随机学习规则下的可学习性和LOO稳定性分析
Learnability and LOO stable under randomized learning rule setting
徐天伟 1黄晓 2周菊香 3高炜1
作者信息
- 1. 云南师范大学信息学院,云南昆明 650500
- 2. 浙江师范大学教育学院,浙江金华 321000
- 3. 云南师范大学民族教育信息化教育部重点实验室,云南昆明 650500
- 折叠
摘要
学习算法的可学习性是统计学习理论的基本问题.本文指出,一个随机学习问题具有可学习性当且仅当存在AERM且一致LOO稳定的学习规则.同时,得到随机学习规则下通用学习算法的一些结果.
Abstract
The problem of learnability is one of the most basic questions in statistical learning theory.We show that a randomized learning problem is learnable if and only if there exists a learning rule which is always an AERM and uniform-LOO stable.Moreover,a generic learning algorithm in randomization learning rule setting will study.
关键词
统计学习理论/学习性/一致收敛/经验风险最小/稳定性Key words
statistical learning theory/learnability/uniform convergence/empirical risk minimizer/stability引用本文复制引用
基金项目
Supported by the National Natural Science Foundation of China(60903131)
Key Science and Technology Research Project of Education Ministry(210210)
出版年
2012
NETL
NSTL
万方数据