摘要
研究采用多神经网络组合结构、改进BP学习算法和训练策略等3个对策来克服BP学习算法的局限性.手写数字识别的实验研究表明,这些对策效果显著.
We present three countermeasures of using a composite structure of the mutiple neural network,improving BP learning algorithm and perfecting training tactics.By these countermeasures,the limitations of BP learning algorithm are overcome and good results in the recognition of hand-written digits arc obtained.
出处
《福州大学学报(自然科学版)》
CAS
CSCD
1996年第6期13-17,共5页
Journal of Fuzhou University(Natural Science Edition)
基金
福建省自然科学基金