Neural-Network-From-Scratch

This is an implementation for neural network without using numpy or any data science or machine learning library.

Neural Network Structure

Neural Network is A Network of fully connected layers:

First layer is the input layer

Last layer is the output layer

in between is just the hidden layers

• Every Node in The layer is called Neuron

We Calculate the Neurons in the hidden layers to the output layers by multiplying every neuron in the previous layer for its weights connected to this neuron

But there are something missing

• We Cant Solve The problems with this equation because the output is just will be linear

• then we will apply non linearity function to every neuron in the layer

  In this code it is the sigmoid

So We will Calculate all neurons through all layers till we get the final layer which is the output layer and calculate the output

.

..

...

Is This Output will be accurate with our weights ?

nope

At First Iteration we initialized the weights at every layer randomly and by applying our activation function to every neuron we will get some output but its not the accurate output

• For the training after calculate the output we will compare it with the real output for this input
• we will calculate the error by Mean squared error
• this error will help us to modify all weights from the output layer to the input layer that is why we are calling it backward propagation
• the whole Iteration of Forwarding till the output layer and Backwarding with changing the weights it is called an Epoch

Perfect References To Understand Neural Network

1. https://towardsdatascience.com/math-neural-network-from-scratch-in-python-d6da9f29ce65

   English

2. https://www.youtube.com/watch?v=twOGEhgYuoI&list=PL5JZLxl_tFCczr7SDjVxZce9lw7ESsaQM&index=5

   Arabic