Pool-based unsupervised active learning for regression using iterative representativeness-diversity maximization (iRDM)

Active learning (AL) selects the most beneficial unlabeled samples to label, and hence a better machine model can be trained from same number of labeled samples. Most existing active for regression (ALR) approaches are supervised, which means sampling process must use some label information, or an model. This paper considers completely unsupervised ALR, i.e., how select without knowing any true information. We propose novel ALR approach, iterative representativeness-diversity maximization (iRDM), optimally balance representativeness diversity selected Experiments on 60 datasets various domains demonstrated its effectiveness. Our iRDM applied both linear kernel regression, it even significantly outperforms supervised when is small.