What we will learn in this session ?¶

• The Neuron
• The Activation Function
• How do Neural Network works ?
• How do Neural Networks Learn ?
• Gradient Descent
• Stochastic Gradient Descent
• Backpropagation

1 : THE NEURON¶

A neuron (also called neurone or nerve cell) is a cell that carries electrical impulses. Neurons are the basic units of the nervous system. Every neuron is made of a cell body (also called a soma), dendrites and an axon. Dendrites and axons are nerve fibres.

• Dendrites : are the receivers of the signal to the neuron
• Axon : is the transmitter of the signal to the neuron

As you can see Dendrites are connected to axon of the other neurons

How it works in Machine Learning? ?¶

In this image you can see Green Neuron receive signal from the Yellow Neuron through synapse

Let’s talk about Input Value

Let’s talk about Output

If the output is Categorical then you might get several output values

Lets talk about Weight

Weights are curcial to Artificial Neural Networks Because weights are how Neural Networks learn by adjusting the weights the Neural Network decides in every single case what signal is poor and what signal is not important to certain neuron. When you’re training an artificial neural network you’re basically adjusting all of the weights. Here Gradient Descent and Back Propagation comes into play.

Let’s talk about Neuron

In the second step we need to apply the Activation Function

2: THE ACTIVATION FUNCTION

We have many types of Activation Function Let’s discuss some of the them which are mostly used.

Sigmoid funtion is very useful when you are going to predict probability

Let if we have Binary Dependent Variable ? Which Activation Function you will use ?

This is how Neural Network Looks

3: HOW DO NEURAL NETWORK WORKS

As we have seen Digramatically Structure of most of the Machine Learning Algorithms looks like below

Now, by adding the Hidden Layer in the above model. How we can imporve the efficiency or accuracy of the model. This is how your Artificial Neural Network comes into picture

Let assume we have already trained our Neural Network. Just see how it works Step by Step

Step 1:

In the first step, We have 4 Independant variable : Area, Bedrooms, Distance to city(Miles) and Age. Out of 4 let say only two parameter are required to for first neural network i.e Area and Distance to city

As in the first case we know that Area Increases, Price will increase. and also as we go far away from the city the price also decreases

Step 2:

For second Neuron : Area, Bedroom and Age are the parameter. This could be the case like people having kids or any modern family or Young Population want big size or want New Property.

So area is not matter, they only want big and brandnew property

Step 3:

This is the case, where age matters, as we know the older property has less value. But what if the property is more than 100 years old than that would become the historical property and that might be the case that someone want to buy historical property so he will pay the higher amount

Here in this case we can simply apply the rectifier function. As we can see after 100 years the price of the property will be higher

Step 4:

This might be the case, Neural Network only picks two inputs

Step 5:

This might be the case, where Neural Network pick all the Inputs

This is how our final model look likes

4: HOW DO NEURAL NETWORKS LEARN