EVEAL - Expected Variance Estimation for Active Learning

Regression problems frequently occur in the surrounding world, therefore are unavoidable in real-world applications. However, to obtain a model with desired generalization performance usually vast amount of labels is required. In many scenarios obtaining unlabelled data is relatively costless, therefore active learning approaches may be used to reduce needed annotation effort.
Most of uncertainty based regression active learning algorithms base on model prediction variance estimation to choose informative objects. Those algorithms do not incorporate knowledge about the data distribution for the given task. In this paper we propose a novel algorithm to incorporate data distribution knowledge and combine it with variance estimation as an informativeness function. Experiments on 4 data sets show that proposed approach outperforms standard variance based sampling by a margin and indicate robustness of the algorithm.

---------------------------------------

Plan wystąpień w tej edycji jest dostępny tutaj

The schedule of presentations can be checked here