An Active Learning Approach with Uncertainty, Representativeness, and Diversity

Algorithm 1

Incorporating uncertainty, representativeness, and diversity for active learning.

Input: labeled data set unlabeled data set

Repeat

 Training on to get the probabilistic classification model

 for each in

  Use (1) to measure the uncertainty of sample

  Use (2) to measure the representativeness of sample

  Use (3) to measure the information content of sample

 end for

 Select the high information content set ;

 Apply kernel -means clustering algorithm to ;

 Select centers from each of the clusters;

 Query true labels of the selected samples;

Output: final high-performance classifier .