中文摘要：

极限最小学习机ELM（Extreme Learning Machine）是一种具有快速学习能力的神经网络训练算法。它通过随机选择神经网络节点的参数结合最小二乘法达到了减少训练时间的目的,但它需要产生大量的神经网络节点协助运算。提出一种利用迭代式Lasso回归优化的极限最小学习机（Lasso-ELM）,它具有以下优势：（1）能大幅减少神经网络隐藏层节点的数量;（2）具有更好的神经网络泛化能力。实验表明Lasso-ELM的综合性能优于ELM、BP与SVM。

英文摘要：

Extreme Learning Machine（ELM） is a neural network training algorithm with rapid learning capability.It reaches the goal of reducing training time by randomly choosing the parameters of neural networks nodes and combining the least squares method,but at the cost of producing a great deal of neural networks nodes to assist the operation.In this paper,we propose an ELM utilising the regression optimisation of iteration expression Lasso（Lasso-ELM）,it has the following advantages：（a） it can significantly decrease the number of the nodes in hidden layer of neural networks;（b） it has better generalisation capability of neural networks.Experiments show,the comprehensive performance of Lasso-ELM outperforms the ELM,BP and SVM.