Systematising clustering techniques through cross-disciplinary research, leading to the development of new methods

Abstract

Clustering algorithms can help scientists gain valuable insights from data. Thereâ–™s a variety of clustering methods in use, which means there are gaps between the methods used in different fields. Associate Professor Kohei Inoue, Department of Media Design, Kyushu University, Japan, wants to bridge these gaps by investigating the relationships among various clustering methods developed in different fields, in order to systematise the world of clustering. He is bringing two decades of research activities in pattern recognition and image processing to this work. In order to clarify the relationships between different clustering methods, Inoue and the team are conducting an interdisciplinary survey. First, the researchers are working to clarify the relationship between the technologies used across different fields. So far, they have successfully clarified the relationship between the rolling guidance filter and the local mode filter. In a previous study, Inoue and his collaborators proposed a robust K-means clustering al-algorithm. The researchers demonstrated the effectiveness of their technique utilising a BBC dataset originating from BBC News. In their work, the team is collaborating with a laboratory at a university in Japan that is studying non-photorealistic rendering. They have so far published several co-authored papers, as well as having obtained results from their joint research. Ultimately, by systemising clustering technology, Inoue believes that the characteristics of each method, as well as the interrelationships between each method, can be explained and clustering technology enhanced, as well as new clustering techniques developed.