两个神经网络优化准则及其在无限制字符识别中的应用

Two Optimization Criterions for Neural Networks and Their Applications in Unconstrained Handwritten Character Recognition

本文根据信息理论提出了模式特征有效提取准则和稀疏表象准则两个新的神经网络优化准则，探讨了它们之间的相互关系，并给出了具体的能量函数形式和优化算法。设计了两个具体的神经网络模型，根据上述算法在无限制手写体字符识别的初步应用中得到了很好的结果。

:Two new optimization criterions for neuro-computiong,Effective Feature Extraction rule(EFE-rule)and Sparse Representation Rule (SP-rule)respectively, are proposed based on recent progress in neural information processing.Possible cost function and learning algorithm are formulated.Two neural networks are constructed and good results on recognition of unconstrained handwritten characters are obtained.