Evolutionary Approach to Clustering by Ujjwal Maulik

This lecture introduces evolutionary algorithms for solving clustering problems, where genetic algorithms are used to optimize cluster centers and memberships simultaneously. The speaker explains that traditional K-means clustering has limitations including sensitivity to initial cluster centers and inability to determine optimal number of clusters, which can be addressed by using genetic algorithms with center encoding (rather than point encoding) and incorporating cluster validity indices. The approach is demonstrated on satellite image segmentation, showing improved results compared to standard K-means and Fuzzy C-Means algorithms.