CNN Research in Recent Year

In recent years, research on Convolutional Neural Networks (CNNs) has advanced significantly, with numerous innovations in architecture design and optimization methods. From 2019 to 2024, a range of studies have focused on improving CNN performance, efficiency, and applicability across diverse domains. These advancements include automatic architecture design, optimization for embedded systems, and the integration of hybrid models. The transition from traditional to more automated and efficient CNN design has marked a pivotal shift in deep learning research.

Recent CNN Architecture Research (2019–2024)

Key Studies and Findings

1. Meta-Heuristic Automatic CNN Architecture Design:

• A genetic algorithm-based framework that evolves CNN models to create lightweight architectures with high validation accuracy, validated on popular benchmark datasets (Ahmed & Darwish, 2021).

2. Optimization for Image Classification:

• Techniques to determine optimal CNN topology for deployment on embedded platforms, focusing on reducing model size and computational cost while maintaining accuracy (Dhouibi et al., 2021).

3. Automated CNN Design Using Blocks:

• A fully automatic method using genetic algorithms to design CNN architectures based on ResNet and DenseNet blocks, achieving competitive performance with reduced computational resources (Sun et al., 2020).

4. Evolving and Ensembling Deep CNNs:

• An ensemble method using particle swarm optimization to design and fine-tune CNN architectures, leading to improved image classification accuracy (Fielding et al., 2019).

5. Evolution of Deep CNNs Using Genetic Programming:

• A method employing Cartesian genetic programming to construct high-performing CNN architectures, optimized for accuracy using evolutionary algorithms (Suganuma et al., 2020).

Previous Research (Before 2019)

Pre-2019 :

Introduction of deep architectures like ResNet, VGG, AlexNetKrizhevsky et al. (2012), Simonyan & Zisserman (2014), He et al. (2016)

2019–2024:

Automated architecture design, optimization for embedded systems, hybrid models(Ahmed & Darwish, 2021), (Dhouibi et al., 2021), (Sun et al., 2020), (Fielding et al., 2019), (Suganuma et al., 2020)

Conclusion