 Okay. Now you can share the screen. I start it. Okay. Hello everyone. Welcome back on Sanjay Gupta Tech School. So we have one more session on Salesforce AI associate bootcamp. So like in last session also Nikita explained you about deep learning and neural networks. So today also she will be covering little bit about this and then we'll be having some quiz session. So we will discuss some question and answers related to the AI fundamentals that we have covered so far. Okay. So over to you Nikita. Thank you sir. Hi everyone. So in the previous few sessions we have covered the AI fundamentals plus some of the neural networks thing also. So the first section of your AI associate Salesforce trial had program has been quite accomplished and as you know that it is some percentage of all the associate program belongs to this one. So we are going to have a final conclusion today. Plus we are also going to see the quiz and attempt a couple of questions from the quiz because there are two questions each in the AI fundamentals topic. So that's what we are going to probably do today. So here we have some I mean you can just pop on to the presentations and all the videos that we have done. And I'll also make you a little bit aware of today's motive that we are going to do. So here I told you about the AI MNDL. So these slides are going to help you learn and have a bit about everything. Plus I'll now screen share the Salesforce trailhead website and I'm going to reiterate the things whatever we have done till date from those particular topics. And if anything is left out we'll discuss it there and then only correct. So just a moment. So guys whatever we are discussing here so it is already available on trailhead. So theoretically you can go and learn from there and in these sessions we have discussed them in detail. So if you want to if you like watching videos and understand want to understand the concepts from there. So I think this bootcamp will be very helpful for you. So we are going to share the screen. Yeah it is there. So you can see like in the last session. Yeah we discussed about AI fundamental like how to turn data into models. So yeah these these three sub modules are available here. So you can learn it from AI fundamentals. You can see that 17% of the exam weightage is on the AI fundamentals and then here we have got AI fundamentals. So this is what we have covered up till now. And you can see that there are three topics over here get started with artificial intelligence turn data into models and understand the need of the new networks. So you will have to have a little bit study of all these things in order to you know have a you know significant number of questions marked correctly in the quiz. So here I just have a re-iteration. We did three sessions like this is our day four. So in three sessions we covered these three topics right. Yes. Okay. Yeah I think we can jump on with the quiz directly. So I'll just quickly you know bullet points I'll just read them out so that they have a you know have this idea of what we are actually doing. So there was a these topics you can see it's learning objectives and then time to get into AI. So these are little bit things to learn. I mean just you can have a you know view of these topics it is not very intense. However the difficulty of defining AI. Yes the first session was based on ML and AI fundamentals. So you can really see the videos and you know you can absorb whatever I have mentioned and it's of the content whatever has been discussed in the videos has been is of importance to this section. So the difficulty of defining AI all the definitions of AI where it started from by what was the need of intelligence and you know what was the need of making a computer independent enough so that it can make its decisions. Everything has been discussed. You can please go and watch the videos. Then you have got main types of AI capabilities. So you have means what can I actually do. So the first thing is that numeric predictions. So I discussed with you that you know different salaries are based on the CGPA and let's say that you know if you have scored as much percentage in your graduation and on that we are finding your package. So this is what a prediction would look like. So you know we discussed a model for this regression model. Then classification. I had you understand the classification. What is classification. Then of course there are various other things like robotic navigation and the most important and the most high thing that's called chat GPT which was based on the L language processing. So on November 30 2022 this it could you know chat GPT is one of the most capable AI is built to interpret everyday language and act on it in some meaningful meaningful way. So you knew that chat GPT is something where you know you can just type your queries and you can get the answers for it like if you if you type in that you know tell me something about AI what is artificial intelligence. So it's going to tell you about what is AI. So what is it doing. It is understanding your language whatever your language you are inputting into the you know section inputting into the input box. It is understanding and then by that it is giving you all the results what has been fed into chat GPT. Correct. So this is known in the industry as natural language processing or NLP. There are various books which are dedicated to the NLP things and you know hence what we are doing is you know we are just reading our books and getting the best content possible to have in the survey and you know service. Now national language processing relies on understanding of how words are used together and that lets AI extract the intention behind the words. For example you might want to translate the document from English to German or maybe you want a short summary of a long specific paper AI can do that. Correct. So this is what NLP that is natural language processing looks like but there is a lot to understand in the NLP. It's not about this small thing. This is just a part of it. But a matter of significant importance shall be discussed later. In the summary it holds that artificial intelligence can be thought of as the ability for a computer to perform skills typically associated with the human intuition, intelligence and reasoning. So at this time AI skills are very specialized and fall into some broad categories like numerical predictions and language processing. And I'm pretty sure that Salesforce also has got a lot of things that are coming up in the artificial intelligence segment. It's not only the associate however this can be your first step into the artificial intelligence thing. Okay. So now let's get on to the enhanced quiz for it. What can distort our understanding of the artificial intelligence? So what do you think can be something that deviates you from the actual artificial intelligence thing? Like what they're wanting to know is do you understand the term artificial intelligence or do you actually upon reading their stuff have realized a little bit of artificial intelligence? So let's see what does it hold? Like superflares or an unclear definition of artificial. Now artificial definitely is one word, but it's not the only thing that you have to be worried about. Hence A and B are actually very discarded options. Then you go on to the C which says that fictional representation of the AI. So of course, if you are taught about artificial intelligence in a manner that is only hypothetical way of explaining AI and not a practical way of explaining AI, then this certainly can distort the understanding of the artificial intelligence. Or what can be the other thing that a narrow view of what constitutes intelligence. So to understand the artificial intelligence, we are first told to understand the intelligence. So if you only if you understand the human intelligence well, can you be able to understand the artificial intelligence well. Now it is very much true that nobody till date has been versatile enough or capable enough to dive into what actually human intelligence look like because there has been a lot of work, a lot of work. But still there's a lot of opportunity for you to catch up with the human intelligence because neurological things have not yet been so appropriate that you are understanding the human intelligence well. So I believe that C&D options would work well. Let's see. So the answer I think should be C&D. And then we have another question. So which broad category would an AI system fit into if it's used to determine the optimal price of an airline ticket? Here is where the videos come into play. So we discussed about regression models where I told you that your salaries and your house flats, you know, the house that you're going to buy is also a thing which is giving you the output in the numbers. Like, of course, a house is priced at 50 lakhs, 55 lakhs, 60 lakhs. So it's a continuous distribution of numbers. So the prediction is all about numericals. It's not a yes or a no. It's not classification. Hence, it's not choosing numbers from four numbers given. Hence, it is not classification. So what we can say is classification, so definitely it is not there. Then robotic navigation. No, certainly not. I don't think we require robotic navigation for which broad category. So we can't put that into robotic navigation. Like, if you want to figure out the airline ticket price, why would you put it into the robotic navigation part? We just want to predict the numerical output over here because airline tickets look like 4,000, 5,000, 5,550, 8,780, like if you go from Bangalore to Jaipur, the ticket might just cost 8,500 or something like that. So these are all numerical predictions. So that's why I think it should be numerical prediction. If you want me to discuss the last option as the language processing, so I told you that language processing is dedicated to making some applications which result into understanding your language and giving some output, which is something written output for your language, whatever you input. So your language processing also doesn't play any broad category role. So I believe it should be numerical prediction. Hence, I'm marking A. If you have anything else to say, you can probably comment and we can just check whatever we have scored so far. Okay, so I will vote the options for today. Yeah, I think we can jump on to the next one. Yeah. Yeah, you can click on top like if you see in the middle, we have a drop down. Okay. Yeah. Okay. Now we discussed about difference between the hand coded algorithm and train models. Now I again told you that when you need automations and all, then certainly we are looking for hand coded algorithms because it is in our control, whatever we want a machine to do, it is going to do and hence we are going to hand coded. Again, however, if I'm going to give my model a data that is a training data and the machine is going to look at it and I'm not looking at it, just process that data and come up with the predictions that it has to predict, then that's called a train model. Right, so this is what I had explained it to you and you can just have a quick read of all of it over here. The next topic is defining the machine learning and how it relates to AIP. So this is also in our videos, the first and the second videos where I have shown that machine learning is a subset of the artificial intelligence in a full view. And then you have forgot for the subsets as DL, neural networks and all as the major and minor subsets. And then you have got distinguished between structured and unstructured data and how it affects training. This is what we are going to pinpoint today. I'll tell you when we reach there. So you can see that there are various things over here. It looks like a semi-perceptron model where we have got some inputs and we've got an instinct over here. So after several trips, it's just an example given over here to explain a perceptron that how much weight it should have been given to a weekday or any other things that are discussed over here. So you can just have a quick read. You will be understanding everything pretty well. Now here, this is what I want to read. Use the right data for the right job. So this is very simple example of using training to make an AI model, but it touches some important ideas. First, it's example of machine learning, which is process of large amount of data to train a model to make predictions instead of hand drafting an algorithm. So machine learning is what is making our models more capable enough of getting the training rather than getting a hand coded algorithm. Second, not all data is seen. So whatever example has been discussed above, the spreadsheet is what we call a structured data. Why do we call a spreadsheet as a structured data? Because we know that everything in a spreadsheet has got a particular column, particular label and everything is very much ordered into it. That's why it's called structured data. However, when the data is not organized, then what is it called? So it is well organized with labels on every column. So you know the significance of every cell in the structured data. In contrast, unstructured data would be something like news articles or unlabeled image files. This again has been explained in the day two video where I have told supervised and unsupervised learning. In unsupervised learning, we do not require labelled data and the models are trained without the image labels or anything without the labels they are trained. So this kind of data is called unstructured data where you do not contain any labels, any categories, any organization. It is all unstructured data. Third, the structured data from our spreadsheet lets computers do supervised learning. Please focus on this. This is very, very important and I'm sure when server attempting the AI associate certification examination, he must have gotten some questions related to the supervised learning and supervised learning. So I'm kind of sure that he must have made them all correct observations with the correct observations. So it's considered supervised because we can make sure every piece of input data has a matching expected output that we can verify. Conversely, unstructured data is used for unsupervised learning which is why AI tries to find the connections in the data without really knowing what it's looking for. So this is what unstructured is related to and structured is related to. So please look at our second video. You will learn about supervised and supervised learning. So let's have a quiz now. What limits programmers from hand crafting algorithms to perform tasks we associate with the human intelligence? So let's understand the question first. What limits programmers from hand crafting algorithm to perform tasks we associate with the human intelligence? Is it like not having enough memory in the modern computers or is it like the laws that prevent the creation of artificial intelligence? Now let me be specific. There are not very hard and fast laws in the AI. However, certain ethical things have to be followed and some lawful acts have to be taken into consideration. Of course, you cannot be illogical or unlawful and unethical in any of the artificial intelligence works because certainly you will be found out. Hence, this is our next topic next to next maybe where we will understand the ethics with the or ethics related with the artificial intelligence. However, here not much of the laws we have discussed and there are no hard and fast laws that have been enlisted for the artificial intelligence. Then you come on to the sheer number of rules to account for many of which are unknown. Yes, certainly there are certain number of rules that we have to be account for and many of which are unknown. So out of the given four options, we will choose this is the most likable option. Last is too little coffee, too little time. It's none of the concern because it's not relating to the question that has been imposed. So what limits program is from hand-crafting algorithm to perform tasks we associate with the human intelligences? Either they are too lengthy because hand-crafting the algorithm can be really tedious and monotonous task for you. In case if you do so, it would take some nights and some days for you to actually come up with the final output. However, something that is related to the human intelligence would do training and come up with the outputs in a better way. So I believe that the sheer number of rules to account for should be the right choice. Lastly, you have got two of all. So a database of business names, zip codes and market values would be an example of structured data. So you can see that they have shown and they have told you that a database of business names, well, you have got business names. You also got the second category, the second label for it that is the zip code. And the last is you have got market values. So this would be an example of structured data. Definitely that's what I believe because if it was unstructured data, the labels wouldn't have been given. So we have got business names, we have got zip code, we have got market value. These are all labels given to the data that we are having. So let's see how is it going. Yeah. Yeah. And so we've got 200 points so far. So I just hope that you understood the meaning of unstructured unsupervised structured supervised learning. And let's move on to the last one, which is understand the need of neural networks. Right. I think we covered it in the last session. Yeah, the last session was based on the neural network theory. Here, many questions can pop up actually based on, you know, Perceptron, which is the inspired model from the human intelligence. That is a human body, which is a biological name called neuron. So these neurons can be, you know, of vital importance. Still, I told you that the study is not enough. But whatever we have studied in the neurons, we have been able to apply and implement into the perceptron as the neural in the neural network. The perceptron, however, is a very basic model, and it is not of use if you use it singularly. There are a lot of perceptrons with a lot of hidden layers that are used in the neural networks. The neural networks are used for bringing out the output for the nonlinear data because linear data we have got a lot of models. Now we will look for the nonlinear data and we will have some implementations for the nonlinear data. So let's see the objectives to explain the limitation of AI models that only consider the weight inputs. They can describe the role of neural networks in machine learning, define the major components of the neural networks. Then you have got to describe how complexity is added to the neural networks and define deep learning. If you have seen the previous video, I told you that there are a lot of hidden layers and based on the output that we are arriving, that's how we decide if we need to backpropagate or we need more hidden layers, whatever is required, that's what we do to the neural network model that we create. Then you have explained how it's possible to interpret the weights and biases determined through training. Well, so as I mentioned, there was a word called backpropagation. That's something through which we realize the weight of the required entity. Suppose, again, I'll take that same example of salary. The salary is based on the percentage that you had in your graduation or whatever, you know, whatever CGPA. So these are all the three things, you know, based on which mostly you get placed. Then similarly, you have got projects, suppose you do projects. So our model should know that what is to be given most importance. Is it the projects that have to be taken more importantly or is it the CGPA that I have to take more important? So that weight is based on these particular things. Whatever is going to be most important, that is going to carry a significant weight. If there's something that is not important for you, not important for the model, then it will certainly be disqualified, right? It will have less weight and that's how our output will be managed. And if we still see that there are some problems with the output, we are going to again go backpropagate and just check what is the weightage problem that we are having. And then we are going to see the outputs and then until and unless we match with the predicted output, we are going to work out it. So the need of neural networks was no conversation about AI is incomplete without mentioning the neural networks. It's better that you go and watch the slides, that is better. Here you can just come up for the examples that they have given to make the... We can just go to the quiz because this we already explained in the previous video. So I think it will be a repeat for them. Yeah, sure. So we will see the quiz then? Yeah. Okay. So for the artificial intelligence change to be considered deep learning, what does its neural network need more of? So do we require more nodes? Do we require more weights? Or do we require more layers or inputs? I just now had a statement that what is required in the neural network? The more the layers that you have in the neural network, but also the optimization that you don't have to create a neural network having more weight, but less optimization and it is taking so much time in coming up with the output that you are just waiting and the things are not working. So you just have to have an optimized output at a significant time having some good number of layers which are able to judge the creative output that you are having. So for AI training to be considered deep learning, I believe that it should be layers that you have to... The hidden layers have to be more and hence the output would be the optimized output. And then last you have the values of weights and biases in a trained neural network usually have an obvious connection to the inputs. Okay. So now to be very precise, the values of weights and biases in a trained neural networks do not have any obvious connection to the inputs. The inputs are not connected to the weights or the bias that the model decides or whatever the machine decides. So they are not connected. The bias is given randomly and you have the output related to the neural network in a manner which can be considered as the output and the whole point of this explanation is that the output is though connected to the whole neural network but the values of the weights and biases in the neural network usually do not have any obvious connection to the inputs. So let's mark it up and let's see how is it going. Yep. I think it is correct and one batch is completed. Right. So this was... I mean you can just read the explanation also. So adding the new layers allows for the possibility of surfacing meaningful connections that are not always obvious. Finding the optimal number of layers is a part of neural network design but more than one is required to be considered as deep learning. As I already mentioned that you should have an optimal output. It's not that you are keeping on waiting and waiting and you are not arriving at the output. That's not what we require. So optimization has to be taken into importance and then if you see the second one... Yeah, I think it was true also. Okay, it is also I think. Yeah, we have the explanation for this also. As a neural network is trained, it adjusts weights and biases so that it produces the optimized output not to make sense to us as observers. As deep learning gets deeper, the connection between the nodes become impossible to interpret in a way that connects to the input. So this is about true and false. However, I do believe that neural networks... It's just the surface study of the neural networks. It is not the holistic study to be very honest and if you really, really want to dive into it, you have to prosper into the path of deep learning and data science towards that side if you are a one who are wanting to transition between the surface matter to the deep learning and the insights of data science. So that's where you will understand actually why and how is of the biases and the weights are playing important. Okay. Yeah, I think we covered all the quizzes. So nice explanation. I think it will help if anybody is preparing for AI associate certification. Yes. Yeah, and do we have anything else to cover? Of course, one second. I was just about to tell them that it is however just the 17% age of contribution has been made as in we have learned. But then next coming up topics are AI for business and customer service and generative basics. My next topic would be generative AI and predictive AI. So we shall cover all of this, you know, whatever you can see over here. The AI basics, generative AI, generative AI versus predictive AI, then you have national language processing basics. So this is what we are going to cover next. Probably in the next video, we might just cover generative AI and predictive AI that full. And then we will mark up all the quizzes probably the way we did today. So firstly, we will have a presentation view. Then the next day we will have a quiz view where we will solve all of these things regarding that. So this was about the Salesforce quiz section. Now we can move on to one of the things that I wanted to mention because some of the questions can arrive from the Perceptron model also. So I would just like to reiterate some of the things that we had discussed and a propagation video, a propagation day of how actually the models are in the cycle of ANN model. This is what we are going to see, right? So just a moment, I'll share. If you can see the data here. Yeah, it is visible. Okay. So somebody who wants to go in depth of the ANNs that is the artificial neural networks can actually understand this part that ANNs are powerful tools in the machine learning toolkit and they are using a variety of ways to accomplish the objectives. As I've discussed over here, whatever objective you are aiming for, like image recognition or anything like that, language processing, natural language processing and then in the game playing as well, they have got a huge contribution. So ANNs learn in a similar way, just the way machine learning algorithms learn and that is by training data. It is better for you to give them the data sheet, give them, I'm saying, it's not give them, it's actually the models read data and the models are, you know, we are going to take example of the model, put it into the Python format and then we are going to run the algorithm. We are going to use different libraries for it and execute by inputting a lot of data. Data can be found anywhere. Data sets can be found anywhere. I'll put some, you know, coming up sessions shall be definitely based on how I am taking the data from different websites and, you know, putting them into the Python files and executing all of them. So they are best suited to unstructured data and where it is difficult to understand how features relate to one another. That's the best, you know, place where you can use the neural networks. So to gain a clear understanding of how ANN is put into the bigger machine learning landscape, we would review. So here is your review that what actually happens in the ANNs. Now look at it carefully. The collection and understanding of the data has to happen. As I mentioned, we will first take the data and put it into the Python files and then prepare it up then our model that is going to be there, whether it be regression model or whether it can be, you know, any sort of model is going to be there because it is also the lifecycle of the machine learning. So you can, you just put this whole lifecycle because there will be some questions if you are from, you know, college or school where deep learning or machine learning is being taught. Usually colleges are also teaching the machine learning and artificial intelligence. So lifecycle of the machine learning you can learn from here. So this is how you can actually dive into the process of DL and ML. So collection and understanding of the data, then preparing data, then you have to train a model, then you have to also test it because if you won't test it, you will not be having an eligibility for imposing that data or implementing that, you know, model into any of the industry works. And the last is to improve the accuracy. As I told you about the neural networks, the best way of improving the accuracy is by back propagation. So, you know, that is the way you can just see what is the output. If the output is not good, go back, change the weights, change the bias, come up with the desired output unless the predicted output and your output is matching, you can go on making it more accurate, more efficient neural network. So this is how it works. This is the basic lifecycle of the neural networks. Then this perceptron model, you can take it down because again, if you are a person who is learning A&Ns and all the machine learning things, so, you know, perceptron model looks like this. So a neuron, let me just take it here, yeah. So the neuron is a fundamental concept that makes up the brain and nervous system. It accepts many inputs from the neurons, from other neurons, processes these inputs and then transfers the result in other connected neurons. So this is how our biological neurons work because certainly there is not only one neuron, there's a chain of neurons present. So if you look into your scientific things, this is how the neuron functions. Then we'll go on to the A&Ns and see how they work. So A&Ns are based on the concept of perceptron. What is a perceptron? It's a logical representation of the single biological neuron. Like neurons, the perceptron receives inputs. Like then writes, of course, you can go and check the video where I have explained the dendrites. Then alter these inputs by using weights like synapses, then processes the weight inputs like the cell body and nucleus and output results like the axons. Axons is where you get the outputs and biologically you get the reflex from the axons. Okay, so you can see then writes is the receivers, then synapses, then we go on to the sum function which is the cell body and activation function which is the nucleus. And then we finally get on to the output which is at the axon level. So hidden nodes are here. They correspond to the sum function or cell body, right? So I have given you this explanation also. You can just check over your inputs, weights, hidden nodes and output. This will certainly help you in getting more quiz answers, correct? The components of perceptrons are divided by variables that are useful in calculating the output. Weights modify the inputs. The value is processed by hidden node and finally the result is provided as the output. So here's a brief discussion about components of the perceptrons. Inputs describe the input values in a neuron, these values would be an input signal. Second is weights. So we need to describe the weights on each connection between an input and the hidden node. And weights influence the intensity of an output and result in a weighted input. We will have to multiply the weight by the input to get the weighted input. In a neuron, these connections would be called as synapse. Then you have got hidden nodes. More of the hidden nodes and more of the hidden layer better is your network. So sums the weighted input values and then applying the activation function to the sum to result. One of the activation functions that you can apply is sigmoid function. An activation function determines the activation or output of the hidden neuron. Last is output which describes the final output of the perceptron. That's it. This is all about neural networks and perceptrons. You don't need to study much about it in the associate program that you are preparing for. So I think this is... It is a good summary. So if anybody will be following this whole explanation and the document, I think they will be having the crux of neural network and other terminologies. Right? So I think AI fundamentals are covered now. Right? Yes. And as you mentioned, in the next session you will be starting with AI, right? Yes. Okay. Awesome. So I think those who are watching this bootcamp are following all the sessions. So I'm sure like if you watch all the session properly and listen what Nikita explained, so you will be able to understand the concepts and basis on those concepts you will be able to clear the exam. I have also cleared AI associate exam and the material which Nikita prepared. So I followed those and prepared terminologies and in the examination like those those really helped. So along with trail head, if you watch these sessions so I'm sure like your learning experience will be great. Okay? So like keep following all the sessions and having one more session, right? Which will be on Generative AI and like it is another session if you are talking about AI associate certification. So this bootcamp is basically targeted to AI associate program, right? Whatever is provided by Salesforce as part of certification. Once it is complete, so like we are in discussion to have another program that we will be bringing for you because here we are discussing all the things theoretically, right? So correct me if I am wrong Nikita. So we will be doing in another bootcamp practical implementation with the help of Python, right? So we will be covering Python first and then we will see like how practical implementation we can do the things which are related to AI, ML, TL. Okay. Thank you. Thank you for sharing your knowledge and we will connect again next week for another session. Okay. Thank you. Thank you everyone for joining the session.