Automated Debugging of Deep Text Classifiers
Proceedings of the Canadian Conference on Artificial Intelligence · 27.05.2024
DOI: 10.21428/594757db.2387ff51 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the Canadian Conference on Artificial Intelligence, online erschienen am 27.05.2024. Verfasst von Shamir Towsif Khandaker und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Deep Learning models are heavily reliant on large datasets. Often, it is challenging to get a large yet perfect dataset. Many datasets might have biases against specific subpopulations or not work in real-world classification problems because of over-fitting. When such datasets are used, the model inherits these problems. Specific models have been proposed to overcome these problems, incorporating human feedback to reduce irrelevant features. But human feedback is very energy and time-consuming. In this paper, we propose a model which replaces the human feedback process and automatically reduces the irrelevant features. Experimental results show that our automated model performs almost as well as the state-of-the-art models, which use human feedback for most cases. In some instances, our model outperforms these models.
Zitieren
APA
Khandaker, S. T., Islam, A., & Trahan, G. (2024). Automated Debugging of Deep Text Classifiers. Proceedings of the Canadian Conference on Artificial Intelligence. https://doi.org/10.21428/594757db.2387ff51
Vancouver
Khandaker ST, Islam A, Trahan G. Automated Debugging of Deep Text Classifiers. Proceedings of the Canadian Conference on Artificial Intelligence. 2024. doi:10.21428/594757db.2387ff51
BibTeX
@article{Khandaker2024,
author = {Khandaker, Shamir Towsif and Islam, Aminul and Trahan, Gabriel},
title = {Automated Debugging of Deep Text Classifiers},
journal = {Proceedings of the Canadian Conference on Artificial Intelligence},
year = {2024},
doi = {10.21428/594757db.2387ff51}
}
EUVIMED