Loading…
Local variational Probabilistic Minimax Active Learning
In the last decade, many excellent active learning methods have been proposed whose algorithms generally deliver an acceptable performance. However, many of these methods rely on measures other than the risk of the classifier. Recently, researchers have proposed Probabilistic Minimax Active Learning...
Saved in:
Published in: | Expert systems with applications 2023-01, Vol.211, p.118538, Article 118538 |
---|---|
Main Author: | |
Format: | Article |
Language: | English |
Subjects: | |
Citations: | Items that this one cites Items that cite this one |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | In the last decade, many excellent active learning methods have been proposed whose algorithms generally deliver an acceptable performance. However, many of these methods rely on measures other than the risk of the classifier. Recently, researchers have proposed Probabilistic Minimax Active Learning (PMAL). PMAL is optimal because, after adding the result of a query, it minimizes the upper bound of the risk of the classifier. Unfortunately, the exact computation of PMAL’s objective function is intractable for likelihood functions suitable for classification problems, such as logistic regression. Employing a variational approach, the present study approximates the PMAL objective when logistic regression is the likelihood function. Experiments show that the proposed algorithm effectively asks queries, which results in superior performance to that of the state-of-the-art.
•Logistic regression which is a workhorse for statistical learning, is used to estimate Probabilistic Minimax Active Learning (PMAL) objective function.•A Local variational approximation of PMAL with logistic regression has been proposed.•An active learning algorithm based on the local variational approximation of logistic PMAL objective shows superior results. |
---|---|
ISSN: | 0957-4174 1873-6793 |
DOI: | 10.1016/j.eswa.2022.118538 |