TY - GEN

T1 - Semi-supervised classification with cluster ensemble

AU - Berikov, Vladimir

AU - Karaev, Nikita

AU - Tewari, Ankit

PY - 2017/11/14

Y1 - 2017/11/14

N2 - We propose a method for semi-supervised classification using a combination of ensemble clustering and kernel based learning. The method works in two steps. In the first step, a number of variants of clustering partition are obtained with some clustering algorithm working on both labeled and unlabeled data. Weighted averaged co-association matrix is calculated using the results of partitioning. We prove that this matrix satisfies Mercer's condition, i.e., it defines symmetric non-negative definite kernel. In the second step, a decision function is constructed on labeled data using the obtained matrix as kernel. Some theoretical properties of the proposed method related to its convergence to the optimal classifier are investigated. Numerical experiments show that the proposed method possesses accuracy comparable with some state of the art methods, and in many cases outperforms them. We will illustrate the performance of the method on the problems of semi-supervised classification of hyperspectral images.

AB - We propose a method for semi-supervised classification using a combination of ensemble clustering and kernel based learning. The method works in two steps. In the first step, a number of variants of clustering partition are obtained with some clustering algorithm working on both labeled and unlabeled data. Weighted averaged co-association matrix is calculated using the results of partitioning. We prove that this matrix satisfies Mercer's condition, i.e., it defines symmetric non-negative definite kernel. In the second step, a decision function is constructed on labeled data using the obtained matrix as kernel. Some theoretical properties of the proposed method related to its convergence to the optimal classifier are investigated. Numerical experiments show that the proposed method possesses accuracy comparable with some state of the art methods, and in many cases outperforms them. We will illustrate the performance of the method on the problems of semi-supervised classification of hyperspectral images.

UR - http://www.scopus.com/inward/record.url?scp=85040514677&partnerID=8YFLogxK

U2 - 10.1109/SIBIRCON.2017.8109880

DO - 10.1109/SIBIRCON.2017.8109880

M3 - Conference contribution

AN - SCOPUS:85040514677

SP - 245

EP - 250

BT - Proceedings - 2017 International Multi-Conference on Engineering, Computer and Information Sciences, SIBIRCON 2017

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2017 International Multi-Conference on Engineering, Computer and Information Sciences, SIBIRCON 2017

Y2 - 18 September 2017 through 22 September 2017

ER -