Journal article
Toward optimal probabilistic active learning using a Bayesian approach
Publication Details
Authors: | Kottke, D.; Herde, M.; Sandrock, C.; Huseljic, D.; Krempl, G.; Sick, B. |
Publisher: | SPRINGER |
Publication year: | 2021 |
Journal: | Machine Learning |
Pages range : | 1199-1231 |
Volume number: | 110 |
Issue number: | 6 |
Start page: | 1199 |
End page: | 1231 |
Number of pages: | 33 |
ISSN: | 0885-6125 |
eISSN: | 1573-0565 |
DOI-Link der Erstveröffentlichung: |
Abstract
Gathering labeled data to train well-performing machine learning models is one of the critical challenges in many applications. Active learning aims at reducing the labeling costs by an efficient and effective allocation of costly labeling resources. In this article, we propose a decision-theoretic selection strategy that (1) directly optimizes the gain in misclassification error, and (2) uses a Bayesian approach by introducing a conjugate prior distribution to determine the class posterior to deal with uncertainties. By reformulating existing selection strategies within our proposed model, we can explain which aspects are not covered in current state-of-the-art and why this leads to the superior performance of our approach. Extensive experiments on a large variety of datasets and different kernels validate our claims.
Gathering labeled data to train well-performing machine learning models is one of the critical challenges in many applications. Active learning aims at reducing the labeling costs by an efficient and effective allocation of costly labeling resources. In this article, we propose a decision-theoretic selection strategy that (1) directly optimizes the gain in misclassification error, and (2) uses a Bayesian approach by introducing a conjugate prior distribution to determine the class posterior to deal with uncertainties. By reformulating existing selection strategies within our proposed model, we can explain which aspects are not covered in current state-of-the-art and why this leads to the superior performance of our approach. Extensive experiments on a large variety of datasets and different kernels validate our claims.
Keywords
Active learning, Classification, Probabilistic active learning
