Overview / Usage

In this project neural network has been implemented from basics without use of any framework like TensorFlow or sci-kit-learn. Training has been done on the MNIST dataset. Implementation has been done with minimum use of libraries to get a better understanding of the concept and working on neural nets.

It gives a clear understanding of neural networks and can be used for research work by manipulating the network.

Methodology / Approach

The following methodology was adopted:

• Functions for random initialization of weights and bias
• Activation functions
• Derivatives of the activation function
• Function for Forwarding propagation
• Backward propagation
• The cost has been written separately and derivative has been found.

Technologies Used

Software used:

• Python
• Jupyter Notebook

Intel

• Intel Integrated Graphics

Tools

• Numpy-For array manipulations
• Matplotlib-For visualization

Repository

https://github.com/manzoormahmood/mnist_neural-network-from-scaratch