An investigation of Semi-Supervised Learning with autoencoders to improve high dimensional image classification with fewer labels
Journal of High School Science · 2024
DOI: 10.64336/001c.116415 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of High School Science, 2024. Verfasst von Fiona Vitali und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-SA 4.0).
Abstract (Originalsprache)
As the field of machine learning expands, so does the need to learn patterns in high-dimensional image data, a problem we explore with semi-supervised techniques. This learning becomes problematic through the Curse of Dimensionality which explains that as data dimensionality increases, performance decreases. Semi-supervised learning has been proposed as a solution to this problem, by leveraging a small portion of labeled data and a larger unlabeled set. We sought to explore model performance under various model sizes, and the percentage of labeled data. We found that in small amounts of labeled data, the moderately sized model performed best. As the labeled data percentage increases, the larger model obtains the higher test accuracy. Finally, the learning patterns of each of these models are “S”-curves, representing a ‘sweet spot’ of linear semi-supervised learning on a moderate amount of labeled data.
Zitieren
APA
Vitali, F., & Showalter, S. (2024). An investigation of Semi-Supervised Learning with autoencoders to improve high dimensional image classification with fewer labels. Journal of High School Science. https://doi.org/10.64336/001c.116415
Vancouver
Vitali F, Showalter S. An investigation of Semi-Supervised Learning with autoencoders to improve high dimensional image classification with fewer labels. Journal of High School Science. 2024. doi:10.64336/001c.116415
BibTeX
@article{Vitali2024,
author = {Vitali, Fiona and Showalter, Samuel},
title = {An investigation of Semi-Supervised Learning with autoencoders to improve high dimensional image classification with fewer labels},
journal = {Journal of High School Science},
year = {2024},
doi = {10.64336/001c.116415}
}
EUVIMED