基于子空间信息量准则的前向神经网络设计理论与方法研究

从理论上提出了子空间信息量(SIQ)及其准则(SIQC)的概念；在此基础上阐述了基于上述准则的前向神经网络设计的相关理论，包括前向神经网络隐含层信息量(HLIQ)、存在性和逼近定理，给出了选择隐含层神经元数、权值向量集和隐含层激励函数的指导方向；提出了基于上述理论的一种可行的次优网络设计算法；最后，详细分析了网络性能指标及其影响因素，上述理论和方法完全克服了传统学习算法的各种弊端，丰富了前向神经网络设计领域的理论依据，具有较大的理论指导和实际应用价值，文中通过具体实例验证了上述理论和方法的可行性和优越性．

Research on Theory and Application of Feedforward Neural Networks Design Based on Subspace Information Quantity Criterion

Presented a new concept of the subspace information quantity(SIQ) and its criterion(SIQC); the related theory of feedforward neural networks design based on SIQC was described, including the theorem on hidden layer information quantity (HLIQ), existence and global approximation of feedforward neural networks, and the direction for selecting the number, weight and activate function of hidden neurons were proposed. A new suboptimal algorithm of feedforward neural networks design was given; The performance of networks was analyzed finally. The above theory and application overcomed the shortcoming of general learning methods, and enriched the theories in the domain of feedforward neural networks design. Their reliability and advantage are illustrated through concretely test.