00:00reinforcement learning. In supervised learning, we have classification and regression,
00:06in unsupervised learning, we have clustering and in reinforcement learning, it is all about
00:12learning from the feedback from the environment. We will see little more in detail in the coming
00:17slide. So, we come to supervised learning. In supervised learning, we train a machine
00:24learning model with something called as labelled data set. So, suppose consider, we have this set,
00:33a rectangle, some triangles, some circles and some hexagons. And if we can label them,
00:45that means, give a name like this, like a square shape is a rectangle, then the three
00:51side shape is a triangle, then circle and hexagon, then that training data set becomes a labelled
00:59data set. And this labelled data set, when we give to a machine learning model, what happens is,
01:05it learns that ok, this is, if it is having four sides, it is a rectangle, if it is having
01:11three
01:11sides, it is a triangle, if it is having six sides, it is a hexagon and so on. It is
01:16like teaching a
01:17kid that ok, this is A, B, C. Over a period of time, you show the letter alphabet A, B
01:23and C to this
01:24child. After seven or eight days, you show the letter B to him, he will be able to recognise.
01:33So, it is called as supervised learning, you are supervising the learning process. And when you send
01:39this labelled data into the machine learning model and train it, after that, you give some test data,
01:45like one triangle and a hexagon, without giving the label, then it gives, it is a triangle and this
01:52is a hexagon. So, this is supervised learning. Coming to unsupervised learning, what exactly it is?
01:59So, here we have some, the same set, where we have hexagons, rectangles, triangles, circles,
02:06but they are not named. Which one is a rectangle, which one is a circle, which one is a triangle,
02:11there is no label that is attached to these items. There is a, you know, mixed set and which we
02:17are
02:18giving to the unsupervised learning algorithm. This unsupervised learning algorithm, what it does is,
02:26it tries to learn a pattern, a hidden pattern from this data. And then it clusters or groups
02:35these input data into different clusters. Like, you know, it will tell these are, the green ones are
02:43looking similar, the red ones are similar, the blue ones are similar and the yellow ones are similar.
02:51Then coming to something about the reinforcement learning. What exactly is the reinforcement learning?
02:58Reinforcement learning is nothing, but basically learning from the feedback that an environment
03:04gives to an action taken. See here, there is a cartoon in which a robot is there and it is
03:14having
03:14two things in its environment. What are the two things that are there in this environment? One is a tap
03:21from which water is falling and there is another one that side which is a fire. So, this comprises of
03:27the
03:27environment and the agent is the robot. Now, the robot is thinking whether I should go towards the tap
03:35or the fire. It is observing the environment and then it is selecting an action using the decision
03:47called as the policy. So, we call it as a policy where it takes a strategic decision to go either
03:52towards
03:53the fire or towards the fire or towards the water. Now, what happens is, when it goes towards the fire,
04:01it gets hit, ok, because that high temperature is hurting him. So, he gets like, ok, this is not good
04:11for me.
04:12So, he gets some feedback from touching the fire which is equivalent to some negative points.
04:18So, next time based on this learning, he will not go towards the fire. He will choose the path
04:29where he will not get negative points, but he will get some rewards. So, this is all about reinforcement learning.
04:37So, after every action, there will be either a reward or a punishment given to the agent. If he gets
04:48a reward,
04:48he will follow that path. If he gets a punishment, he will reiterate and then he will go to some
04:55other,
04:55he will take some other decision. So, this is learning from the feedback in the environment.
05:01This is about reinforcement learning. Then we come to applications of machine learning, a quick look.
05:07So, we all know supervised, unsupervised and reinforcement learning, just now we have seen.
05:12And in supervised learning, we have two basic algorithms like regression and classification.
05:19In regression, we predict some values and in classification, as the name suggests, we classify some
05:26items, ok. So, for example, weather forecasting is an example of regression problem, where we want
05:34to predict what will be the weather in the next few days. Whereas, in classification, maybe diagnostic,
05:40whether an MRI scan or a CT scan is showing some tumor or not inside the human body. Yes, tumor
05:49is there,
05:50no tumor is not there. So, that kind of thing is a classification task.
05:54Then in unsupervised learning, we learnt how the clustering happens. And this clustering is used in
06:01various kind of recommended system, targeted marketing and customer segmentation.
06:08So, based on whatever previously we have given, our history, there is a recommendation that comes in
06:16the system, ok. So, that is called as clustering algorithm. Like, you know, Netflix gives recommendation
06:23for movies. It is based on whatever we have seen before. And then it has, if I am seeing more
06:29number of Hollywood movies, then it will put me in a cluster where there are audiences who watch
06:36Hollywood movies. And then next time it recommends me, it will recommend based on the history that he
06:43is having from that cluster. So, he will see that, ok, this person belongs to an audience,
06:50target audience who are watching Hollywood movie. So, let us recommend some Hollywood movies to her.
06:55So, this is the way the clustering algorithm works. And in reinforcement learning, we saw that it is
07:02based on the feedback that we get. And it is used in gaming AI, robot navigation, real-time decision making,
07:10like this kind of applications. Now, we come to deep learning. Now, what is this deep learning?
07:19It is a subset of machine learning based on artificial neural network. Now, we will keep on hearing this
07:29term artificial neural network in all our subsequent lectures. And that is the building block of deep
07:38learning. So, what is this artificial neural network? It is nothing but a mathematical model of
07:45the neurons or the network of neurons that are present in human brain. And it mimics how
07:54activities happen in human brain. The similar things are mimicked in artificial neural network.
07:59that offer an artistic leaning in artificial neural network. And it deserves still
07:59about a world of大家都 backwards, which means that I need help it. танies
07:59Yeah.
Comments