A Novel Hybrid Active Learning Strategy for Nonlinear Regression
Karel Crombecq, Ivo Couckuyt, Eric Laermans, Tom Dhaene · Ghent University Academic Bibliography (Ghent University) · 2009
In many supervised machine learning problems, the labeling of data points is a very expensive operation, often requiring the intervention of a human agent to ac-complish the task. Therefore, many meth-ods have been developped to minimize the number of labeled data points required to achieve a certain accuracy. In active learn-ing, information gathered from previously labeled data is used to determine which data points need to be labeled next. This guarantees more efficient resource usage by focusing data points on areas which are estimated to be interesting or which have a high uncertainty. In this paper, we propose a novel hybrid exploration-exploitation ac-tive learning strategy. 1