IMAGE CLASSIFICATION WITH STOCHASTIC GRADIENT DESCENT METHOD

• Main Author: Raffa Reyhan, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84282
• Institution: Institut Teknologi Bandung

As one of the branches of Artificial Intelligence, Machine Learning is actively developing in various fields of technology and human life. With the increasing popularity of big data, large-scale Machine Learning problems have emerged that must be solved. One way is to use a mathematical model. Many forms of mathematical models are built to solve cases in the Machine Learning domain, one of which is the problem of image classification. In this study, the focus will be on solving image classification problems using a perceptron model or simple neural network model. In building a model, the model that is to be built is an optimal one with good accuracy and high computational efficiency. In practice, the model is built using the gradient method. The main focus of this research is the influence of stochastic elements in the Gradient Descent method on model accuracy and computational efficiency. The experimental results show that overall computational efficiency can be higher when combining stochastic elements in the gradient descent method with certain stopping criteria. Model accuracy is also influenced by adding stochastic elements to the gradient descent method where the increase or decrease in model accuracy is influenced by the batch size chosen.