An Efficient Iterative Framework for Semi-Supervised Clustering Based Batch SequentialActive Learning Approach

Abstract

Semi-supervised is the machine learning field. In the previous work, selection of pairwise constraints for semi-supervised clustering is resolved using active learning method in an iterative manner. Semi-supervised clustering derived from the pairwise constraints. The pairwise constraint depends on the two kinds of constraints such as must-link and cannot-link.In this system, enhanced iterative framework with naive batch sequential active learning approach is applied to improve the clustering performance. The iterative framework requires repeated reclustering of the data with an incrementally growing constraint set. To address incrementally growing constraint set, a batch approach is applied which selects a set of points based on query in each iterative. In the iterative algorithm, k instances select the best matches in the distribution, leading to an optimization problem that term bounded coordinated matching. Leveraging the availability of highly-effective sequential active learning method will improve performance in terms of label efficiency and accuracy with less number of iterations.