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