论文部分内容阅读
在基于任务自适应的统一框架下讨论了神经网络的生成方案，重点研究了两类典型而适用的神经网络构筑算法，即基于隐节点自适应增长的神经网络结构算法和基于子网自适应增长的神经网络结构算法。还结合所提出的层次式多网络模型，对基于任务的神经网络的自适应结构方案、发展前景与存在的问题作了详细的比较研究。
In the unified framework based on task adaptation, the neural network generation scheme is discussed. Two types of typical and applicable neural network construction algorithms are studied, namely neural network algorithm based on hidden node adaptive growth and adaptive growth algorithm based on subnet Neural Network Structure Algorithm. Combined with the proposed hierarchical multi-network model, a detailed comparison and research is made on the adaptive structure scheme, development prospect and existing problems of the task-based neural network.