Evaluating Interactive 2D Visualization as a Sample Selection Strategy for Biomedical Time-Series Data Annotation
A new research paper explores interactive 2D visualizations as a strategy for selecting samples for annotating biomedical time-series data. The study compared this method, termed 2DV, against random sampling (RND) and farthest-first traversal (FAFT) across infant motility assessment and speech emotion recognition tasks. Results indicated that 2DV generally performed best when aggregating labels, particularly for capturing rare classes and with expert annotators, while FAFT excelled when models were trained on individual annotators' labels due to budget constraints. The research also noted that 2DV made the annotation process more engaging for participants. AI