Modular—tree：一个自构筑的神经网络结构

提出了一种新颖的具有自构筑能力的神经网络结构，称之为Ｍｏｄｕｌａｒ－ｔｒｅｅ和两个相应的自构筑算法。在此结构中，任何现存的前馈神经网络均可以作为子网。对于一个给定的学习任务，利用提出的生成算法通过对输入空间递归地划分，自动生成一树状的模块神经网络，从而避免了网络结构预置问题。由于使用了“分治”原理，Ｍｏｄｕｌａｒ－ｔｒｅｅ具有良好的性能及快速训练的能力。此结构已用于多个监督学习问题（包括：标准测试

Modular-tree:A Self-architecture Neural Network Architecture

Presented a novel self-architecure modular neural network architecture,called modular-tree for supervised learning.In the architecture,any kind of feedforward neural networks can beemployed as componenets and a modular neural network with the tree structure is generated auto-matically with a growing algorithm by partitioning input space recursively to avoid the problem ofpre-determined structure.Due to the principle of divide-and-conquer used in the proposed architec-ture,the modular-tree can yield both a good performance and significantly fast training.The pro-posed architecture has been applied to several supervised learning tasks including both benchmarkand real-world problems and achieved satisfactory results.