OBEBS (Optimally Balanced Entropy-Based Sampling)

Published: 14 February 2020| Version 2 | DOI: 10.17632/7cz7rgg76d.2


In active learning, Optimally Balanced Entropy-Based Sampling (OBEBS) method is a selection strategy from unlabelled data. At active zero-shot learning there is not enough information for supervised machine learning method, thus, our sampling strategy was based on unsupervised learning (clustering). The cluster membership likelihoods of the items were essential for the algorithm to connect the clusters and the classes; i.e. to find assignment between them. For best assignment, Hungarian algorithm was used. We developed and implemented adaptive assignment variants of OBEBS method in the software.



Software, Machine Learning, Active Learning, Clustering, Image Classification, Sampling