Neural Network and Deep Learning

This repository contains projects, practices for Neural Network and Deep Learning

Project1: SGD Neural Network

In this practice I built a vanilla neural network using Mini-batch Stochastic Gradient Descent. The network was designed with configurable settings such as network structure, number of training Epochs, mini-batch size and learning rate. Finally, the network was trained on XOR, Iris and MNIST datasets.

Project2: Enhanced Neural Network

This project is meant to enhance and expand on what was achieved from project 1. Building upon the vanilla neural network, I have added more features to allow me explore the performance of NN with more flexibility:

• Early Stopping Criterion
• Activation Funcitons * Sigmoid * Tanh * ReLU * Softmax
• Cost Functions * Quadratic * Cross-Entropy * Negative Log Likelihood
• L2 Regularization
• Momentum Parameter Updates
• Returning Cost and Accuracy for Plotting
• Returning Learned Network