随机权网络稀疏正则化算法及其应用

Sparse regularization algorithm of random weight networks and applications

随机权网络是一种有效的前馈神经网络(FNNs),尤其是内权和偏置值的随机选取极大地提高了网络的学习速率,并克服了其他学习算法的一些不足.但是,其在计算外权的过程中也存在着一些不足.我们就此不足提出了一个新的算法——稀疏正则化算法,并结合梯度投影算法给出了一种迭代解,进而提出了相应的参数选择方法和算法终止准则.实验证明所提出的算法的优势,尤其是当隐层神经元数较多和训练样本较大时,所给出的算法具有明显的优势.

Random weight networks was an effective feed-forward neural networks(FNNs).The random choice of input weights and biases of networks could improve the learning speed and overcome challenges faced by other learning techniques.However,it had shortcomings in calculating the output weights.To solve these shortcomings,we proposed a new algorithm called sparse random weight networks.We put forward an iterative solution combining with the gradient projection algorithm,the way of parameters choice,and the termination criteria of iteration.The experimental results indicates that the proposed algorithm has advantages in the large number of hidden neurons and training samples.