Abstract
Active learning is mainly to select a part of unlabelled samples from a big dataset. The selected samples are then submitted to domain experts to label and added to the training set. Suppose that the price of labeling samples is far more than the computational cost of training algorithms, we propose a scheme of active learning based on support vector machines, which follows the traditionally inductive learning model of general-specific. In terms of the number of selected samples, the training cost, and the generalization ability, a comparison with some existing active learning algorithms is conducted. The advantages and disadvantages are demonstrated experimentally. ©2010 IEEE.
Original language | English |
---|---|
Title of host publication | Conference Proceedings - IEEE International Conference on Systems, Man and Cybernetics |
Pages | 1312-1316 |
DOIs | |
Publication status | Published - 2010 |
Externally published | Yes |
Keywords
- Active learning
- Order in hypothesis space
- Sample selection
- SVM