Abstract: This paper looks at the big improvements in classifying images using deep Convolutional Neural Networks (CNNs) from 2020 to 2023. We review how CNN designs have evolved, including EfficientNet and Vision Transformers (ViTs), and the rise of hybrid models that combine both convolutional and transformer techniques. We also examine new training methods like self-supervised learning and clever ways to enhance data, which have greatly boosted model performance. Additionally, we discuss optimization strategies like neural architecture search (NAS) and the use of advanced optimizers, as well as how hardware accelerators and distributed training have improved computational efficiency. By summarizing recent research, this paper gives a clear overview of the current state of CNN-based ImageNet classification, emphasizing key innovations and their importance for future research and applications in computer vision.
Keywords: Artificial Intelligence- Convolutional Neural Networks
|
DOI:
10.17148/IJARCCE.2024.13641
