摘要
树形结构数据包括多个结点的属性信息与结点间的连接结构信息,然而传统的机器学习对树形数据的处理方法比较单一。为此,提出一种适用于树形结构的树形回声状态网络方法,使用树形回声状态网络对树形结构数据进行建模,得到固定维数的空间模型,从而将复杂的树形结构数据转换为模型空间中的点。基于模型空间的思想,通过模型空间中点的距离来度量树形结构数据之间的相似度,并将模型与核方法相结合以提高分类器的判别能力。实验结果表明,树的回声状态网络方法与传统方法相比,在相关数据集上有着较好的测试性能。
Classical machine learning methods are not enough for dealing with tree data because the tree contains not only node information but also structure information. Therefore, this paper proposes an approach of tree echo state network,applicated to tree-structured,it uses the tree-structured echo state network to model tree-structured data, gets a fixed-size space model, and converts the complex tree-structured data into points in the model space. Based on the idea of model space, the similarity between the tree-structured data is measured by the distance between the models. It combines the model with the kernel methods to facilitate classification performance. Experimental results show that, compared with traditional algorithms,the tree echo state network has better performance in related datasets.
出处
《计算机工程》
CAS
CSCD
北大核心
2017年第4期194-199,206,共7页
Computer Engineering
关键词
机器学习
回声状态网络
水库
模型空间
树结构数据
machine learning
echo state network
reservoir
model space
tree-structured data