【摘要】：極限學習機(ELM)是一種新型的單隱含層神經(jīng)網(wǎng)絡(luò)的訓(xùn)練方法,同傳統(tǒng)的基于梯度的網(wǎng)絡(luò)訓(xùn)練方法相比,具有快速的學習速度和更好的泛化性能。ELM在實際應(yīng)用中往往需要大量的隱含層神經(jīng)元,由于隨機設(shè)定輸入權(quán)值和偏置值,容易導(dǎo)致病態(tài)問題的出現(xiàn)。為解決上述問題,提出一種應(yīng)用量子粒子群(QPSO)優(yōu)化包括隱含層節(jié)點個數(shù)在內(nèi)的網(wǎng)絡(luò)參數(shù)的方法。這種優(yōu)化基于驗證集的均方根誤差,考慮到了輸入權(quán)值矩陣的范數(shù)。在典型的回歸和分類問題上進行試驗證明了算法的有效性。
[Abstract]:Extreme learning machine (ELM) is a new training method of single hidden layer neural network, which is compared with the traditional training method based on gradient. ELM has fast learning speed and better generalization performance. In practical applications, a large number of hidden layer neurons are often required. Due to random input weights and bias values, pathological problems may occur. In order to solve the above problems, a method of optimizing the network parameters including the number of hidden layer nodes by using quantum particle swarm optimization (QPSO) is proposed. This optimization is based on the root mean square error of the verification set and takes into account the norm of the input weight matrix. Experiments on typical regression and classification problems demonstrate the effectiveness of the algorithm.
【作者單位】： 魯東大學信息與電氣工程學院;海軍航空工程學院飛行器工程系;
【基金】：National Natural Science Foundation of China(61602229) Natural Science Foundation of Shandong Province(ZR2016FQ19)
【分類號】：TP18
，

本文編號：2246446