单隐层神经网络输入权值的新算法

针对传统极端学习机输入权值与隐层阈值随机设定的问题，提出了输出值反向分配算法。算法在传统极端学习机的基础上，通过优化方法得到最优输出值分配系数，并利用最小二乘法确定网络输入参数。将该算法应用到常用数据集进行实验，并与其他极端学习机改进算法进行比较，显示该算法有良好的学习以及泛化能力，能够得到简单的网络结构，证明了算法的有效性。

New algorithm for input weight of single hidden layer neural network

This paper proposed a back distribution algorithm for output values, which was to solve the random setting problem about input weights and hidden layer thresholds on traditional extreme learning machine. On the basis of traditional extreme learning machine, the algorithm obtained the optimal output value distribution coefficient through the optimization method, and used the least squares method to determine the network input parameters. Experiments on common data sets show that, the proposed algorithm not only achieves good learning and generalization ability compared with other improved extreme learning machine algorithms, but also can obtain a simple network structure and it is effective.