown) function
goal: design a classifier : such that = ,
2
=
3
Perceptron Learning
is simple stochastic gradient descent on the cost
set = 0,
= 0,
= 0
set = max
do {
for
= 1: {
if
+
< 0 then {
•
=
+
•
=
+
• = + 1
}
}
} until
+
0, (no errors)
4
Perceptron Learning
the interesting part is that this is guarantee to converge in finite time
Theorem: Let =
,
, … ,
,
and
If there is
,
such that
= 1 and
then the Perceptron will find an error free hyper-plane in at most
the main problem is that it only implements a linear discriminant
iterations
= max
.
+
> , ,
5
Linear Discriminant
Q: when is this a good decision function?
clearly works if data is linearly separable
• there is a plane which has
• all 1’s on one side
• all 1’s on the other
271A: it was also showed that it is optimal for
• two Gaussian classes
• equal class probability and covariance
but, clearly, will not work even for only slightly
more general Gaussian cases
Q: what are possible solutions to this problem?
剩余49页未读,继续阅读
资源评论
sdq388
- 粉丝: 1
- 资源: 1
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功