图 书 馆
一种用于极端聚类层次算法An Online Hierarchical Algorithm for Extreme Clustering |
|
| 课程网址: | http://videolectures.net/kdd2017_kobren_extreme_clustering/ |
| 主讲教师: | 信息不详。欢迎您在右侧留言补充。 |
| 开课单位: | 马萨诸塞大学 |
| 开课时间: | 2017-10-09 |
| 课程语种: | 英语 |
| 中文简介: |  |
| 课程简介: | Many modern clustering methods scale well to a large number of data items, N, but not to a large number of clusters, K. This paper introduces PERCH, a new non-greedy algorithm for online hierarchical clustering that scales to both massive N and K--a problem setting we term extreme clustering. Our algorithm efficiently routes new data points to the leaves of an incrementally-built tree. Motivated by the desire for both accuracy and speed, our approach performs tree rotations for the sake of enhancing subtree purity and encouraging balancedness. We prove that, under a natural separability assumption, our non-greedy algorithm will produce trees with perfect dendrogram purity regardless of online data arrival order. Our experiments demonstrate that PERCH constructs more accurate trees than other tree-building clustering algorithms and scales well with both N and K, achieving a higher quality clustering than the strongest flat clustering competitor in nearly half the time. |
| 关 键 词: | 现代聚类方法; 分层聚类; 数据科学 |
| 课程来源: | 视频讲座网 |
| 数据采集: | 2023-12-27：wujk |
| 最后编审: | 2023-12-27：wujk |
| 阅读次数: | 11 |