Built an MNIST Neural Network Classifier with 88% accuracy
This repository contains a Neural Network for the MNIST database, built from the scratch using numpy and math. It is a great way of understanding deep learning concepts. Just through this activity I was able to understand the fundamentals of Neural Networks.
-
Loading the data and splitting the data into training and testing. (Shuffling the data before splitting the data)
-
Creating the functions for different steps in the neural network
- Initialize parameters
- Activation Function (ReLU)
- Forward Propagation
- Derivative of Activation function
- One Hot function
- Backward propagation
- Updating the parameters
- Gradient descent
- Predictions and Accuracy Functions
-
Running the gradient descent and finding out the accuracy of the model
The neural network has a simple two layer architecture. Input layer