论文部分内容阅读
流形数据的查询需要使用流形的嵌入表示,因此查询流形数据需要访问大量的样本数据.提出一种选择标注分层流形学习算法,选择出的标注点集用来帮助查找流形数据.首先采用自适应近邻算法求出每个数据的最优近邻,然后构造测地线矩阵,最后逐步迭代随机选择标注点,求出每个标注点的极大单元子集,直到流形数据集变成空集,形成初始标注点集.此外,还要优化标注点集.实验结果证明所选择的标注点集保持流形的拓扑特性,可有效帮助查询流形数据.
In order to query manifold data, we need to access a large amount of sample data.This paper proposes a selective labeled hierarchical manifold learning algorithm, and the selected set of labeled points is used to help find the manifold data First of all, we use the adaptive nearest neighbor algorithm to find the optimal neighborhood of each data, then construct the geodesic matrix, and then iteratively iteratively select the label points randomly to obtain the maximum unit subsets of each label point until the manifold data set The set of initial annotation points is formed, and the set of annotation points is also optimized.Experimental results show that the selected annotation point sets maintain the manifold topological properties, which can effectively help to query the manifold data.