Hierarchical Clustering for Semi-Supervised Ground Truth Generation

Supervised learning tasks can require a large collection of labeled data for accurate pattern recognition. For recognition of handwritten characters, manually producing ground truths can be very tedious. In this paper, we propose a semisupervised hierarchical clustering method to reduce the necessary amount of human effort required for labeling a dataset of handwritten characters. The experimental results demonstrate that the approach can improve labeling accuracy over baseline methods.