含有L_(21)范数正则化的在线顺序RVFL算法  

Online sequential RVFL algorithm with L_(21) norm regularization

作　　者：季江飞 郭久森 JI Jiangfei;GUO Jiusen(School of Information Science and Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学信息学院,杭州310018

出　　处：《智能计算机与应用》2022年第10期150-153,共4页Intelligent Computer and Applications

摘　　要：单隐层前馈神经网络(SLFN)以其量级轻、参数量少、训练成本低等优点,目前被广泛地运用于函数逼近处理、模式识别和控制领域中。随机向量函数连接网络(RVFL)作为SLFN的一种,能够将输入层与输出层做直接相连,加强输出层与输入层的关联。然而目前的预测任务中,已经训练好的网络在面对批量数据会随时间不断变化的情况时,则容易显露出泛化能力不足问题。为了提升网络的泛化能力,并防止重复训练,本文提出了一种在线顺序的RVFL算法,使用L_(21)范数实现正则化。在UCI数据集上经过对多种相关参数的最佳选择后,与同类型的RVFL算法和LR_(21)-RVFL算法相比,本文提出的LR_(21)-OSRVFL算法在多种评价指标下均有更优表现。Single-hidden layer feedforward neural network(SLFN)is widely used in function approximation processing,pattern recognition and control due to its advantages of light weight,few parameters and low training cost.Random vector functional-link neural network(RVFL),as a kind of SLFN,can directly connect the input layer with the output layer to strengthen the association between the input layer and the output layer.However,in the current prediction task,the trained network is prone to show insufficient generalization ability in the face of batch data changing over time.In order to improve the generalization ability of networks and prevent repeated training,an online sequential RVFL algorithm is proposed in this paper,and is regularized by L_(21) norm.After the optimal selection of a variety of related parameters on UCI dataset,the LR21-OSRVFL algorithm proposed in this paper has better performance under a variety of evaluation indexes compared with the same type of RVFL algorithm and LR_(21)-RVFL algorithm.

关 键 词：单隐层前馈神经网络 随机向量功能连接网络 在线顺序 L_(21)范数 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]