What is it about?
Semi-supervised learning is a way to teach computers by using both labeled and unlabeled data. Imagine you have a bunch of data, but only a few pieces are labeled with correct answers, while the rest are not. This method is useful because it often does a better job than using only labeled data. This paper explores semi-supervised learning and makes changes to two popular methods: self-training and co-training. While many studies tweak these methods, no one has automated the process of labeling the unlabeled data—until now. This paper introduces three new techniques—FlexCon-G, FlexCon, and FlexCon-C—that vary in how they calculate confidence and choose labels. The results show that all three new methods usually perform better than the original self-training and co-training methods.
Featured Image
Photo by Jan Antonin Kolar on Unsplash
Why is it important?
This paper introduces three new techniques: FlexCon-G, FlexCon, and FlexCon-C, which are variants of two popular methods: self-training and co-training. These new methods perform better than the original self-training and co-training methods.
Perspectives
I hope this paper inspires other researchers to explore semi-supervised learning further and consider using the FlexCon family of algorithms.
Flavius Gorgonio
Universidade Federal do Rio Grande do Norte
Read the Original
This page is a summary of: An Efficient Approach to Select Instances in Self-Training and Co-Training Semi-Supervised Methods, IEEE Access, January 2022, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/access.2021.3138682.
You can read the full text:
Contributors
The following have contributed to this page
