SLAMVis: An Interactive Visualization Approach for Smart Labeling on Multidimensional Data

Multidimensional data is widely applied in various machine learning tasks such as regression, classification and pattern recognition. Their labels are significant, especially in supervised learning tasks. Considering the difficulties encountered in the actual situations including the lack of labeled data, low-quality labeled data, and expensive generation of labels, we propose a novel smart labeling method named SLAMVis, which combines active learning and visual interactive labeling, to obtain effective models and labels through an iterative labeling process. The algorithms are tightly integrated with an interactive visual interface, which is composed of multiple coordinated contextual views. Based on the pattern recognition algorithm combining SOINN and K-means, we also introduce a new query strategy to recommend informative candidate instances. Through quantitative experiments and example usage scenarios, we demonstrate the effectiveness of SLAMVis.