Unlock all 58 lessons of Statistics and Probability for Data SciencesFull course ₹40,000 · paid
Introduction to AI, Machine Learning and Learning Types
A free lesson from Skill-Lync's Statistics and Probability for Data Sciences course, on Grow by Skill-Lync · Last updated 16 Nov 2026 · Session: “Week 1”
What will you learn in this free Statistics and Probability for Data Sciences lesson?
The instructor introduces AI through movies and real examples such as Google Translate, chatbots and driverless trains. The lesson then explains supervised, unsupervised and reinforcement learning, features versus labels, and the steps of building a model with a train and test split. No mathematics is done yet.
- What AI is and what machines can mimic
- Supervised, unsupervised and reinforcement learning explained
- Features versus labels, machine learning versus deep learning
- Steps to build and test a model
What comes after lesson 1?
The next free lesson is lesson 2, “Set Theory Basics: Number Sets and Set-Builder Notation” (17 min). Later in Skill-Lync's Statistics and Probability for Data Sciences course (paid): lesson 4, Week 2 Session 3; lesson 5, Week 2 Session 4; lesson 6, Week 3 Session 1. The full course has 58 lessons (21 h 38 min of video); see the course overview or its price and syllabus.
Lesson transcript
Machine transcript of the lesson audio (lightly cleaned). Technical terms may contain errors; the video is the reference.
Hey guys, welcome to your new classes. This classes will be about Mathematics, Statistics and Probability for Machine Learning. Okay, so after this classes, you will be able to appreciate, you know, about data, how you can manipulate data, pre-process data, how data cleansing can be done, how data can be made more useful for further analysis. Okay, so this is a course on Mathematics, Statistics and Probability. Although I am saying Mathematics, Statistics and Probability, your first week would be an Introduction to Artificial Intelligence. Okay, all the Mathematics you will be doing will start from the second week.
Okay, so let us, you know, go into the topic. So what we will be covering in this week is, we will be giving you a brief introduction of what is AI. So the, after the introduction of AI, we will be talking about machine mimicking human. So first of all, why do we even want machine to mimic human? Then, if at all you want to, you know, have a machine that is going to mimic human, how do you teach it how to mimic like a human? Then, I will like motivate you to understand why we need systems, why we are aiming to have successful machine learning or artificial intelligence.
And thereby, I will also discuss about supervised learning, unsupervised learning, reinforcement learning. These are some things we keep hearing all the time, right? The words like computer vision, augmented reality, virtual reality, face recognition, all of these jargons do belong to this artificial intelligence subject. Okay, then we would proceed into data type. Then we will identify the pattern. And then finally, I will like brief you on how to create a ML and DL algorithm. Okay, this will be like a very brief introduction to a very vast subject. Okay, so let us go.
So the learning objectives for this week is, you will know about AI and actually understand the importance of mathematics in the field of AI. Once you understand what is all is done in AI and then you know the problem we face in the field of AI, then you will understand why is mathematics, statistics, probability more important and you will learn to appreciate the subjects. Okay, so let us go into introduction to the AI. So before you know, drilling into the topic, I am going to ask you a question.
We are all movie buffs, right? So what is common between these five movies that are listed here? Age of Ultron, Ravan, Indyvan, Terminator and Matrix. Okay, pause the video here and try to see if you can find anything common. Take your time to find, okay? Don't rush forward. See, we all have brains. Our brains do processing. So make use of it. See if you can find the pattern in this. And for those who are not movie buffs, let me tell you what the movie actually showcases.
So you have the movie Age of Ultron. So in this movie, what happens is, you have Tony Stark who is a philanthropist, billionaire, intelligent who, you know, creates Ultron. Ultron is actually a robot. So what it does is, something goes wrong and then it takes over the world. It tries to destroy the world. And then finally, Avengers assemble back and try to save the world from Ultron. That is Age of Ultron. So essentially, you have a mission that is not behaving the way it should behave.
It is not following the Asimov's rule. It is like hurting human beings. So that is Age of Ultron for you. Then you have this movie Ravan. So villain in the game becomes a stronger character and starts attacking human beings. Okay, so that is with regards to Ravan. Indyvan is a movie. So this is from South India. Yeah, the whole point about Indyvan is, you have a machine that is able to replicate itself. And then it, of course, tries to take over the world and then Kivo saves the world, you know, from the mission.
But the point is, the mission is trying to replicate. Replication, reproduction is all important property of a living thing, right? You know, bacterias reproduce, human beings reproduce, animals reproduce, birds reproduce, viruses reproduce. Everything that is living reproduce. But machine, they cannot reproduce or replicate, right? But in this movie, you have a machine, which, you know, learns to replicate itself and tries to take over the world. Okay, so I want you to remember, you have a machine and the machine is replicating. Then in the next stage, you have the movie Terminator.
So what happens in Terminator is, human beings are, you know, happily using all the equipments and machines. One day a machine takes over them. One of the guy, you know, he realizes that, no, this is, there is something wrong. We have to do something else. Otherwise, the machines will completely take over. So this guy is going to, you know, attack all the machines and finally save the humanity. So that is, you know, the backstory of Terminator. So what happens here is, the machine here now realizes that there is this guy who is going to attack us.
So what he does is, what the machine does is, the machine actually sends a cyborg to kill the mother of the child so that the child is never born. So you can see that here is a machine which is able to predict the future. The machine is able to see that there is going to be some incident which is going to attack its survival. So what this machine does is, this machine tries to, you know, eliminate the source of such destructive event. So in this case, he is sending a cyborg back to the 1980s when the boy's mother was young.
So if the mother is killed, the boy will not be born, right? If the boy is not born, he is not going to be, you know, there to save the humanity from the machines. So again, you see here, there is a machine and this machine, you know, is able to predict. It is making a choice, it is making a decision from the surroundings. Then the next movie you have is Matrix. So for those who do not know what Matrix is, so the story is that you have Earth, Earth is completely polluted, there is too much of cloud cover, sunlight is blocked, there is absolutely no way, you know, to extract energy from the sun.
Machines realize this, they know that they need energy to survive. So what they do is, they find that human beings are the most efficient energy producers. So they literally use human beings as a battery. So what they do is, instead of our inverter, UPS, these machines create a very big battery bank sort of a thing. And this is powered by human beings. And how are they keeping human beings intact with that? Simple virtual reality. So how do you see, how do you speak, how do you understand things?
It's all because the information comes to you and it gets converted to electrical signals and the brain analyzes, right? For instance, if you are seeing the screen, you are reading introduction to AI, age of Ultron. What happens? This visual input goes via eyes, falls on the retina. Retina converts the optical signal into the electrical signal. And this electrical signal is processed by the brain. And you identify that introduction to AI is written, right? If suppose, instead of, you know, having a visual input, somehow if I am able to simulate the brain, maybe the optic nerve, into thinking that some input has come, which resembles introduction to AI.
Then even without the visual input, I will perceive introduction to AI, right? So that's what happens in matrix. So every human experience that the human being has in the movie matrix is all simulated. All the nerves, all the nerve endings, everything that goes to brain contains all signals, okay? So in fact, there is one scene where the hero walks through and then a lady passes by. And then when the hero feels, oh my God, this girl looks good. The computer engineer says, hey, that is my creation.
You see that red girl, that's my creation. He created the exact electrical input needed to, you know, create the red girl walking in the virtual reality, okay? So you can see that in all these movies, there is machines. Machines are taking over the world, okay? It sounds so scary, right? What do you think? Very scary, right? I mean, a lot of people might think, do we even need artificial intelligence if they are going to take us as slaves? It is very scary, right? Do you really think AI can do this?
Just pause the video here and take some time to think about it, okay? So let me, you know, give you an introduction to a funny event that happened maybe some three, four months back. So due to this COVID, all the spectator heavy events were canceled, right? So there was this football match. Already there were no spectators and they didn't want to have too many cameramen. So was they hired, you know, team, they made an AI camera. So this camera has to, you know, literally focus on the ball.
So this camera has to just, you know, whenever there is ball, it has to follow the ball. So what happened was the machine was trained, the camera was high-fied, the algorithm, everything was fine. But the only thing is what the AI camera did, it started following the referee's bald head because it resembled the football. So because of this, an entire match contained only the footage of the bald referee who was running around. So you can see, we are the people who landed, you know, on Mars.
You have Perseverance on Mars. You have Injunity helicopter flying on Mars. But still, you see, here we are standing, taking one and a half hours footage of bald referee. So that is where we are. But having said that, these are only few simple mistakes that have happened in AI, okay? Again, let me also introduce you to self-driving trains. If you take trains, trains travel in Wales, right? Trails are already predetermined and fixed. So ideologically, technically, you don't need a driver, right? Why do you need a driver?
Driver needs to drive you somewhere. For instance, if you are driving on a road, you need someone who is able to take decision, who follows the traffic road. But in case of trains, the rail is laid and the signal control is in the stations. Then why do we need drivers? Has anyone thought any time? The reason is, first thing is human lives are involved. Secondly, everything is fixed, man. Like you have the lanes are fixed. You even know the timing at what time which train will be where.
Even in Japan where trains are considered late if they are running by one minute, we know which train is where, at what time. But still! But still! But still! We need a driver, right? Why? Because even for such a mundane job, we don't, you know, have a perfect system which we can trust and, you know, live the entire human life. Okay, please go ahead. This train will not have driver and my human beings will be safe. No, we have not reached that stage. However, having said that, I would also like to say that by November 2021 or probably December 2021, first time Germany launched the driverless train.
So this train is completely based on artificial intelligence. There is absolutely no human control. But then there is a human driver just in case because it's a matter of human life, right? You cannot risk. So this guy is just going to sit in the room without doing anything. He is in the engine room. The train is navigating by itself. Okay, it is a driving by itself on the lane, detects the signal and goes on. Okay, so this is where we have reached in when it comes to artificial intelligence.
So what is AI? AI means artificial intelligence. The desire to make machines that can think and act like human actually drives the field of AI. So AI is the branch of study devoted to make machines behave like human beings. So these both are same thing said in a different language. So what do we humans do that we want the machines to do? Okay, we understand that you want artificial intelligence. We want a machine. We want it to behave like a human being. But what is it that we all do that we want the machine to do?
So let us see what we all do. We look at the surroundings and identify the object. Essentially, this is nothing but the visual input, right? You look into the screen. You see what is AI written on the right hand side. There are two fingers pointing at each other. One looks like a human hand. Other one is looking like a robotic hand. You see all this, right? This is because you are able to, you know, like your brain is able to visualize this from the input it receives from the eyes.
Similarly, we speak and listen to each other and convey our thoughts and feelings. For instance, if I tell you, hey, hi, how are you? You will reply back most probably, hey, hi, I'm fine. Or at least you will smile, right? We do all these things, right? Similarly, what do we do? We write mail. We use Google Translate. We use chatbots. We are able to communicate with each other effectively. So what does this mean? Our brain is able to process the natural language. We process the language. We understand it and accordingly reciprocate our feelings, right?
If you ask me a question, what is your name? I will reply an answer to you. Similarly, you ask me, what is the time? I am going to tell you the time. How do we do this? Because we have a common understanding that if somebody asks you, what is the time? It means that the question asked is, what is the time? And we have to reply the answer saying, maybe it is 5 o'clock or sorry, sir, I am not wearing a watch, right? This is because we have an understanding. Our brain processes that way.
Okay, this is something we can all do. Similarly, based on the cues from the surrounding, we take the decision. For instance, imagine you are walking on the road. You see a, you know, lot of pits and broken road. Definitely, you will not accelerate with a high speed, right? You know that your shock absorbers will be damaged. Your bike will be damaged. You will immediately go slow. Given you have to do the same thing in highway, I am sure you guys will be flying at a much faster speed, right?
So you look at your surroundings and you take a decision. For instance, you look outside, you see dark clouds, you know that, okay, chances are there is going to be a heavy rainfall. So what do you do? You immediately, you know, let me not, you know, put my clothes outside today or let me take an umbrella, right? So these are the decisions we take after looking at the surroundings. Similarly, we can think, predict some of the outcomes and identify the pattern. So we all know that if we don't study for the exam, we are going to fail, right?
And we all know for sure how much we have prepared. Our parents may ask, how did you do? You may say, I have done well. But deep inside, we know how much we have studied and we can, you know, sort of predict what would happen in the exam tomorrow if we go the same way without preparing. Similarly, we are able to identify patterns. Patterns as in, see, we get so many news, right, around us. We all sit together, not sit together. I mean, we inherently see patterns.
For instance, I know this is a controversial one, but I can tell you. So for instance, every time when Vivat Kohli says, you know, like, say whatever, unholy, we always know that it becomes controversial. There is some group of people who may be targeting that, why are you supporting this? There may be some people who are saying that he is speaking truth, right? Maybe Vivat is genuinely saying, maybe. We don't know whether he is being paid to say or whether he is normally saying. That is something we do not know.
But whatever he says, just because he is a celebrity, it gets blown out of proportion, right? And then you have great pundits who will find patterns saying that he is saying this because this is this, this is this, this is this. We identify the pattern, right? All of our gossip mongering is because of the fact that we identify the pattern, right? So this is something we can do. We do more things also, but I am saying the most important things. So now the main question is, can machines mimic human?
Can we do this? So currently, see, we are using deep learning techniques to identify the objects. When it comes to vision, you are watching the screen. But at the same time, if somebody throws an object to you, you can always catch the object, right? Imagine you are watching this, a reading machine mimicking human. Accidentally, your hand pushes a mobile and your mobile is going to fall down. Immediately, you know, you go and try to save your mobile, right? So we can do multiple things at a time.
But right now with deep learning techniques, what we can do is we can only identify the objects. Probably, you know, identify what is cat, what is dog. Similarly, what we can do is we can do face recognition. Face recognition is still a very tough job, okay? So those are the things which we have been able to do. Similarly, augmented reality, computer vision, virtual reality or some of the latest achievement of mankind. Okay, so what happens in augmented reality? Supposing, imagine that you are running a news channel and then you are like, you know, interviewing a minor child.
Now, as per rules, you cannot, you know, expose the face of a minor. So what do you do? You blur the face, right? Now the show might be real. Real in the sense might be running live. At that time, the child is moving here and there. So you have to exactly scan the movement of the child and those pixels alone should be blurred. Or, you know, you put something else on it and then for instance, some smiley or something, right? Just to hide the child face.
So these things come under augmented reality and computer vision. So these are some of the latest achievements which mankind have. So we also have, you know, face scans, thermal scans, analysis of biomedical records. These are the current achievements we have when it comes to visualization, okay? Then let us go to the next topic. So language processing. In language processing, what we do is we basically try to analyze the pattern in text and words in natural processing. So these are something available in chatbots, translator, grammar correctors, reply predictor, right?
So you can always see. When you go to Google Translate, suppose, imagine that you want to translate from English to other language. You write what is your name and if you like select Tamil, it says, or if you select Hindi, it says, or if you select Hindi, it says, you select Kannada, it says, you know, all that, right? How do you think Google Translate does? Can you make a guess? Actually, behind it is a very good, robust neural network that is like constantly, you know, working to get the correct translation.
So let me tell you, we have so many languages in the world. Similarly, in India alone, we have 800 languages. But in Google Translate, if you see, there are only like say 70 or 80 languages available for translation. And even in Indian languages, not all the 800 languages are available for translation. Hardly, I would say 10 or 15 major languages available for translation. Why do you think so? So not because Google is being selective, it's because the number of available volunteers who are willing to translate is less.
So what happens is, when Google wanted to create Google Translators, it asks for volunteers who could translate simple, simple sentence. So for instance, imagine I ask a question, what is your name? People might write different, different answers. So you know, all these things are fed into neural networks. Chances are that 99% of the people who probably written for what is your name? So the neural network learns that for what is your name, is the most correct translation. I can also complicate this, you know, like saying, This also indirectly means, please tell me your name.
But that doesn't translate to what is your name, right? Although the information conveyed is same. So what happens is, this will also go into the neural network. A neural network, you know, keeps learning, learning. And finally, it gives you the most optimal correct result. Now, from my knowledge, I personally feel that Google Translate is doing an amazing job for Hindi and Tamil. Other language, I do not know much, so I cannot comment. Maybe they are doing a good job. But all this is because they have a robust neural network that is constantly, you know, learning and learning.
Similarly, when it comes to, you know, chatbots. What are chatbots? You have a lot of websites, right? In the corner of the website, there will be a small chat, like pop-up menu. It will say, hi, how can I help you? Now, even in Skilllink, you will have, you know, a chatbot. Okay. So you have the chatbots. So what are chatbots? Chatbots are usually a small pop-up, you know, usually usually in the website. You can ask them any question and then they reply. So how do you think they are able to reply?
So for instance, imagine Skilllink is an yet tech and it offers a lot of, you know, courses related to engineering. So, I ask the question, hello, I would like to know some, you know, courses in the mechanical domain, you know, civil domain. It is automatically going to answer me, saying that these are the domain related courses, blah, blah, blah. Now, if suppose imagine that I write a very big story, sir, sir, sir, sir, I am an engineering student, I am from this place, I am from a very educated family, I want to learn a lot of things, I want to learn this, blah, blah, blah, blah, blah.
I am from mechanical engineering background, I want to do this. I can write a very big story also. Still, what the chatbot will do, it is going to finally direct me to this mechanical engineering course. So, how do you think these chatbots work? So, essentially, what happens is, these chatbots, you know, home on to certain words. Depending upon the presence of certain words, your direction, the where you are directed to change. Supposing I said, I want mechanical related, I will be sent to the mechanical courses.
If I wanted electrical, it would go to electrical. But then, having said that, it is not that easy, okay. Supposing, imagine that I am collecting a database of review about some company. So, I am making a very simple algorithm. If I see the word good, best, better, call it a good review. And if I say words like crap, not very bad, something like negative words, call it a bad review. Now, supposing, imagine that I have a customer who gives review, not bad, not at all good.
Then, what will the system do, right? So, you can see that, although, you know, it sounds very easy. Sometimes, it can happen that you may have a lot of negative statements, negative, negative statements. So, at this time, you know, you have to construct a neural network. And then, like, these neural networks needs to be very specific. For instance, you will have LSTMs. You will be learning that later. These kind of network that will, you know, look for certain choice of words. And it will try to predict what the next word would be.
So, in fact, in reply, predictor and all, such a thing happens. When you write a letter in Gmail, imagine that I send you a email, please submit your homework by today. Immediately, your Gmail is going to give you a predicted reply. Yes, I have completed. No, I would take some time. How do you think Gmail does that? It is because the neural network sort of predicts that if a question or a statement, please submit your homework is written. Then, the reply should be, maybe it is homing on to the word submit.
And it is, you know, homing just because the word submit was there in the input. The output predicted by Google is, no, I will take some time. Yes, I can submit by time, right? So, these kind of connections neural networks make depending upon the presence of some available keyword. Similarly, you have grammar connectors. Grammar characters like Grammarly or Ginger, they also have a very big neural network behind their software. A lot of people on everyday basis use Grammarly. So, their neural network keeps updating and every time you feed your input, it gets corrected.
Grammarly is one of the very good. It's a proprietary software, but the free version also does amazing grammar correction. Okay. Then, the next one we are going to go is, can machines make informed decision? So, now the thing is, with the machine learning algorithms, we are able to classify or predict some situation. It is not as efficient as human beings. For instance, when you are playing in computer, if your mobile falls, automatically you decide that mobile is more important and you know, stop playing the games.
So, these are some rapid decisions which human being can take. But, machine, that is not possible. Irrespective of whether something falls or not, it is going to do the same thing over and over again. Okay. So, yes, we cannot say 100% to making informed decisions, but yes, to an extent, we can give it a go. Okay. Then, how efficient is machine in finding patterns? Machine can identify some features, but it is not as efficient as humans. Okay. So, now let us teach the machine the human way.
Okay. So, how do we learn? We learn from experience, right? We use our experience to make decision and predict the next step. So, again, the experience we gain is either by ourselves or our parents teach it. So, sometimes it happens that, you know, we learn a language by ourselves due to close proximity with the particular language speakers. For instance, you learn your mother tongue by just being with your family members, right? Your parents did not teach you grammar, right? They didn't teach you how to, you know, use sentences.
We very easily learned up to 50 words per day when we were young, right? These are something we learned by ourselves. But then, again, there are some things that are taught to us. For instance, you are being taught that in a traffic signal, if you see a red light, stop. If you see a green light, go. If you see a yellow light, move with caution, right? This is something taught to us. Probably, your parents taught you or maybe you learned in school. But this is something taught to you, okay?
And again, there are some things which we learn by, you know, rewarding our punishment. For instance, imagine you are playing and then you see, you know, a beautiful verse. You go and pick it. Maybe this is your first time in your life you are seeing a verse. Then a thorn tricks you and you start bleeding. Or it pains. Then immediately, what do you learn? Okay, next time when I am going to pick a rose, I should be careful because a rose plant has a thorn, right?
So, let me repeat again. There are literally three ways by which human beings learn. One, we learn by ourselves. Just like languages, we just observe things around us and we learn. Secondly, something is being taught to us. Like in the case of traffic signal. And in the third case, depending upon the reward of punishment, we learn. So, if we get a punishment, we learn not to repeat it. When we get a reward, we learn to do it again. Okay, let us go to the next one.
So, mission two, you know, requires an experience to learn. This experience is the data. Okay, the data can be in any number of formats. The data is fed into the mission for learning purpose. So, it is possible, you know, to learn again in supervised and supervised and recommend, sorry, reinforcement learning. So, wait a minute. From where did this supervised and supervised and reinforcement learning come? This is what you will ask me, right? But I told you, right? There are three ways by which we learn. One is, we just learn from the surroundings.
That is unsupervised learning. There is no one to, you know, teach you anything. Next thing is, somebody is teaching us something in case of traffic lights. That is supervised learning. And finally, you have the reinforcement learning. Where you are being rewarded or punished for certain activity based on which you modify your behavior. Okay, so these are the three things. Supervised learning, unsupervised learning and reinforcement learning. Okay. They are also called as SL, USL and RL. So, here we are supervised, unsupervised and we in forced learning.
Okay. So, in supervised learning, what do we do is, we teach a machine what in the data and help it to learn. For instance, if your machine learning algorithm is to classify between cat and dog, what you do is, you take a lot of images of the cat and dog, feed it to the computer saying, this is cat, this is dog. This is cat, this is dog. In unsupervised learning, what do you do is, you just give the data of pictures of cat and dog and allow the, you know, system to identify the cat and dog by itself.
Then finally, you have a reinforcement learning. Whenever the system identifies the cat correctly, maybe you reward it. Then it starts identifying the cat better. Okay. So, now that we have learned the three types of learning, let us get into how to learn. Okay. Now we have the data types. So, data types, they are essentially like numerical data. So, most of the machine learning and the deep learning techniques would use numerical data. You have the data of weather parameters, real estate data, stock price data. Data can be numerical or categorical.
Numerical means numbers. Categorical means yes, no, good, bad, those kind of answers. Then you have the text data. Text data is predominantly used in natural language processing and machine learning. So, data of review of products, text for translation, question and answers. These are, you know, usual text data that natural language processing people use. Then original data is sometimes written or spoken. All those things can be, you know, converted into text format and fed into the machines. And finally, we have image data. So, image data is mainly for deep learning and computer vision.
So, the data is in the form of pixel values. The numerical values of the pixel undergoes mathematical processes and machine learns. Okay. So, these are the main different types of data. Okay. Three main different types of data. These are the three main different types of data. So, now we have machine learning and deep learning. So, deep learning is a subset of machine learning. I am saying you this. I know this could be confusing to you, but what is the difference between them? So, to know the difference between them, you have to understand something called features.
So, what is a feature? Do you remember, I asked, I was telling you about this identifying the dog and cat from pictures. So, whenever I show you a picture of a dog, you will be able to identify the dog. So, similarly, when I show you the picture of a cat, you will be able to identify the picture of the cat, right? How do you do it? Can you take a guess? Maybe you are looking for something that is making it a cat, right? Maybe it has a, you know, triangular face.
It has a here on top. Maybe it has gray eyes. It is looking very smaller. Maybe these are the features that, you know, distinguishes a cat more from a dog. Like most of the dogs are big. You do have small dogs, but in general, cats are smaller than dogs. So, maybe you are looking for some specific quality that distinguishes a cat from a dog. This distinguishing quality is called as a feature, okay? Remember, this feature with which you are able to identify an object, okay? So, what happens is, when you have a machine learning problem, you have...
Problem statement. You need to have, I am going to repeat this part again. Okay. So what happens in machine learning and deep learning is that in deep learning, we don't identify the features. In machine learning, we identify the features. Now this could be very confusing to you. You may be thinking, hey, is it not supervised learning and unsupervised learning, right? But we will get back to that. Now let us go to identifying the pattern. I show you this picture. You will immediately recognize that this is a cat, right? Let me show you a few more pictures. I am sorry, some of my own drawing, but I know in all the four of them, you will say this is a cat. In fact, the one in the corner more, right, is probably the first cat picture we all drew when we were in UKG or LKG, right? How is it that we all identify that this is a cat? It is because probably we are looking for the, you know, triangular here, maybe the gray eyes, maybe the entire fur, maybe those are the identifying pattern with which we identify the cat.
Okay. So what do we do? We even without thinking for a moment, we say cat. Our brain looks for some patterns and told it is a cat. So the problem is when you have a ready-made pattern with which you can make an identification, then we are in the machine learning domain. But then if we do not say the machine, what is what, like what is the feature and allow it to find the feature by itself, then it is a deep learning. Now you can ask me, is it not supervised learning?
In supervised learning, somebody told this is this and in unsupervised learning, people tells nothing and ask the machine to find things by itself. The answer is no. I will tell you why. So there are two things here. One is called feature. Another is called label. Feature means telling the machine that it has a fur, it has two ears, it has an head and it has a gray colored eye. That is called feature. Okay. Saying the machine that it has a cat. Okay. So when you give the features, two ears on top, it has gray eyes, then you are giving features. If you say the machine that it is a cat, then you are also, you know, supervising it. So you are giving two things.
You are giving the name of the object and you are also telling how to identify the object. So you can see it is like supervised mission learning. Mission learning, why? Because you are giving the feature, how to identify it. Supervised because you are telling what is what. So what is supervised deep learning? You just tell the machine it is a cat. Essentially, you are not telling the machine that it has two ears, it has gray eyes. You are just telling the cat. Now you do. You find whatever features you want from all the cat pictures and do the work. Okay. Then let us go to unsupervised machine learning. So what happens in unsupervised machine learning is you don't tell the machine that it's a cat, but you tell it that it has two fur, it has ears and it has gray eyes. Okay. So that is unsupervised machine learning. And in unsupervised deep learning, you neither tell it is a cat nor do you say that it has gray eyes or fur. So do you see that?
There are four levels. Okay. Let me write it down for you. Okay. Let me reiterate it to you again. So you have two things. One is label and another is features. Label means essentially you are giving the name of the object and features are just saying that you have to look for certain qualities to identify the object. So if you give the label and features, then this is a machine learning supervised. If you don't give the label, but give the features just alone, it is an unsupervised machine learning. Okay. If you don't give the label and don't give the features, it is unsupervised deep learning. You give the label, but don't give the features, just give the label cat, dog, then it is a supervised deep learning. Okay. So let me give you another example. Supposing imagine that you want to identify handwritten digits 0, 1, 2, 3. For instance, I am writing 5, 6, 7, 8, 9, 10, 10 like this, 10 letters. So I want my computer to identify 0, 1, 2, 3 like that. Suppose I tell the system, give the folder full of pictures of 0 and then I have another folder with pictures of 1 like that. I have say 10 folders.
Okay. Suppose I tell the system that this is all 0, this is all 1, this is all 9. Then it is a supervised learning because I am telling 0, 1, at least I have given the label. But if I also give the features that if something like this is circular, of course I am generalizing it too much, but in reality it could be a different set of feature. Okay. So if I am telling, you know, the machine to say, if it is like this, call it 0, if it is just a straight line, call it 1, then I am also giving it a feature. Then in that case, it becomes a supervised machine learning. Why machine learning? Because I am telling what to look for. Now I told you look for a curve, but in reality it is not like that. There are a lot of process. So all the zeros, whichever way I put, some people have the habit of doing like that. So, you know, there are many ways zero can be written. Similarly, some people write 7 like that. Some people write 7 like that. Some people write 7 like that. Some people write 4 like this. You don't know whether it is Y or 4. Some people write 2 like that. You don't know whether it is 2 or Z.
So, you know, feature that makes one digit unique from other. If you are feeding it to the system, then it is machine learning. Otherwise, if you want the machine itself to find the feature, then that is deep learning. Okay. So in the next topic, we are going to learn about how to create a machine learning or a deep learning algorithm. Okay. So what do you want to do? Why do you want to do? And when do you want to do? First, you have to identify the problem statement.
For instance, if your problem statement is to identify cat among other animals, then you must, you know, collect the data accordingly. So in this case, it is images of cats. Okay. Once you collect the cat pictures, what do you do is you have to check for bad images, clean your data by removing other animal pictures or bad quality images. And then you pre-process your data in some models also. Okay. Once you do that, then now you have to identify the model. Your task is to classify the cat, right? So preferably, you know, use a classification algorithm. Once you have chosen a classification algorithm, you just make a strong justification. Like why do you need this algorithm? This is, you know, just to make sure that you are in the right way.
Okay. Okay. And once you have decided on the algorithm, now what you are going to do is you are going to create a model and then train them with your data set. So how do you train them with your data set? So you take the data set, you split the data set into test and train data. Okay. The original idea is you train your model with the training data. And once when your model is learned, you use that model on your test data. Okay. Once you use them on your test data, you look at your predicted values and the actual values and then you compute your model's accuracy. If you are satisfied with the model accuracy, then you can use it for classification. Okay. So I'm going to just give you a brief of different kind of machine learning algorithms that are around. So in machine learning, only you have supervised learning. So here you are essentially, you know, teaching the machine what it is and when it is and you are asking it to learn. In that also, you have two types. You have classification classification and you have regression. In classification, you are classifying the objects. This is a cat. This is not a cat. In regression, you are trying to do a prediction. Okay. Similarly, in unsupervised learning, what you do is you have no labeling. Essentially, you don't know whether it is cat or dog. You just have images of pictures, images of various animals. You use various, you know, algorithms. So here what you do is you use various algorithms and unsupervised learning and try to do the same classification or prediction. Okay.
Finally, you have the reinforcement learning. In reinforcement learning, based on the reward or award, punishment, you know, your machine learns and then gives a better result. Okay. So these are some of the names of famous algorithms that are used most often. So you have KNN, which is KNMS, neighbors, SVM. Then you have decision tree and random for us, naive bias, neural network, SDGM. Then linear regression, logistic regression, polynomial regression, principal component analysis, high-wav-kill question. Independent component analysis, K-means clustering. So these are algorithms that have always been, you know, used more often. There are many more algorithms, but these are, you know, often used. Okay. Don't worry. You will be learning all of these in your machine learning courses. This is just an introduction. So you don't have to, you know, take time to, you know, memorize this. I am just seeing that these are commonly used. Okay.
Okay. So minimum requisites to process data and, you know, to set up an ML algorithm, you need to know mathematics. You need to know probability, calculus, linear algebra, basics. Then in statistics, inferential statistics, descriptive statistics, hypothesis testing, database management. Then you need to have, you know, very good programming skills. For database management, MySQL is a separate course. Okay. So with this, we conclude this week. I hope you enjoyed this session. In this week, we learned what ML DL AI is, how machine learns via data. Then we also gave a very small brief on various algorithms of ML.
And we also saw the minimum free requisites. So set up, you know, an ML algorithm. I encourage you, you know, to attempt the challenges. So in this week, what we do is the challenges are extremely general. It is essentially to make you to think. I would have just asked you, you know, to create, to think about applications in different fields where, you know, artificial intelligence can be used and made our life easy. Okay. I have not given you, you know. History of artificial intelligence and all that. It is because these are some things you can always find in Google. I want you to go and search those things. Okay. So have a look at the challenges. It's an open book challenge. You can refer any Google search engine or any research paper. Try to answer them. This will also, you know, open up your ideas and you would also start exploring very much. Okay. Thank you very much.