Abstract
Recent advances in deep learning have witnessed the emergence of transformer architectures beyond their initial dominance in natural language processing (NLP) tasks. Meanwhile, semi-supervised learning (SSL) remains a crucial strategy for leveraging limited labeled data alongside abundant unlabeled data. This paper investigates the comparative performance of transformer models against classical convolutional neural networks (CNNs) and recurrent neural networks (RNNs) within semi-supervised settings, focusing on general machine learning benchmarks. The study highlights the growing efficacy of transformer-based models in SSL scenarios and discusses challenges and limitations.
View more >>