Invited Speaker---Assoc. Prof. T. Hitendra Sarma


Department of Computer Science and Engineering, Srinivasa Ramanujan Institute of Technology, India

Biography: Dr. T. Hitendra Sarma received Ph.D in the faculty of Computer Science and Engineering from Jawaharlal Nehru Technological University Anantapur, India, in 2013. He has more than 10 years of teaching experience, presently he is working as Associate Professor and Head in Department of Computer Science and Engineering at Srinivasa Ramanujan Institute of Technology, Anantapur, Andhra Pradesh, India. He has more than 15 Publications in various peer reviewed International Journals and Conference proceedings published by Springer, Elsevier and IEEE. As a coauthor, he published a monograph titled “Improvements to Nearest Neighbor Classifier”. Recently, he presented his work in the International Joint Conference on Neural Networks (IJCNN) organized under the IEEE World Congress on Computational Intelligence (WCCI), Vancouver, Canada, in July 2016. He has been a reviewer for Pattern Recognition, Pattern Recognition Letters and Journal of Machine Learning and Cybernetics. He is one of the editors of the book titled “Emerging Trends in Electrical, Communications and Information Technologies” published in LNEE by Springer. He is the Secretary for IEEE CIS/GRSS Society chapter (2016-17) IEEE Hyderabad Section, Region10. He is a member of IEEE, ACM, IE(I) and CSI. His areas of research include Pattern Recognition, Data Mining and Image Processing.

Speech Title: Prototype based hybrid techniques to Speed up k-means and kernel k-means clustering methods for large datasets.
Abstract: Data clustering is a process of identifying the natural groupings that exists in a given data set, such that the inter-cluster similarity is less and intra-cluster similarity is more. It has various applications including pattern recognition, image processing, data mining, remote sensing, etc.. However, identifying the natural clusters of various shapes such as isotropic, non-isotropic, linearly separable, non-linearly separable, etc. is still a challenging problem particularly in case of large data sets. K-means clustering method has been considered as the simple and efficient data clustering algorithm in the literature. However, it fails to identify linearly inseparable and non-convex shaped clusters in the input space. Kernel k-means is often called as the non-linear extension of the conventional k-means clustering method shown better results in such inseparable and non convex shaped clusters. However, the kernel k-means method cannot be applied for very large datasets, because of its quadratic time complexity w.r.t the size of the dataset. This talk presents the recent prototype based hybrid techniques to speed-up the k-means and kernel k-means methods in order to work with large datasets.