基于启发知识的神经网络训练策略

摘要

基于对目前神经网络存在问题的具体分析,认为将启发性信息引入神经网络训练将是提高网络学习能力、质量以及效率的重要途径。进而讨论了启发知识的来源与种类,将启发性知识分成诱导性约束和强制性约束两类,进而建立了引入网络训练的相应策略,给出了启发性知识引入与选择的具体原则,并建立了两种基于导数关系的启发知识模型。最后建立了神经网络的具体训练算法。具体应用结果证明了所提出策略与方法的有效性。%Based on the detailed analysis of the problems in current neural network,it is considered that the utilization of heuristic knowledge in the learning of neural network would be an important approach to improve the ability,quality and efficiency of neural network learning.This paper discusses the source and classification of heuristic knowledge,and divides them into two categories:the inducing constraint and obliged constraint.Then it establishes the strategy and detailed principle of the utilization of heuristic knowledge in neural network learning.Two heuristic knowledge models based on derivative relations are also established.At the end of the paper,a detailed algorithm of neural network learning is presented.The effectiveness of the strategy represented has been proved by results of practices.