一种基于最大相关熵准则的核极限学习机

A Kernel Extreme Learning Machine based on Maximum Correlation Criterion

针对传统的基于最小均方误差准则(Minimum mean squared error,MMSE)构造的核极限学习机(Kernel extreme learning machine,KELM)对噪声敏感、容易出现过拟合现象的问题,提出利用最大相关熵准则(Maximum correntropy criterion,MCC)替代最小均方差准则构建核极限学习机.采用拉格朗日乘子法对最大相关熵准则下的目标函数进行求解,推导出适用于高斯噪声环境的核极限学习机(MCC-KELM)模型.UCI公用数据集和噪声数据集上的实验表明该算法具有良好的预测精度和泛化性能.

In view of the fact that the traditional kernel extreme learning machine(KELM)based on the minimum mean squared error(MMSE)is sensitive to noise and prone to over-fitting,this paper proposes to utilize the maximum correlation criterion(MCC)instead of the traditional MMSE to construct the KELM.The Lagrange multiplier method is used to solve the objective function with the MCC,and a new kernel extreme learning machine(MCC-KELM)suitable for Gaussian noise environment is derived.Experimental results on several UCI data sets and noise data sets demonstrate that the proposed model has better prediction accuracy and generalization performance.