Deep Neural Network

Description

This project implements a deep neural network, from scratch, without the use of external machine learning libraries. It features a NeuralNetwork class which allows to create a feedforward network of personalized shape and activation functions. This can be used to train simple deep learning models.

Structure

The code is structured as follows:

• NeuralNetwork.py: Allows to create a feedforward network of personalized shapes and activation functions. Contains a trainModel function to train the model for a given number of epochs and learning rate.
• Initializer.py: Different way to initialize weights and biases for a given layer.
• Var.py: Object which holds a floating variable and enables gradient computations.
• Utils.py: Provide various utility functions.
• FNN_Test.py: Instanciate a feedforward network and trains it on a generated dataset.

Built With

This project was build using the following technology:

• Python

Sources

• Neural Networks and Deep Learning Book
• Deep Learning course, DTU