Adaptive habitat biogeography-based optimizer for optimizing deep CNN hyperparameters in image classification.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision

Background:

• Deep Convolutional Neural Networks (DCNNs) excel at image classification but face hyperparameter optimization challenges.
• Traditional Biogeography-Based Optimization (BBO) can suffer from premature convergence and limited exploration in complex tasks.

Purpose of the Study:

• To develop an improved optimization algorithm, the Adaptive Habitat Biogeography-Based Optimizer (AHBBO), for tuning DCNN hyperparameters.
• To enhance the exploration capabilities and population diversity of BBO for better optimization performance.

Main Methods:

• The Adaptive Habitat Biogeography-Based Optimizer (AHBBO) was developed with variable habitat sizes and regulated mutation.
• AHBBO was evaluated on 53 benchmark optimization functions.
• The DCNN-AHBBO model was benchmarked against 23 established image classifiers on nine diverse image classification datasets.

Main Results:

• AHBBO demonstrated improved initial solutions, faster convergence, and superior performance on benchmark functions.
• The DCNN-AHBBO model achieved significant reductions in error rates, up to 5.14%, compared to existing classifiers.
• AHBBO outperformed 13 benchmark classifiers in 87 out of 95 evaluations, showcasing its effectiveness.

Conclusions:

• The proposed AHBBO algorithm offers a high-performance and reliable method for optimizing Deep Neural Networks (DNNs) in image classification.
• This research advances deep learning by providing an effective optimization technique to enhance DCNN efficiency.