College of Engineering, Huazhong Agricultural University, Wuhan, Hubei, 430070, China.
Math Biosci Eng. 2019 May 23;16(5):4692-4707. doi: 10.3934/mbe.2019235.
Extreme learning machine (ELM) is a kind of learning algorithm for single hidden-layer feedforward neural network (SLFN). Compared with traditional gradient-based neural network learning algorithms, ELM has the advantages of fast learning speed, good generalization performance and easy implementation. But due to the random determination of input weights and hidden biases, ELM demands more hidden neurons and cannot guarantee the optimal network structure. Here, we report a new learning algorithm to overcome the disadvantages of ELM by tuning the input weights and hidden biases through an improved electromagnetism-like mechanism (EM) algorithm called DAEM and Moore-Penrose (MP) generalized inverse to analytically determine the output weights of ELM. In DAEM, three different solution updating strategies inspired by dragonfly algorithm (DA) are implemented. Experimental results indicate that the proposed algorithm DAEM-ELM has better generalization performance than traditional ELM and other evolutionary ELMs.
极限学习机（ELM）是一种单隐层前馈神经网络（SLFN）的学习算法。与传统基于梯度的神经网络学习算法相比，ELM 具有学习速度快、泛化性能好、易于实现等优点。但由于输入权重和隐藏偏差的随机确定，ELM 需要更多的隐藏神经元，不能保证最优的网络结构。在这里，我们提出了一种新的学习算法，通过使用改进的电磁机制（EM）算法 DAEM 和 Moore-Penrose（MP）广义逆来调整输入权重和隐藏偏差，来克服 ELM 的缺点，从而通过解析方法确定 ELM 的输出权重。在 DAEM 中，实现了三种不同的基于蜻蜓算法（DA）的解更新策略。实验结果表明，所提出的算法 DAEM-ELM 具有比传统 ELM 和其他进化 ELM 更好的泛化性能。
