中文摘要：

针对神经网络训练数据缺损，造成逼近精度和推广能力大幅下降的问题，提出一种在数据缺损情况下的收敛算法。理论和试验证明该箅法能够有效地提高神经网络在数据缺损情况下的精度和推广能力。

英文摘要：

The absent training data often leads to bad generalization and accuracy of network. To solve this problem, a new algorithm is proposed based on wavelet theory. The theory analysis and simulation show that this algorithm learns the absent data effectively and improves the accuracy and the generalization of network.