Hello,
I am Mohit from LearnVern.( 6 Second pause ; music )
In our previous session of Machine Learning, we studied about Distribution Models, and today we will learn about its Paradigms.
The meaning of Paradigms is types.
To understand how machine learning learns through different types and then solve it.
So, let’s begin..
So majorly we have 3 types of Paradigms that you can see on the slide
For instance,
Supervised learning,
Unsupervised Learning,
Reinforcement learning..
So, let's begin to understand them.
So at first we will start with supervised learning,
So as you can see in this slide it looks like we have suddenly arrived in a nursery class where we are learning about fruits and their names.
This is the way we go about in Supervised Learning, where we learn about input with its output.
Here, input is the image of the fruit and output is the name of that fruit.
We human beings also learn in the same way, where we also have some input in front of us and we get to know it's output also.
For example when we were small, somebody told us to see a 100 rupee note, we carefully looked at its picture and learned that it is a 100 rupees note.
Similarly somebody gave us an input by showing us the shape of a coin and the output which came out of it is that this is 1 Rupee coin.
How ? The information of the shape of the coin is reached to the brain , by the information captured through the eyes.
Right !
So, here you can see that it is a guava so guava looks like this, so this input and output that we can see,
This approach is known as a supervised approach.
(02/08)
Now let's understand through this slide, It will be more clear by understanding this slide, now you can see this picture at number 1. What is it ? and you can easily identify that it is grapes, so input is the shape size and colour which you can identify because your parents or siblings must have told you about it when you were young.
Similarly, second example,
what is it ?
you can easily identify it as banana and this over here, is apple.
So, this approach is a supervised approach which is driven out of supervision.This is supervised approach, we have already done supervision of input and output.
Let's understand it through a definition.
“Supervised learning is a task of machine learning, where a function is developed or a learning function is created whose function is to map an input or output, and to implement that function we provide it with input and output pair before only.”
In our previous slide also we have seen that we provided an image with the name of the fruit so that way we gave both input and output.
Now, let's see its flow, so we gave an input, in between a mapping function is given which maps it with the output.
So even in the future, if you provide any input, it would give proper output without any trouble and errors or maybe with less errors.
So, this is our first approach that is supervised learning.
Clear friends !
Now, let's see our second approach where there are times when the output is not present,(repeat)
so in such cases where there is no output we go towards, unsupervised learning.
Here, you can see that there are different types of shapes present such as star shape, some plus sign, some heart shape.
So, there are different types of shapes present.
Here we need to create a grouping in them and divide them in groups, but here there is nothing defined like group 1 group 2, there is only input, but no output as such, so what can we do?
Here, there can be many ways but we will try and understand with one example in following a way,
So here if I observe, I can see that in one the image itself is filled with white colour and the other has a background filled with that colour, so there is a difference in them based on colour like white and other than white, so I can do grouping on that basis.
So here this one and this one, these shapes which are not filled with colour will come in one group and the others having backgrounds filled with colour will come in another group, so in this way based on features we differentiate between them and form a group.
So, we are not sure if such grouping is defined and proper, but we did it based on mere observation.
So, in the next slide, you can see that we have grouped them and named them as cluster zero and cluster one accordingly.
So, now if I have to ask you which group this image belongs to, you will be able to answer really quickly that it belongs to cluster zero.
And this one over here is from cluster one, on which basis ?
based on the colour filled in the shape.
So, this is known as unsupervised learning.
Now let’s move ahead and see its definition.
This identifies the old undetected patterns, and those who do not have any labels, that means there is no output and the best objective of machine learning is that it should not require any human support, and its algorithm should function without any human supervision.
So, we have to design such an algorithm which does not use any human intervention and is able to identify the patterns, similar looking objects it should be able to place them in one group, and those dissimilar should be grouped by the algorithm in other group, and those found similar in them are clubbed in other group.
So, this is how unsupervised learning works. Its flow is that input data comes, it finds a pattern and thereafter forms a cluster which is nothing but groups.
So, making a group or cluster , both things are similar.
(06/41)
Now, let’s move ahead and learn about third paradigm that is reinforcement learning,
Now, you can see that in the picture a small baby is trying to take steps and climb the stairs, now as the baby is small, it does not have any prior experience of climbing stairs,
So, reinforcement learning is something of that sort.
You already saw in supervised learning that we had data which is nothing but experience. So, it has an input and an output.
In unsupervised learning, we have the data that is input is there but there is no output.
Here in this particular approach there is no data itself, so in reinforcement learning there is no data.
So, as the baby is trying to climb the stairs, for the first time two things can happen: either he will be able to climb or will not be able to climb and even fall down and get hurt.
So we consider that he was not able to climb and fall down, and got a little hurt so it's painful a little bit.
Now, what he will do, again there are two chances, first he will completely quit climbing, next will be I will climb however, but in some other ways, and will build a strategy, either he will cry and call someone and with their support then climb, or he will search for something to hold upon which can help in climbing, he will try to do something different.
So you must have observed here that, on the baby's first action he got a penalty, that is he got hurt, so he got a negative reward, from which he learned things and changed his actions and strategy.
But as he changed his strategy, it doesn’t matter if it wins or not, if he is successful he will be rewarded or if it fails, he will have to change the strategy again that’s it.
So, this is what happens in reinforcement learning.
So, let's understand it through an example.
In this we have an agent, so in this case the agent is the baby, the stairs and things around are the environment and thereafter he learns from the action through penalties or rewards. On whose basis, it decided whether to take that action again in the future.
So this is how reinforcement learning works.
Through this diagram also you can understand that when an agent does an action it receives a state and rewards from the environment back, so this big box represents the environment, the agent is rewarded based on actions and whose basis it takes its next action.
So, this was supervised , unsupervised or reinforcement learning paradigms, which we will see in great depth ahead.
As you can see our next topic is on classification, which comes in supervised learning, where we will learn about classification problems and algorithms.
Till then keep learning and remain motivated.
If you have any queries or comments, click the discussion button below the video and post there. This way, you will be able to connect to fellow learners and discuss the course. Also, Our Team will try to solve your query.
Share a personalized message with your friends.