二分法的极端学习机  

Bisection extreme learning machine

作　　者：翟敏[1] 高海瑞[1] 

机构地区：[1]西北大学数学系,陕西西安710127

出　　处：《纺织高校基础科学学报》2014年第1期109-112,共4页Basic Sciences Journal of Textile Universities

摘　　要：基于传统的神经网络理论,并结合原始的ELM算法,提出一种新的确定网络结构的方法,即二分法的极端学习机.该算法无需人为调整任何参数,整个训练过程都是全自动的,这为今后寻求更好地确定网络结构的方法提供了一个好的思路.数值实验结果表明,该方法不仅收敛速度快且能获得较紧凑的网络结构,同时还具有较好的泛化能力.A new method to determine the networks is proposed. Based on the traditional neural network theory, combining the previous ELM algorithm,a new method to determine the network structure which called the bisection ELM is put forward. This new algorithm is full-automatically to determine the net- work architecture which needs not to adjust any parameters,and it gives a good idea to find better method in the future. Simulation results demonstrate and verify this new method not only achieve extremely fast convergence rate and obtain a more compact network structure, but also have good generalization performance.

关 键 词：ELM算法 二分法 网络结构 隐节点 

分 类 号：O29[理学—应用数学]