应用单层函数连接神经网络进行字符书写轨迹实时预测
首发时间:2003-12-10
摘要:本文应用单层函数连接神经网络(Flat Functional-link Neural Network)来对字符书写轨迹进行实时预测。对于非平稳随机信号的预测问题,传统的神经网络需要大量的时间和样本进行训练,而本文采用的神经网络在这方面有较好的表现。考虑到汉字字符结构自身的特征,本文对神经网络训练样本的选择上作了一些优化,并对预测算法进行了改进,取得了较好的实验效果。
关键词: 神经网络 时间序列 伪逆矩阵 预测
For information in English, please click here
Application of the Flat Neural Network to Predicting the Time Series of Handwriting Characters Trajectory
Abstract:This paper studies the application of the Flat Functional-link Neural Network (FFNN) to predict the moving penpoint position. For the problem of prediction of a non-stationary time series, convectional neural networks need a lot of time and samples to train, where FFNN can solve this problem very well. Considering the structure of Chinese characters, the paper makes some improvements for the predicting algorithm, and promising experimental results have been obtained.
Keywords: neural network, time series, prediction, pseudoinverse matrix
论文图表:
引用
No.82231070997728****
同行评议
共计0人参与
勘误表
应用单层函数连接神经网络进行字符书写轨迹实时预测
评论
全部评论0/1000