kruthi-sb / dl_basics

Practice deep learning and neural networks

Implementing Neural Networks with Numpy from scratch

Logistic Regression

• folder: basics
• The math behind a sigle unit of Logistic Regression is beuatifully implemented using numpy.
• The intricacies of initializing weights and biases, forward and backward propagation, computing gradients, and updating weights and biases are implemented.
• The code is tested on various sythetic datasets and on the cat vs non cat dataset. Decision boundary and learning curves are plotted for better understanding.

Shallow Neural Networks

• folder: shallow_nn
• Neural Network using 1 hidden layer is implemented from scratch including forward and backward propagation, computing gradients, updating weights and biases.
• The number of hidden units used can be changed and the resulting model is testing on the noisy moon dataset.

Deep Neural Networks

• folder: deep_nn
• l_layer_nn is implemented from scratch. The architecture of the neural network can be provided as input and the model is trained using the given dataset.
• The intricacies of computing cost, gradients of linear and activation functions, updating weights and biases are implemented.
• This implementation is tested on the synthetic dataset and the cat vs non cat dataset. It achives a very good accuracy compared to the shallow neural network.

Improving Deep Neural Networks

• folder: improving_nn
• improved_nn incorporates various techniques to improve the performance of the neural network such as:
  • Initialization: He initialization, Random initialization, Zero initialization
  • Regularization: L2 regularization, Dropout (Not Implemented)
  • Optimization: Mini-batch gradient descent, Momentum, Adam, RMSprop (Not Implemented)

Diagnosis of Pneumonia using Deep Neural Networks trained on Chest X-ray images

• folder: application
• The l_layer_nn is trained on the chest x-ray images to diagnose pneumonia.
• The current model gives very less accuracy.
• The improved neural network techniques are being implemented to improve the accuracy of the model. (Not Implemented)

Keras Basics

• folder: keras_basics
• 1. Basic Applications: Covers MNIST, Fashion MNIST, and IMDB datasets.
• 2. Operations: Covers basic operations in Keras such as Convolution, Pooling, Flatten, Dense, Dropout, BatchNormalization, etc. (In Progress)