In this project a maximum likelihood estimation for a multi-class logistic regression model is implemented, including gradient and stochastic gradient descent implemented from scratch. The model is trained on the MNIST dataset (Modified National Institute of Standards and Technology database). This was a part of an assigment in a course in Machine Learning at National University of Singapore (NUS).
Loss for standard gradient descent
Loss for stochastic gradient descent
Validation accuracy for stochastic gradient descent based on no. of iterations