Describe Machine Learning - AI for Everyone - video Dailymotion

SkillTech Club

Explore Machine Learning (ML) in this insightful lecture! Discover what ML is—a subset of AI where systems learn from data—and dive into its process, from data preparation to model training and evaluation. Plus, uncover the power of Deep Learning, a key ML technique driving innovations like Azure AI. Ideal for mastering ML fundamentals!   Explore our other Courses and Additional Resources on: https://skilltech.club/   Stay updated with real-world tech skills  Follow us on other social media  LinkedIn: https://www.linkedin.com/company/skilltechclub  Instagram: https://www.instagram.com/skilltech.club/  Facebook: https://www.facebook.com/profile.php?id=61572039854470  YouTube: https://www.youtube.com/@skilltechclub  Reddit: https://www.reddit.com/user/skilltechclub/  X: https://x.com/skilltechclub    #Machine learning #AI #ChatGPT #SkillTech Club #Copilot #Techfriday

Transcript

00:00let's move forward to the next thing where we need to understand machine learning conceptually

00:13okay now this slide is actually talking about what is machine learning now obviously we know

00:19that machine learning is going to create predictive models for us but what I want you to understand

00:23here is the process of machine learning and that's why you can see here we have stepwise

00:27one two three four steps step number one is actually talking about that we are going to

00:33have a data which we have collected from the past observations now when I say past observations it

00:39can be a data which we have collected from last few weeks or maybe last few years this data we are

00:45going to provide as an input into this particular process most of the time this data is going to

00:50have two things inside that is going to have values which normally we call features and it's going to

00:55have labels associated with those values so features and labels are going to be there as

01:01a data this values and the label associated with that is what we are going to add on into this

01:06particular system and the second step which is actually going to be some kind of algorithms are

01:12going to be applied on that so on top of your data we are going to apply some algorithms and the goal

01:18of this algorithms is mostly going to find out the relationship between X and Y so how this values are

01:24actually related to that particular label is what this algorithms are going to find out once this

01:30process is done we are going to get something which we call predictive models now this machine learning

01:35models are basically nothing but some kind of a logic every time we understand this thing that every

01:41time when we have a logic we provide some kind of an input we execute the logic and we are going to get

01:46some output this is also that kind of a logical function this function is going to take your values

01:52as an input and once you provide this values as an input into the function is going to process that

01:57and it's going to give you an output which is nothing but the predicted label this process is what your

02:03model is going to do and that's why we call it predictive models now every time when you're using machine

02:09learning this is the same process which is going to be followed and ultimately you're going to get one logical

02:13function which is your model now the obvious question people ask me at this time is Maruti

02:19what kind of algorithms are going to be applied in that well most of the time these are going to be

02:25statistical and mathematical algorithms but if you want to understand this algorithm properly you need

02:30to understand something which is known as types of machine learning so we understood the process somehow

02:36very high level overview I'm showing you right now but we understand the process of machine learning

02:40with these four steps if that is a case then we need to understand something which is called types

02:45of machine learning now this slide is actually showing you that there are two types of machine

02:51learning which is available you have supervised machine learning and that is something which is

02:56going to be based on a data which is having known labels associated with that so basically if you have

03:01data with labels you're going to do supervised machine learning and if you have data which is having

03:07unlabeled data with that so basically it's not having labels then you're going to do unsupervised

03:11machine learning now both of the types are equally important let's say we want to understand and do

03:18supervised machine learning first in supervised machine learning we have different types again

03:23so we have some types like regression classification now each type here is actually focusing on

03:29predicting some kind of a label associated with that for example regression type of algorithms

03:35are actually focusing on predicting a label which is always going to predict a numeric value associated

03:40with that which means is always going to give you one answer which is nothing but a number let me

03:46give you an example to simplify this process like in this I'm showing you an example that you want to

03:51predict the number of ice creams which are sold based on the particular day now let's say I want to find out

03:57this thing for an ice cream shop business and if I provide some data like current temperature weather

04:02information or some festival kind of a date then when you do this thing we want to see how many number

04:08of ice cream will be sold on that particular day obviously the answer of this is going to be one number

04:13the number of ice creams or maybe if I give you another example let's say you want to find out a

04:19price of a property which you want to sell so let's say in the real estate you want to sell the price of

04:24the property and you're going to provide some kind of an input information like size in the square feet

04:29number of bedrooms or maybe some social economic metrics associated with that but when you're

04:34providing this thing is going to give you the exact price on which you can sell that property now that

04:40selling price of the property is also nothing but a number so when you have this kind of things you're

04:44going to use regression kind of algorithms and it's going to give you a predicted number associated

04:49with that which is always a numeric value on the other hand sometimes we have to do classification

04:56now classification is always going to predict a label which will tell you that a particular data belongs

05:02to which class in that also we have further types like we have binary classification and multi-class

05:08classification as the name suggests binary classification is going to find out that the particular data

05:14belongs to which class either one or two is always going to have two options in that

05:19and then it's going to be mostly like true false kind of thing like I'm giving you an example here

05:23that let's say you want to predict whether a patient is at having a risk of diabetes based on

05:29this clinical data so let's say you're going to provide a patient information data like his age

05:34his weight his blood glucose and when you provide this details is going to predict whether the patient

05:39is having a risk of diabetes or not now obviously there is no third option here it's going to give

05:44you either true or false kind of two options and one of them is going to pick so that is why this is

05:49something which is binary classification on the other hand sometimes we have more than one class

05:55associated with that and at that time you have to go for multi-class classification let me give you an

06:00example of multi-class classification also in this case let's say you want to predict the species of a

06:06penguin based on this measurement now penguins are actually having different species like we have

06:12species like Adelie, Gentoo, chain strap now don't worry I'm not very expert of penguin species but in

06:19this example let's say we have different species of penguin and then we want to provide different

06:24kind of an information like I want to provide penguins flipper size or maybe I want to provide penguins

06:30weight penguins height and other information when I provide this thing it's going to predict that this

06:35particular penguin belongs to which species now this can have multiple classes so it can have

06:40multiple species associated with that it's going to pick one of them as a predicted label or if I give

06:45you a simplified example don't get confused with penguins let's say everyone's favorite is movies so

06:51let's say you want to find out a genre of a movie based on some input values which you're going to

06:56provide when you provide the movie details is going to give you whether this movie belongs to

07:00comedy horror romance science fiction is actually going to give you this information

07:05and that is also nothing but multi-class classification because ultimately that movie

07:10belongs to one of the class from the multiple options so when you have two options two classes

07:15go for binary classification when you have more than two classes you go for multi-class classification

07:21this is how simple it is then sometimes as we said we do not have a data with labels and when you

07:28have unlabeled data you have to focus on unsupervised machine learning one specialized type of

07:33unsupervised machine learning is clustering in which we try to find similarities between items

07:38and we try to group them now this clustering kind of algorithms are mostly going to find similarities

07:44and because the data label is not there the initial goal of clustering kind of algorithm is to find the

07:50initial label associated with that for example I'm giving you an example where we want to have a data of

07:57let's say separate separate plants let's say you have a data of hundreds of plants information and you want to find out

08:03the similarities between these plants and based on that you want to group them first

08:07so I'm going to do what I'm going to provide some data like what is a what kind of a flower this plant is having

08:12what kind of leaves are there how many leaves are there what is the size of the leaf

08:17what is the size of the petal now all this information we are going to provide into this

08:20and based on that we want to group those plants based on the similarities

08:24at the end of this based on the grouping you are actually going to get at least similar

08:28three to five groups and then based on those groups you're going to give them some label

08:33that initial label is going to help you to further process the data using supervised machine learning

08:38so basically this is how the process works if you ask me this obvious question which people ask me is

08:45that how many total number of algorithms are there in machine learning well there are hundreds of if

08:51you're using something called azure machine learning kind of services and all you'll be able to use those

08:56machine learning algorithms but right now it is out of the scope of this particular thing so I'm not

09:01going into that right now we have different types of machine learning algorithms and based on the goal

09:06based on the label which you want to predict you can actually pick which one is perfect for yours

09:11now once we have this information that what is machine learning and how the process is working

09:18and what types of machine learnings are there you need to understand this slide now this is actually

09:23a slide which is combining whatever we have understood so far if you focus on the slide I have training data

09:29which is in the step number one but remember in this data where we have features and label initially when

09:35we do the machine learning model training process it's actually going to divide the data into two parts

09:40we are going to divide data into some kind of a ratio let's say I'm saying 80 20 kind of a ratio

09:45I'm going to do 80 percent data I'm going to provide for model training and 20 percent data I'm going to

09:52keep it aside for validation purpose when we are doing this thing this 80 percent data is going to be

09:57used in the process and on top of it we are going to apply some algorithms once the algorithms are

10:03applied once the process is done it's going to create what is going to create my model and as I told you

10:09already model is nothing but some kind of a logic so we want to check before we give this particular

10:14model to our customers or before we associate this with our AI systems we want to check whether this

10:20model is performing well or not now in that case we are going to try this model with this validation

10:27data which we have kept aside initially because this validation data is already having known labels

10:32associated with that we will apply this particular data into this particular function and we will try

10:39to get the new labels and why we are doing this thing well because we want to evaluate model we

10:45want to compare the predicted labels to actual labels and if both are matched then it means my model is

10:51predicting correct things if both are not matched then it means that my model is not performing well

10:56now in real time your data scientists are actually going to do this job

11:00and they are going to check couple of metrics there is something called confusion metrics precision

11:05curve aoc curve there are so many things which they are going to check but based on that they are

11:09going to evaluate model and they are going to check whether the model is performing well or not

11:13if model is performing well we can use that model for our AI software applications but if model is not

11:20performing well you have to maybe repeat this process multiple times you're maybe going to redo the same

11:27same thing by adding more amount of data by applying different kind of algorithms in step number two and you're

11:32going to do the model evolution once again now this evaluation you're going to do until you're getting best

11:38performing model maybe this process you have to repeat four times five times or more that is something which

11:44is a task of your data scientist and that's what machine learning is actually doing ultimately at the end of this

11:51machine learning process you are going to get one good performing model and that model is what we are

11:56going to use with our AI systems i hope this concepts of machine learning is clear intentionally i'm not

12:04going very in-depth because this video is for everyone so i'm not focusing only on the technical

12:09concepts this technical concept is something which is simple to understand even if you belongs to any

12:15particular domain of the industry keeping that in mind let's move forward to the next thing which is

12:21deep learning as i say deep learning is that particular part of machine learning which is focusing

12:27on mimicking human brain capabilities so basically deep learning has created something which is known as

12:33neural network or if i be specific this is something which is known as deep neural network dnn for your kind

12:40information neural network has also gone through an evolution in the history initially there were some

12:46neural networks which were known as rnn recurrent neural network then we got something called lstm then

12:52we got something which is called cnn convolutional neural network there are different kind of neural network

12:57came up the one which we are using right now with the modern generative ai application is deep neural

13:03network because this is actually trying to create something like a human brain capability so basically they have

13:09created what they have created some kind of a network of multiple neural nodes which are nothing but

13:14some kind of a small small computers now you can think like this right now that you are going to

13:20have some kind of a model logic and that model logic which you have created with the help of machine

13:24learning model you are actually going to execute when you execute this function you're going to provide

13:30values into that and with those values we want to check whether this model is predicting proper or not

13:36that's why we are going to apply some kind of a weight associated with that this different weight

13:40we want to associate with this logical function execution and we want to execute this function

13:45maybe hundreds of times so that we can check for different values and we can actually associate

13:50that which is the right label which we are predicting now i know this sounds confusing let me just simplify

13:56this with the examples which we have seen so far so have a look at this slide in this slide if you

14:03check right now i'm showing you an example of multi-class classification which is the same example of

14:08that penguin kind of a thing and then i'm showing you how exactly deep neural network is going to

14:13process that from the left side you're going to provide some data now let's say you are providing

14:18some penguin measurement data here assume that x1 is something which is a length of the penguin's

14:24flipper x2 is something which is a depth of the penguin's flipper same way maybe x4 is something which

14:30is weight of the penguin now all these things you are providing as an input and then once you provide

14:35that input data this each blue dot which is here is nothing but a small kind of a compute node

14:40that's going to execute this functional logic on that data and it's going to apply different different

14:45kind of a weight with that throughout this process in this diagram right now maybe we have 15 to 20 nodes

14:51only but in real neural network you're going to have hundreds of nodes these hundreds of nodes are going to

14:57process this information is going to execute this thing parallelly and then it's going to pass on that

15:02to the further nodes this kind of a multi-parallel process is what we are able to do with the help of

15:07deep neural network so remember deep neural networks are helping you to do what processing the data in a

15:13much faster way in the modern ai agents like copilot and all i hope you agree with me that when you provide

15:19your prompt by typing some natural language input it's actually going to give you response within fraction of

15:25seconds suppose if you say i want to write a blog maybe a 500 word blog on a specific topic it's

15:31actually going to do that thing in few seconds only how exactly it's going to understand that how exactly

15:36it's going to process that obviously there should be a faster processing engines which are running in

15:41the background well those engines are actually nothing but this deep neural network kind of nodes now based

15:47on this processing in this case is actually able to predict that out of the three species of penguin

15:53maybe there are 70 percent chances are there that this penguin belongs to one particular species

15:58which is there maybe this is a gen 2 species of the penguin now this is how the processing is going

16:03to be working now if i connect all these dots remember guys generative ai is what we are going to use

16:09which is actually a subset of ai but if you want to use generative ai in the background machine learning

16:14has created model for that and those models are actually going to be useless if you cannot process the

16:20data with a much faster way that processing is something which is provided by deep neural network

16:26so basically the model logic of machine learning and the processing of deep learning neural network

16:31is what which is combined and giving you the response associated with that i hope you're able

16:36to understand this still in all these things if you are getting confused or if you have any questions

16:41feel free to ask me a question in the comment of this particular video i'm always there to answer that

Describe Machine Learning - AI for Everyone

Category

Transcript

Recommended