Semi-automatic data annotation guided by feature space projection

Data annotation using visual inspection (supervision) of each training sample can be laborious. Interactive solutions alleviate this by helping experts propagate labels from a few supervised samples to unlabeled ones based solely on the analysis their feature space projection (with no further supervision). We present semi-automatic data approach suitable and semi-supervised label estimation. validate our method popular MNIST dataset images human intestinal parasites with without fecal impurities, large diverse that makes classification very hard. evaluate two approaches for learning latent spaces, choose one best reduces user effort also increases accuracy unseen data. Our results demonstrate added-value analytics tools combine complementary abilities humans machines more effective machine learning.