Abstract
dc:descriptionIn the information age we live in, collecting new data is now a very easy process as mobile devices, the applications we use and the internet generate an unlimited amount of data. However, the majority of this data is unlabeled, which makes the classification process difficult. In addition, the labeling process itself can be a very time-consuming and expensive as it may require expensive equipment or/and specialized personnel. For this reason, Active Machine Learning methods have been developed that exploit a small set of labeled data to identify those unlabeled data that provide the most information which are then offered for labeling by some skilled personnel. So the goal of Active Machine Learning is to minimize the labeling process and build performant models using a small set of labeled data. This thesis mainly focuses on the development of new Active Machine Learning algorithms to build performant models using small sets of labeled data. In addition, it presents applications of Active Machine Learning to recognition problems and proposes methods for selecting the most appropriate query strategy for selecting unlabeled data using Decision Trees. Finally, it proposes two ensemble of classifiers that show very encouraging results in terms of classification accuracy.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kazllarof, Vangjel
- Καζλάροφ, Βαγγέλ
Subjects
dc:subject × 12- Ενεργητική μηχανική μάθηση
- Αλγόριθμοι μηχανικής μάθησης
- Επιβλεπόμενη ταξινόμηση
- Active machine learning
- Machine learning algorithms
- Supervised classification
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57222