摘要
本文通过对函数集容量的分析 ,得出用函数的微分来控制函数集容量的学习方法 .该方法不仅能用支撑矢量核函数而且可以采用其他的函数作为基函数 .基于样本的机器学习 ,要求学习机在容量控制和过拟合之间取一个折衷 ,从而保证学习机的推广能力和误差精度 .本文通过在微分容量控制和最小化经验误差之间作一个折衷 ,提出基于微分容量控制的学习机 .仿真实验验证了我们的学习机具有良好的推广能力 .
A differential method for capacity control is presented based on the analysis of the capacity of learning machines. Our method that can be applicable to the set of nonlinear hypothesis functions as well as the set of linear ones generalizes the theory about the capacity control of SVMs. A new learning machine is proposed based on differential capacity control method. In our learning machine, a good generalization performance can be obtained by the right balance struck between the empirical risk and the differential of the set of hypothesis functions that controls the machine capacity. Simulation results show the feasibility of our learning machine.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2003年第10期1526-1531,共6页
Acta Electronica Sinica
基金
国家自然科学基金 (No .60 0 730 53
60 1 330 1 0
69831 0 4 0 )
