首页 | 本学科首页   官方微博 | 高级检索  
     检索      

基于时间序列采样的正则化回归算法的收敛性
引用本文:张静,王建力,盛宝怀.基于时间序列采样的正则化回归算法的收敛性[J].宁波大学学报(理工版),2011,24(4):51-55.
作者姓名:张静  王建力  盛宝怀
作者单位:1. 宁波大学理学院,浙江宁波,315211
2. 绍兴文理学院数学系,浙江绍兴,312000
摘    要:给出了基于时间序列采样的正则化回归算法的误差分析,即先将误差收敛归结为正则误差和样本误差,借助于Hoeffding型不等式给出了正则误差估计;同时,使用覆盖数及基于时间序列的Hoeffding型不等式给出了样本误差估计,从而算出学习速度一种上界的估计.最后的结果也说明学习速度能达到指数收敛性.

关 键 词:时齐的  一致遍历  误差分析

Convergence of Regularized Regression Algorithms Based on Time Series Sampling
ZHANG Jing,WANG Jian-li,SHENG Bao-huai.Convergence of Regularized Regression Algorithms Based on Time Series Sampling[J].Journal of Ningbo University(Natural Science and Engineering Edition),2011,24(4):51-55.
Authors:ZHANG Jing  WANG Jian-li  SHENG Bao-huai
Institution:ZHANG Jing1,WANG Jian-li2,SHENG Bao-huai2(1.Faculty of Science,Ningbo Univerrsity,Ningbo 315211,China,2.Department of Mathematics,Shaoxing Univerrsity,Shaoxing 312000,China)
Abstract:The error analysis for regularized regression algorithm based on time series is conducted. The error is divided into regularization error and sampling error. The regularization error is given by a Hoeffding-type inequality based on time series; the sample error, on the other hand, is defined by the covering number and the other Hoeffding-type inequality based on time series, with which an upper bound for the error analysis is provided. The results show that the learning rates can achieve exponential convergence.
Keywords:time-homogeneous  uniformly Ergodic  error analysis
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号