International Research Journal of Education and Technology

In recent years, deep learning has revolutionized artificial intelligence, particularly in computer vision and natural language processing. Two dominant architectures have emerged: Convolutional Neural Networks (CNNs) and Transformer models. CNNs are highly effective in extracting local spatial features using convolution operations, while Transformers excel in capturing long-range dependencies through self-attention mechanisms. However, each architecture has inherent limitations when applied independently. CNNs struggle with modeling global context efficiently, and Transformers often require large datasets and computational resources.

Hybrid CNN–Transformer models combine the strengths of both architectures to overcome these challenges. By integrating convolutional layers with attention mechanisms, hybrid models enhance feature representation, improve contextual understanding, and achieve superior performance in tasks such as image classification, object detection, medical imaging, video analysis, and multimodal learning.