Abstract:
Deep learning has been incredibly successful in many areas, including computer
vision, speech recognition, and natural language processing, because it is such a
powerful tool, and the power of deep learning lies in its ability to learn from vast
amounts of data. However, much of this success has been attributed to supervised
learning models, which means that deep learning models rely heavily on labeled
data. The process of labeling data manually is often time-consuming and requires
domain expertise, making it a significant challenge in the deployment of deep learn-
ing solutions. Many solutions have been developed to address the challenge of large
amount of unlabeled data, such as self-supervised learning and semi-supervised
learning and others. In this work, we aim to explore and implement the concept of
semi-supervised learning, a machine learning approach that uses both labeled and
unlabeled data, to perform the task of radio signal classification. Classifying radio
signals is crucial for various wireless communication applications, but it requires a
large volume of labeled data. Semi-supervised learning offers a promising solution
by enabling the creation of data-efficient classifiers that can learn from a smaller
set of labeled data combined with a larger amount of unlabeled data.
