Optimal sampling in unbiased active learning
Uloženo v:
| Název: | Optimal sampling in unbiased active learning |
|---|---|
| Autoři: | Imberg, Henrik, 1991, Jonasson, Johan, 1966, Axelson-Fisk, Marina, 1972 |
| Zdroj: | Statistical sampling in machine learning 23rd International Conference on Artificial Intelligence and Statistics (AISTATS), Online Proceedings of Machine Learning Research. 108:559-569 |
| Témata: | Optimal design, Weighted loss, Sampling weights, Generalised linear models, Unequal probability sampling, Active learning |
| Popis: | A common belief in unbiased active learning is that, in order to capture the most informative instances, the sampling probabilities should be proportional to the uncertainty of the class labels. We argue that this produces suboptimal predictions and present sampling schemes for unbiased pool-based active learning that minimise the actual prediction error, and demonstrate a better predictive performance than competing methods on a number of benchmark datasets. In contrast, both probabilistic and deterministic uncertainty sampling performed worse than simple random sampling on some of the datasets. |
| Popis souboru: | electronic |
| Přístupová URL adresa: | https://research.chalmers.se/publication/536361 https://research.chalmers.se/publication/520253 https://research.chalmers.se/publication/519957 http://proceedings.mlr.press/v108/imberg20a/imberg20a.pdf |
| Databáze: | SwePub |
Full Text Finder 
Nájsť tento článok vo Web of Science | Abstrakt: | A common belief in unbiased active learning is that, in order to capture the most informative instances, the sampling probabilities should be proportional to the uncertainty of the class labels. We argue that this produces suboptimal predictions and present sampling schemes for unbiased pool-based active learning that minimise the actual prediction error, and demonstrate a better predictive performance than competing methods on a number of benchmark datasets. In contrast, both probabilistic and deterministic uncertainty sampling performed worse than simple random sampling on some of the datasets. |
|---|---|
| ISSN: | 26403498 |