Clustering-based graph is a successful approach due to its rigorous mathematical foundation. It has been used in many applications such as image and signal processing, data clustering, document analysis, web ranking, etc.
However, standard graph clustering algorithms are off-line algorithms, and hence they cannot be directly applied to dynamic data. Therefore, to handle evolving data, there is a need to develop efficient algorithms for inductive spectral clustering to avoid computation of the eigen-system solution from the scratch. For that, two approaches are used: incremental and evolutionary spectral clustering. Incremental spectral clustering, initialized by a standard spectral clustering, handles evolving data by incrementally updating the eigen-system in order to generates instant cluster labels as the data is evolving. In evolutionary spectral clustering, a good clustering result should fit the current data well, while simultaneously not deviate too much from the recent history. More precisely, it should provide more stable and consistent clustering results that are less sensitive to short-term noises while at the same time are adaptive to long-term cluster drifts. Incremental and evolutionary spectral approaches are applied to dynamic partitioning of transportation network.
Biography: Denis Hamad is professor at the University of Littoral Cote d'Opale. He has been involved in many industrial projects related to machine learning systems: Glass bottles inspection, wind turbine supervision, robot control, fault diagnosis systems and transport security. Actually, his research is in the area of feature selection and spectral clustering for image processing and machine learning.