关于增量极限学习机的逼近阶估计

The approximation order of Incremental extreme learning machine

与以往探究增量极限学习机的全局逼近能力有所不同,文中给出了该算法定量的收敛性分析。源于纯贪婪算法估计逼近阶的思想,文中运用数列构造与不等式缩放等方法估计增量极限学习机的逼近阶,最终用定理证明了它对于连续目标函数f的逼近阶为o(n-16)。这就对增量极限学习机的快速收敛性能给出了清晰解释。

Different from the existing approximation results of Incremental extreme learning machine （I-ELM） , this paper aims at giving the quantitative analysis of this algorithm. Originated from the pure greedy algorithm, the approximation order of I-ELM has been proved by constructing the appropriate sequences and inequalities. 1 At last, in the form of the theorem, this paper proves that the approximation order of I-ELM is （n ^-1/6） for any continuous target function, so the fast Learning essence of I-ELM may be explained clearly.