Trust Issues in Active Learning and Their Impact on Real-World Applications

The research area of active learning (AL) has been a subject of continuous attention over the past few decades. However, the application of AL in real-world scenarios remains considerably less popular than might be expected, given the success and advances of the research in this area. As this discrepancy has already been identified by previous researchers, an overview of the literature concerning this issue is given. The identified challenges are grouped and categorized, and existing approaches and solutions that address them are presented. Based on the identified challenges, we propose the hypothesis that a lack of trust in the success of AL is a key factor contributing to the disparity between the advancements of applications and research. In particular, the difficulty of evaluating the performance of an AL process, due to the absence of independent test data, is a major contributor to this trust issue. Finally, we identify research areas and questions that should receive greater attention in order to enhance trust in AL methods and encourage their adoption.