革新知能統合研究センター

Title: Fast Approximation for a general clustering algorithm

Abstract:
In large-scale clustering problems, simple methods such as k-means with low computational cost are usually applied for large-scale data, whereas these methods may not sufficiently capture the hidden cluster structure. Therefore, more complex algorithms, such as spectral clustering, are required to obtain high-quality clustering results. However, due to the high computational costs of such methods, it is not realistic to apply complex methods for large-scale data. In this talk, we develop a general computation-reducing method that can be applied for various clustering methods. We show fundamental theoretical properties of the proposed method. Moreover, we demonstrate the effectiveness of the proposed method by comparing its performance with existing methods through numerical experiments and real data analysis.

Bio:
https://sites.google.com/site/teradahp/home