050621 Emerging Technologies AI and Machine Learning
to all the participants, I welcome you all in today's session about emerging technologies, artificial intelligence, and machine learning. Today's session is all about artificial intelligence and machine learning which are emerging technologies for today's era. So, without wasting much time here, I would like to welcome to today's Speaker, miss Maria. JOHAR in on behalf of Pyro POLYTECHNIC INSTITUTE University. welcome them to give a quick introduction about her. I would like
She's also holding a strong software development background in the past. work positions. She's currently work on various AI domains like conversational AI, Chatbot development, automation, workflow machine learning, and modeling based on the latest technologies like Python and our programming. So, she's also passionate about shaping the Young minds, feeding their curiosity with guidance, and equipping them with She have some faith in transforming lives with Motivational thoughts and moving them towards an optimistic approach for life. So, now, we are ready to start our session. Over to you, ma'am. Thank you so much Fatima Mam for such a warm welcome and I would like to thank for inviting me over for this session. So, let's start
So, today's topic is about the emerging technologies That is Artificial intelligence and machine learning. So, these are the words of the industry these days and if we go ahead and see what is this AI So, by definition, it is concerned with the design of Intelligence in an artificial device When we impart Intelligence in an artificial device to perform tasks is known as a Now, if I talk about the intelligent tasks, what do we what do we expect out of it? So, we expect something to perform like a human because human is the only entity that is intelligent right? So, there is always a What do do we say that is a controversy? between understanding of AI and a robot? So, what is a robot? So, robots are programmable machines that can autonomously or semi autonomously carry out the task So, robots are basically programmed and its they don't have their own any Intelligence They are program to perform a particular task. So, looking ahead, what is the difference between an artificial intelligence device and a robot? So, an AI device is programmed to think not thinking is the ability of a human and they have Data around when they have situations around So, they perceive the data and then perform in a particular direction. depending upon their understanding about the situation. They see that there is a danger around so they run They know that there is a hot object around. They
don't test that object. So, that is an intelligent task and robots does not have that ability to perceive the surrounding They are programmed to perform a particular task. So, they are programmed to do do a particular action Then, we have that AI has a solution interaction. Now, if I talk about the social interaction, then, we all know that there are devices coming up that are actually talking to the humans the AI devices. we all know about the Alexa. What is Alexa?
It is conversational AI in which we can talk to the device and the device gives us the answer and talks to us So, that is a social interaction. We can interact with AI and AI devices but in case of the humans can only have a low level of interaction. Now, if you, if I say that low level interaction, what does that mean? That means that hardware is programmed to perform a particular task right? and then AI Loans. Now, what is learning? Learning is we provide it with the data and based on the data they gain knowledge out of it. The launch
of the data and the trends and then understand how to perform and in the case of they're only smartest programs. Smartest programs means we have programmed to to perform a particular function So, that's what is the role of the robot Moving ahead. How do we define Intelligence? So, by dictionary, the definition is the faculty of understanding the faculty of understanding. understanding
the situations and the conditions around The human takes up a particular decision. in a similar way when we feel the AI device with the data the the device understands the data and the data trends and then take a particular action That's what intelligent means in AI. an intelligent Machine and can complete a task in the the presence of an unreliable and dynamic working environment. The data is not constant. It is
ever changing and so is the situations around us, right? We are comparing all through my presentation will be comparing a human with an AI device So, in case of humans in we ever changing situations around the nations are ever changing and we have to take our decision depending on the most optimum solution based on this situation at all, its depending on the risk. So, that is what how an AI device is programmed. to pick up the most optimum solution given a situation.
So, when the device is able to perform in an unreliable and a dynamic in the ever changing Data environment and it's able to take up a decision and a correct decision then, we say that this is the intelligent device Moving ahead, we talked about the understanding the faculty of understanding is Intelligence. Now, what is understanding? So, to comprehend something or to recognize it's significance, whenever we have this situation around, whenever we analyze it, like whenever we analyze it, and then try to understand to calculate the risk metrics in the normal human environment. So, that is what understanding is how you understand the situation and based on that, you take up a decision and Good decision. Your day. decides how intelligent you are. I hope that's good. Understanding
about the situation. and then taking a bus stand. or choosing a depending on the risk metrics. So, how intelligently you pick up the solution or how intelligent you behave in a particular situation? defines your intelligence in a human environment. perceive the data or understand the data. and based on the
understanding of the data how it is intelligently able to take a particular action So, that is all about understanding I hope that's good. So, this particular figure sum up the intelligence and understanding So, we have the knowledge. The knowledge is the data and the data. the
understanding basically of the data is the knowledge the understanding of the data is the knowledge okay? after having the knowledge after having the knowledge and perceiving the situation. or the conditions around whenever we take a particular decision, and that decision making is all dependent on how intelligent the device is, So, once a decision has been made, depending on the intelligence, The device takes action. and it's just very very similar. to working of a human and after understanding how intelligently we behave after having that knowledge, and then we take up a decision. It is a
very very similar thing. A human environment and an AI environment. and how they behave Moving ahead. So, one of the main AI approaches They give me various processes One is a top-down knowledge representation. The second is the bottom of machine learning but to conclude it, there is machine learning. and deep learning that helps us achieve artificial intelligence system.
And these are something the buzzword of the industry. We have been talking about the AI. the machine learning, and the deep learning. So, let's try to understand what is all this about So, you see this finger? in which the outermost flare is about the artificial intelligence and the inner circles are talking about the machine learning and the deep learning So, AI is the major part in which it encompasses machine learning and the day.
So, these are the two things that enable us to make an in the device So, that helps us in constructing an AI device Now, what is basically machine learning? So, it is a sub feel that actually relies on numerical optimization and the statistical approaches. So, that's what we do in a machine learning algorithm We have the data. We apply some high water derivatives and some mathematical equations to the conclusion That's how Machine learning algorithm algorithm works. So, there are three main types of learning. One is a supervised Learning Second is the unsupervised Learning and then we have the reinforcement learning So, in supervised Learning what we have is basically, we have a data that is labeled Data Like we have a particular pattern. given in
the form of input and we have a corresponding output to it. We provide that particular Data to machine Like if this particular pattern is coming, then label it as this right. So, if I say that a particular person entering a bank with all face cover with a black mask a black mask, not that mask which we are using during the COVID but the whole mask that thieves use to cover themselves so that they can escape the Pos So, if that particular piece is identified It might be an alarm. to figure out a thief. So, this particular type of patterns with the labor thing that this type of human is coming with a black mask Then we can label it as a trick. not directly as a thief but we can figure it out as a threat. It might be a threat
So, this is this type of data when we provide our machine It's called a supervised Data. and this type of learning when we provide to the machine, it's called the supervisor Learning. Moving ahead to the unsupervised Learning In this case, we don't have a label Data We have just the trends and then based on the action that the model performs, it learns continuously and figures out if this particular is the label So, we have an unlabeled Data and we figure out similarity between the data and we cluster them. that this particular type of data trends belong to a particular class on a particular label. and the other pattern belong to a different level or a different class or a different category. If I'm using a class or label. So,
basically, it means a cat, okay, Then, we have the reinforcement So, in reinforcement learning the model loans based on the on the on the instructor that actually gives a reward or punishment if the model is able to interpret, right? It is given a reward. and if not, then, a punishment in this case, reward and punishment are some mathematical weights that we give to a model to mathematical model, right? But for an understanding, we are using such as a reward or punishment. So, in that kind of scene, we call it as a reinforcement learning Moving ahead to the basic task that a machine learning algorithms is the classification clustering, and the prediction we have already talked about classification like a particular human that we talked about and tagging it as a threat that comes under the classification Then, we have the clustering. that we we talk and the unsupervised Data and the unsupervised Learning in which we don't have a particular label or a category but we try to cluster out in a big set of data We we divide the whole lot of data into different clusters that this particular cluster is belonging to a particular Data category. Then, we have the prediction.
depending on the trails, Trails would say that the data we have, we actually predict the outcome outcome can also be a number Like in case if I give an example, in an example, then, the prediction of the housing price, it's a lot of many factors depend in identifying the price of the house How many floors you have? How many bathrooms you have? What is the neighborhood? How many rooms you have? Kitchen. Ventilation thing. So, all these factors depend on the price of the house. So, even a model can protect that. given that we have certain features like I talked about the number of floors, the number of bathrooms, and more features that side the rate of the house I think that's good.
So, yes, we have another examples of categorizing or classifying the data We have a particular Data Check out how the data is about. Like if we have a lot many news articles, and we classify it into a sports category, a commercial category, on a Data News that's talking about the agriculture right? So, this comes under the classification where we have defined categories. It's a defined category and that's what we call it as a part of classification by in case of clustering, we don't have a defined category You draw patterns in a large chunk of data and they just say that this particular piece of data among the whole lot of data, a lot to a particular category like it's a whole lot of data is divided into different different channels Oh, I didn't defy it as an individual. and then in prediction. we talked about the housing prices. So, that's how it works.
Moving ahead to the deep learning algorithm. now, deep learning algorithm is the fastest and one of the most accurate algorithm How deep learning a lot of them Let me tell you. It works and the logs to human brain.
How are human? brains. based on the stimulus. the neurons pass on the data from one neuron to another. and on a similar note, We have the deep Learning Algo The each note is a neuron. and
we receive a stimulus and depending on the whole days, the transfer of the stimulus from 1 year on to another neuron depending on the threshold that's in the human brain. and on a similar note, in deep learning algorithm when a particular Data is received on a note, We use the optimization algorithm on the activation function. and after a threshold is met. the data is transferred to the next note. like from the first green. No to another red. No from red to blue and blue.
green. in a very simplified way. If you see. So, there's a transfer of data from the first layer which is an input layer.
to the last output layer which is also green in color here in the figure So, depending on the threshold, the data is transferred from one layer to another. And how does the learning take place? So, the learning take place in the back propagation When the data reaches from the first layer to the last layer. and it comes back to the input. so that each and every layer We checked the wheels. like, what is the amount of the chain in the weights that's observed that the data loss and we basically try to tune them to raise a particular optimum loss or minimum loss. So, we have to minimize the loss in terms of output. That is how a deep
learning algorithm works. Moving ahead. as a dog. has particular number of neurons and then you and Data transfers from one to another layer. right? and similar to other machine learning models, we talked about three types of problems there. The
classification, the cluster, and the prediction In a similar way, the deep learning functions So, there are particularly two main deep learning neural networks. algorithms, one of the CNN. That is a convo. neural network
and the second one is the recurrent Neural Network. So, in completion of neural network, It generally processes the images. So, when we evaluate the images, Neo CNN's and if I give you an example of a CNN of a convo neural network, so we can take a simple example of classifying a particular image. into various
categories of flour. So, I've drained my machine learning model. on a whole lot of flower images. like hundred images of roses, hundred images of sunflower.
hundred images of disease, and so on. We train the the model. that data. and whenever it estimates comes So, my model behaves intelligent enough. enough to identify the category of the floor. So, that is how CNN Works. It takes out the features from the images, The main features and then passes on to the next level. and identifies a
particular label. That's how CNN Works. Well, in case of RN, We have a particular text data and each particular token is identified. to make a prediction for the next in the sentence or the stream. That's what RNN are useful for. for expedition for the future.
prediction of the sequences. right? Now, if you talk about the domains of AI, What are the main taquitos? So, there are three main identifiable task domains. First is the mundane tasks. Second is the four Th. and the third is the expert task. What are the mundane tasks Perception. perception of
the data What it is second vision. who it is and identification. then speed. We are talking to the Alexa. speed
identification. natural language. understanding. what we are writing or whatever it takes is available in the If my A device is able to read that and understand it, and take out the knowledge out of it and then use it to behave intelligently. Then, we say it is a mundane task of a Then, we have the generation. Many of you have a data pattern There is one kind of problem in which we generate the pattern So, that comes from the dead. If I give you an example. if I
have a topic, topic is sports and particularly in sports, if it is stable time, So, if my AI is made so intelligent, let it understands the topic for the table tennis. and then the writes a particular article on it. So, that comes out of the generation. and that's the most intelligent task and AI machines are able to achieve that today. Then, we have the translation. We have manual translators. in the form of
humans. Similarly, we have the A devices train in such a way that they can translate it from one language to another. So, that also comes in the mountains. in the common sense. it is still unachievable. but scientists are working on it. Then the reasoning. How did you take action? Like, how do you perceive the So that reasoning. is also one
of the hardest task Moving ahead. We have the fours. So, we have the games, the computer games that we play with. That's a kind of you say intelligent tasks. We have the
mathematics. stars in which we give certain parameters and the machine calculates and gives us an output So, this kind of tasks are known as the forms Then, we have the expert in this machine performs accurate accurate. and as an expert. like we have the medical diagnosis. So, there are diagnostic systems that have come up that actually identifies the history of the patient. and suggest a particular line of treatment. So, that is what
medical diagnosis We even have the robotic surgery. So, the robots are trained. in a in such a way. that it can perform
a particular surgery. in the most exact way. So, that's an expert us Then, we have the engineering design. the
financial planning, the scientific analysis, all the export, Us under this category. Now, if I say what is easy and what's hard So, basically, it has been easier to mechanized many of the high level tasks that to perform like an expert It isn't easy. It is easy to basically devise a device that performs in a static environment. and achieve a particular task. Hey, given a
particular set of situations, if that's fixed, and the machine is strained to achieve that. So, it is easier to conquer that. like it has been very hard to make a noise. the tasks like human in this. The machine behaves as intelligent as human all the human activities to mimic those are tough. Give the example. walking
around and without running into things. So, this is a hard thing. a lot. made a perceptions are required in that was sensing. with the user vision, you need to identify how far is the object and then how much time you will reach near to that. so
that you need to take a turn to run into that right. Moving ahead. Talking about more applications. So, we have the manufacturing robots.
working in the manufacturing industry. in the self-driving cars. How does that work? identification of the traffic. the object, the car in front of it. calculating the distance.
and then managing the speed according to that. In fact, even identifying the traffic signals or other both sides and then taking up an action based on that That's how driving our book "smart" assistants CD Alexa are an example of that. They're trained on a whole lot of data to give out. output. and the
proactive health care management. depending on the data patient data take a decision disease mapping and the came up. There are not many scientists Who worked on those? just to see these scans. to identify the superiority of the COVID. That's some diseases. in the conversational marketing more They are the smart assistants. for ordering a particular item.
So, in two countries, there has been introduced by the ordering system based on the conversation. AA, They identify what the human is trying to say and what they need to order. and then completing the transaction. by delivering the So, that's how the conversational works in the field of particular industry.
Then we have, then we have the natural language processing. We talked about the extraction of knowledge out of the textbook leading a whole lot of data and then taking out the main points out of it and gain knowledge through it. based on the so that comes under the natural language processing each and every has their own way. to talk to from the sentences. So, that's another challenge and healthy. and we have certain
tools for NLP. So, building of those tools to ease the processing of natural language. is a part of application of a Now that we have moved to the end of the session, I would like to give some motivation tips. to the audience watching this. plan to achieve something. We
define the aim. It's not easy to climb up the ladder of success. on the first round. you see a level of fear. and that fear. is nothing tangible but is it is all cooked up in your head? It's all man made and that's all what you have made So, that first round climbing the first round. is a tough
task. because the level of fear is standing there you about the failure. right? the fear of failure. stops a person. to climb the ladder. and that's what you need to overcome your own thoughts of fear. and I
always pick up the example of the Edison thousand times. for building the electric bill. and he was never ashamed about his failure. He never stopped. He kept on working. and then one
fine day, the discovery happened. on a similar note, if I talk about myself, 3 years back when I was working on my research, I feel a many, many times and I'm not ashamed of it. I was also going through a lot of many personal things But I had that stubbornness to achieve my goal. My ML started with 52% of Accuracy. but I work day and night. to make it accurate. to
97%. I feared my failure. but still that stubbornness overcome my fear. That's stubbornness to
achieve. So, I would advice each and everyone who is watching this Keep your milestone. in your head. Keep smaller milestones. Keep on
achieving them. and strike off them of the milestone list. That's very satisfactory. Don't fear. Just keep on going. What if you fail? You're back to the wrong one. But what if
you reach the highest point? Don't fear. overcome it. and that's where I end my session. Thank you. We are having some questions like what do we have over here from the question is that we are having different languages like a CC java, Okay? So, in AI which which languages we are using to develop the programs we generally use Python and R. So, in industry, Python is more and we use Python more. So, if
someone is trying to work on machine learning and AI, then, get your hands on over. There's another question that you know, I mentioned that about the deep learning and learning, Okay. So, what is the difference between that like a machine learning and the deep learning Deep learning is a part of learning It is the as we talked about the deep learning functions as a human brain. it is a part of machine learning. but it is a more
accurate way. of achieving machine learning. in machine learning. you have the simple mathematical algorithms but the learning you have the models that imitate the human brain and the functioning of the human just not as statistical function or a linear of them. No it is a whole lot of model Inc many.
algori thms in it. I hope that answers your I have another question that you have in your in your presentation also that Chatbot, can you elaborate on what is the Chatbot means? Yeah. So, as I talked about the conversational AI, the charts.
So, in that basically, you talk to the AI system behaves as a diligent as a human. that actually understands what you're trying to say. and carry out a particular conversation. So, a chatting. is a Chatbot.
like I'm the answers coming from the world is not something mechanized. but it is able to understand what the human is trying to say. understand that based on the knowledge that it is taught and the understanding It gives out a particular answer. own way of defining the sentences. So, that's how the
chat bots work. They understand what the human is trying to say and then gives out a response Yeah. Okay. that is another question related to like how we can relate the game. The with the AI. That's a good question. If I
talk about the chest, There are all possible steps. that can be taken by a particular object in the chess in the chest bone. Let's define already the machine is strain on those particular steps. and if that particular steps come from the open, it how the machine has to behave. If there is a threat, threat to the king, right? Threat to the pond, So, in one direction to move That's the part of intelligence. based on the knowledge So, it's strain on all the various possible steps that a particular object of the chess piece can take. and how
to behave in case of each and every move of the open So, that's how games work. with all the possible combination of strips and then giving out a particular action That is another question like for the application like some of the people we are having the application like a Facebook, Whatsapp, Instagram. okay. In some of that, they are using the okay. So, how this space which they are using that will be an AI. How they are using on Facebook and Instagram or other applications. That's a good
question. It comes under the AI. biometrics. The hottest you're using for even unlocking of the phones. and the unlocking of the apps.
So, basically, it identifies the facial features the facial features of the human and then store it in the database. that if this particular trend this particular feature is identified, then, it belongs to this particular person. So, we have like a big data because millions of users are using.
this particular social media apps. So, we have different features. of all the users stored in the master Data. and
based on that on the recognition of the facial features, we are identifying the person That's how it works. and restore the principal components. The principal components means the identifiable and the most prominent features of the human face.
So, yeah, that's how biometrics or the facial recognition recognition books. Okay, ma'am. One last question that there are the main fraud detection of cases are going on. Okay? Like some we are what the application we are using. So, how we can use AI to the detection of the pros, Okay, that's a good thing. As I talked about first example, in case of the bank, So, in that case, if you talk about the broadly.
We can tap the IP addresses. from which IP. IP. address. Big question is coming and there are certain flagged it Good night. If the request is coming from the flag, I pay then we can market it as a threat. or
market. that this particular request is a fraud request. So, there are a lot many proxies that people use to bug this anti fraud department. The Department of a particular company takes care of tapping those requests. Those IP
addresses from where the request is coming. which country, and which is if this is a particular valid ID or not or if it's a proxy so, it stops. That's how it works. Okay. So, one last question from my side. This is personal
question from my side that how we can use machine learning in that blockchain thing I actually, I have not worked on blockchain. So, I think yeah, I won't be able to answer that question. Okay, fine, ma'am. No, it's you. Okay. So, I think students get the whole idea about artificial intelligence and machine learning and also how it works So, thank you so much for your valuable time, ma'am. Also, I would like to
thank our principals professors to provide endless Thank you. Thank you so much. Thank you, sir. Thank you for inviting me over. Welcome. You're welcome, ma'am.