论文部分内容阅读
极限学习机(ELM)发展自单隐含层前馈神经网络算法,其理论简单,运行快速,应用非常广泛.为了提高ELM的泛化性能,提出了一种带有双并行结构的优化ELM算法(DPELM).在DP-ELM中,建立输入层和输出层间特殊的连接,使得DP-ELM的输出节点不仅可以接收隐含层节点的信息,也可直接接收输入节点的自信息.利用3组回归数据集验证算法性能,实验结果证明,与ELM相比,DP-ELM可以达到更好的回归精度以及更稳定的泛化能力.
Extreme learning machine (ELM) developed from a single hidden layer feedforward neural network algorithm, the theory is simple, fast operation, is widely used.In order to improve the ELM generalization performance, an optimized ELM algorithm with a double parallel structure (DPELM) .In DP-ELM, a special connection between input layer and output layer is established so that the output node of DP-ELM can not only receive the information of hidden node, but also receive the self-information of input node directly. The experimental results show that DP-ELM can achieve better regression accuracy and more stable generalization ability than ELM.