Place: Large Lecture Room – CVC
Affiliation: Humboldt Fellow, German Cancer Research Center Heidelberg, Germany Dep.of C.S and Engineering, Jadavpur Univ. India
Clustering, a popular exploratory data analysis tool, attempts to partition a given data set into a number of cohesive groups such that patterns that belong to the same cluster are much more similar to each other than patterns that do not. In many situations, it is difficult to assign a point to just a single cluster. Hence in fuzzy clustering, each point is provided a membership value, indicating its belongingness, to multiple clusters.
Often, the clustering problem is mapped to one of searching for an appropriate number of suitable partitions such that some goodness measure is optimized. Consequently several researchers have been actively investigating the effectiveness of genetic algorithms for clustering, since it involves appropriate parameter selection and efficient search in complex and large spaces in order to obtain robust and close approximate solutions. Evidently, it is difficult to define a single measure of cluster goodness. Hence in recent times, the clustering problem has been posed as one of multiobjective optimization, where multiple, often conflicting, objectives are simultaneously optimized. In this talk we will deal with the problem of multiobjective fuzzy clustering using some recently proposed multiobjective optimization techniques. Since such methods provide a set of Pareto-optimal solutions, a novel way of integrating these with support vector machines will be described in order to arrive at a consensus solution.
Some real life applications of the said multiobjective fuzzy clustering method to gene expression data as well as remote sensing images will be demonstrated.