Optimizing Image Classification: Automated Deep Learning Architecture Crafting with Network and Learning...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision

Background:

• Convolutional Neural Networks (CNNs) are crucial for image classification.
• Optimizing CNN architectures is complex and computationally intensive.
• Existing methods often struggle with diverse and complex classification tasks.

Purpose of the Study:

• To introduce ETLBOCBL-CNN, an automated approach for optimizing CNN architectures.
• To enhance the discovery of novel and effective CNN structures.
• To improve classification accuracy across various image datasets.

Main Methods:

• ETLBOCBL-CNN utilizes an encoding scheme for network and learning hyperparameters.
• It incorporates competency-based learning and stochastic peer interaction.
• A tri-criterion selection scheme optimizes fitness, diversity, and improvement rates.

Main Results:

• ETLBOCBL-CNN achieved high accuracies on nine image datasets, including MNIST (99.72%) and Rectangles (99.99%).
• The method demonstrated superior performance compared to state-of-the-art techniques.
• Significant improvements were noted across datasets with varying complexities.

Conclusions:

• ETLBOCBL-CNN effectively automates CNN architecture optimization.
• The approach shows strong potential for enhancing smart device infrastructure development.
• Its competency-based learning and selection schemes drive innovation in CNN design.