The effect of choosing optimizer algorithms to improve computer vision tasks: a comparative study.

Area of Science:

• Machine Learning
• Optimization Algorithms
• Deep Learning

Background:

• Optimization is crucial for enhancing machine learning model performance.
• Various strategies exist to address challenges in the learning process.
• Methodical analysis of optimization techniques is vital for future research.

Purpose of the Study:

• To analyze and summarize diverse optimization strategies from a machine learning perspective.
• To evaluate the effectiveness of specific optimizers on image classification tasks.
• To provide insights for future advancements in machine learning and optimization.

Main Methods:

• Comparative analysis of multiple optimization algorithms including Stochastic Gradient Descent (SGD), Adaptive Moment Estimation (Adam), and Root Mean Square Propagation (RMSProp).
• Application and testing of selected optimizers on the ISIC skin cancer dataset and the COVIDx CT image dataset.
• Evaluation of model accuracy and performance metrics post-optimization.

Main Results:

• The Adam optimizer achieved 97.30% accuracy on the skin cancer dataset.
• The Adam optimizer achieved 99.07% accuracy on the COVIDx CT image dataset.
• Both SGD and Adam demonstrated improved accuracy across training, testing, and validation stages for both datasets.

Conclusions:

• Optimization algorithms like Adam and SGD are effective in enhancing machine learning model accuracy for image classification.
• The choice of optimizer significantly impacts performance, with Adam showing superior results in the tested scenarios.
• Further research into optimization complexities can guide future machine learning developments.