作者: Nurjahan Begum , Bing Hu , Thanawin Rakthanmanon , Eamonn Keogh
关键词:
摘要: In the last decade plunging costs of sensors/storage have made it possible to obtain vast amounts medical telemetry. However for this data be useful, must annotated. This annotation, requiring attention experts is very expensive and time consuming, remains critical bottleneck in analysis. Semi-supervised learning an obvious way mitigate need human labor, however, most such algorithms are designed intrinsically discrete objects, do not work well domain, which requires ability deal with real-valued objects arriving a streaming fashion. we make two contributions. First, demonstrate that many cases just handful annotated examples sufficient perform accurate classification. Second, devise novel parameter-free stopping criterion semi-supervised learning. We evaluate our comprehensive set experiments on diverse sources including electrocardiograms. Our experimental results show approach can construct classifiers even if given only single instance.