Foto von Christiane Schneide ©

Dr. Chrsitiane Schneide : "Cluster detection using network measures and spectral clustering"

09. Jan

Im Rah­men des For­schungs­kol­lo­qui­ums Wirt­schafts­in­for­ma­tik und Data Sci­ence re­fe­riert Frau Dr. Chrsitiane Schneide vom Institut für Mathematik und ihre Didaktik, Leuphana Universität, über "Cluster detection using network measures and spectral clustering".


Da­tum und Ort:  09. Januar 2020     12.15 Uhr     Raum C 40.255


For large data sets, dimensionality reduction is often crucial. One possibility is to perform clustering. This means to search data items that are similar to each other and rather dissimilar to all other items in the data set. In the case of a high-dimensional data set with many items, it is useful to construct a network where links between nodes (data items) have associated weights that express the pairwise similarities between the nodes. Thus, the network can be represented by its similarity matrix. Subsequently, if hard clustering is not requested, we can use different network measures, such as the node degree, to visualize similarities of data items. Finally, spectral clustering techniques in combination with k-means clustering can be applied for hard clustering. 
In this talk, we show how these individual methods perform for different data sets from fluid dynamics.