一种自适应确定隐层节点数的增量半监督超限学习机算法  

An Incremental Semi-Supervised Extreme Learning Machine with Capacity for Determining Number of Hidden Nodes Adaptively

作　　者：梅颖[1] 沈洋[1] 叶思语 卢诚波[1] MEI Ying;SHEN Yang;YE Siyu;LU Chengbo(Faculty of Engineering,Lishui University,Lishui,Zhejiang 323000,China;School of Public Health and Management,Wenzhou Medical University,Wenzhou,Zhejiang 325035,China)

机构地区：[1]丽水学院工学院,浙江丽水323000 [2]温州医科大学公共卫生与管理学院,浙江温州325035

出　　处：《复旦学报（自然科学版）》2022年第1期27-33,共7页Journal of Fudan University：Natural Science

基　　金：国家自然科学基金(12171217);浙江省自然科学基金(LY21F020004,LY18F030003)。

摘　　要：单隐层前馈神经网络中,隐层节点个数是影响网络的学习能力和复杂程度的重要因素。在实际应用当中,如何确定网络的隐层节点个数仍然是一个开放的问题。在半监督超限学习机(SS-ELM)的基础上,本文提出了一种增量半监督超限学习机(ISS-ELM)算法,对于给定的学习精度,该算法能够逐个或者成批地增加隐层节点,并自适应确定隐层节点数量。在此过程当中,网络的外权矩阵不需要重新训练,只需逐步更新。理论分析和仿真实验表明:ISS-ELM在保持SS-ELM泛化能力的基础上,大幅提高了学习速度;此外,与另一种监督学习类型的增量超限学习机(EM-ELM)相比,ISS-ELM具有更好的泛化能力。The number of hidden layer nodes is a key factor affecting the learning ability and complexity of the network in single hidden layer feedforward neural networks.One of the open problems in neural network research is how to determine the number of hidden layer nodes for given applications.In this paper,an incremental semi-supervised extreme learning machine(ISS-ELM)was proposed based on semi-supervised extreme learning machine(SS-ELM).For a given learning accuracy,ISS-ELM can add hidden nodes to SLFNs one by one or group by group,and determine the number of hidden nodes adaptively.During the growth of the networks,the output weight matrix is updated incrementally.The theoretical analysis and simulation results demonstrate that ISS-ELM improves the speed of learning while keeping the generalization ability of SS-ELM,in addition,ISS-ELM has better generalization ability than EM-ELM which is a kind of incremental supervised extreme learning machine.

关 键 词：单隐层前馈神经网络 隐层节点 半监督学习 超限学习机 增量学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP183[自动化与计算机技术—控制科学与工程]