Week 1

Basics of deep learning

Neural network

Let's see what exactly a neural network is. Let's consider an example on houring price prediction.

So suppose we have the data of 6 houses along with their prices and we're going to use it to make a machine learning model.

Graph of our dataset

Now in machine learning, we would simply plot a line in the graph using some like linear regression. Let's plot it in the above example.

Graph of our dataset

Now we have drawn a simple blue linear line that can work pretty well on the above dataset.

But heyyy, wait! Price can-not be negative, atleast in our case, it seems illogical to sell a house at negative price. So to fix that, we'll put a simple straight line at the x-axis so that the blue line can never go below x-axis.

Graph of our dataset

Now if you see closely, you would see, we have built a simple machine learning model. But in fact, this is also a deep learning model.

Our most simple neural network

And as you can see, this is our neural network, basically a single perceptron for our dataset. In fact, the function we used above is very widely used in deep learning and is known as ReLU.

In the above example, we had nothing but house size. But suppose, now we have a lot more information about the house, in such cases, we create a more complex neural network like below.

A more complex neural network

Supervised learning

Supervised learning is basically training a deep learning (or machine learning) model on a data, which has all the labels including the target label.

It can be done in any of the way.

Structured data

A data that is in more of a tablular format.

Un-structured data

Some data like images, audios or something else that is not structured properly.