All other TH-cam videos just explain the basics, but no one explains the maths behind the SVM. But you explained the whole mathematics of SVM. I was stuck on understanding the logic behind the SVM, and even I read many books on ML, but I had not get brief explanation. But your video helps me a lot.
It's commendable how painstakingly you explain every single detail! Wish we had more professors like you. Thanks a ton for these videos, you're a life saviour..
04:22 The main goal of SVM is to find the equation of a hyperplane with the maximum margin between positive and negative hyperplanes. 08:44 In machine learning, a decision rule can be derived using the equation w * u + b > 0. 13:06 The equation of the hyperplanes is assumed to be W transpose x + b = 1 for the positive hyperplane and W transpose x + b = -1 for the negative hyperplane. 17:28 The distance calculation involves simplifying equations to x squared and creating positive and negative lines for convenience. 21:50 Multiplying by a factor greater than 1 shrinks the margin, while multiplying by a factor less than 1 expands the margin. 26:12 Maximize distance between positive and negative hyperplanes with constraints 30:34 The distance between two support vectors in SVM can be calculated using the dot product of the vectors and the unit vector 34:54 The expression for distance calculation and the optimization function in SVM Crafted by Merlin AI.
i am a PhD scholar at NIT in Mathematics this video is for actual working of SVM rather than those 10 min video whose aim is just views but the viewer doesn't know.
Sir I really like your in-depth explanations. It is actually needed rather than a 10 min short overview where we get no depth on the model. Thank you for making these in-depth mathematical videos.
The main goal of SVM is to find the equation of a hyperplane with the maximum margin between positive and negative hyperplanes. 04:22 In machine learning, a decision rule can be derived using the equation w * u + b > 0. 08:44 The equation of the hyperplanes is assumed to be W transpose x + b = 1 for the positive hyperplane and W transpose x + b = -1 for the negative hyperplane. 13:06 The distance calculation involves simplifying equations to x squared and creating positive and negative lines for convenience. 17:28 Multiplying by a factor greater than 1 shrinks the margin, while multiplying by a factor less than 1 expands the margin. 21:50 Maximize distance between positive and negative hyperplanes with constraints 26:12 The distance between two support vectors in SVM can be calculated using the dot product of the vectors and the unit vector 30:34 The expression for distance calculation and the optimization function in SVM Click to expand 34:54
Don't have words to thank you sir for this more than amazing explanation ! Tried multiple sources but it became more of rote learning but your video actually made me understand how SVM works ! Thanks a Ton Sir ! :)
Just watched the same lecture from MIT and you covered every aspect of what they taught in way more better way Good to know Indians are competing nation wide
Ammmazing guy i just wnated to know about maths of svm to write in my sem exam( b.e) and now i know much more than just that please keep doing tye great work
Thanks for giving such a good explaination on this topic on I was struggling to get clearity.It really helped me to get good picture of mathematical concept behind svm.
Bro your video is extremely good kudos to you. Just improve the sound quality as every thing is hazy while listening that is manageable. But overall awesome❤❤❤
To determine the distance of that point using that perpendicular line ... And the later whole video is based on that ... What happens if that distance is greater and what happens if it's less... If I'm wrong,feel free to correct me 🎉
I have a doubt, pi+ and pi- are basically hyperplanes passing through the closet points from the hyperplane wx + b = 0(let's say hyperplane p), what if the distance between pi+ and p is not the same as pi- and p, there can be a case in which the distance between the nearest point of one class and the hyperplane p and the distance between the nearest point of the other class and hyperplane p may not be equal.....what to do in this kind of a case? And how do we optimize the distance function?
Hi sir, As we were trying to achieve hypothesis line in Logistic regression such that it achieves equilibrium among dataPoints. And in SVM also we are doing same thing . Then what is the difference/ how it is solving logistic regression's problem/ or what is the problem in Logistic regression when we have achieved Equilibrium hypothesis line ? @CampusX
The optimization problem to obtain the plane is different in each . In svm the loss function has two terms: margin and misclassification components. In logitreg, it's the binary cross entropy. So the method in which we obtain the hyper plane is different
Great video! One question, at 5:00 you made the decision rule vector(w).vector(u) + b >= 0. How did you put 'b' here which is a characteristic of the hyperplane? Thanks for clarifying in advance.
Bismillah Obviously, you need to know the intercept of the line. Only knowing the vector w will not help because without b, you don't know where the line is clearly situated. With vector w, you know the direction of the line but not it's exact postition. The line could have a gradient of 3, for e.g., but if you don't know where the line cuts the y axis, how can you make your decision?
you very casually said that projection is dot product, can you please provide any source where its written that dot product is actually projection of one vector over another
perhaps there was a slight mistake it should be w.u - c = 0.. such that when any point is greater than 0 than it must lie in positive region else negative region... b and c are just contant for positive n negative hyperplane respectively.
pls anyone answer these question 1) the planes pie(+) and pie(-) are decided by the support vectors then how come their equation can be guaranteed to be wx+b=1 and wx+b=-1 2)if the equation are wx+b=1 and wx+b=-1 , the margin(d) we aiming to maximize is basically the distance bw these two planes only, and if we see the distance bw these two planes is 2 which is a constant then how come we can maximize a constant
2.) you're confused about the chronological order mate, FIRST we find the hyperplane ( pie neutral) and THEN we find ( pie + and pie -) AND THEN calculate the distance between the parallels ( pie + and pie -)
1.) the hyperplane SHOULD be equidistant from pie + and pie - , so the magnitude SHOULD be opposite to each other....and about why +1 and -1 and why not our custom favourite numbers like +69 and -69 ?.. It's because essentially we're concerned about the relationship between the numbers and not the numbers themselves because they're arbitrary ( the numbers) .. this is the fundamental essence of feature engineering... For 100 and 1000 , on log scale are 2 and 3 .. 100 and 1000 , and 2 and 3 they are arbitrary, but essentially represent the same thing (in its magnitude essence)
This video made me emotional ....I couldn't understand my professor's slides at all but now its damn easy. This channel is underrated for sure.
All other TH-cam videos just explain the basics, but no one explains the maths behind the SVM. But you explained the whole mathematics of SVM. I was stuck on understanding the logic behind the SVM, and even I read many books on ML, but I had not get brief explanation. But your video helps me a lot.
Raat me video bnake dark circle bn gye hain, but really thanks for this content. well explained.
It's commendable how painstakingly you explain every single detail! Wish we had more professors like you.
Thanks a ton for these videos, you're a life saviour..
04:22 The main goal of SVM is to find the equation of a hyperplane with the maximum margin between positive and negative hyperplanes.
08:44 In machine learning, a decision rule can be derived using the equation w * u + b > 0.
13:06 The equation of the hyperplanes is assumed to be W transpose x + b = 1 for the positive hyperplane and W transpose x + b = -1 for the negative hyperplane.
17:28 The distance calculation involves simplifying equations to x squared and creating positive and negative lines for convenience.
21:50 Multiplying by a factor greater than 1 shrinks the margin, while multiplying by a factor less than 1 expands the margin.
26:12 Maximize distance between positive and negative hyperplanes with constraints
30:34 The distance between two support vectors in SVM can be calculated using the dot product of the vectors and the unit vector
34:54 The expression for distance calculation and the optimization function in SVM
Crafted by Merlin AI.
how did you do this by using AI?
What an explanation! Beyond words. Thank you for knowledge spreading
Your explanations are simply best .. if u understand with patience point by point ....u can answer questions on them even in sleep ....thnkx bro
i am a PhD scholar at NIT in Mathematics this video is for actual working of SVM rather than those 10 min video whose aim is just views but the viewer doesn't know.
I am pursuing eco Hons now first year and it is easily understandable what he is trying to teach , most underrated channel
Sir I really like your in-depth explanations. It is actually needed rather than a 10 min short overview where we get no depth on the model. Thank you for making these in-depth mathematical videos.
Beautifully explained. Stumbled upon this video after a long search. You are better than many of the top ML teachers on YT. Thank you very much :).
Best explanation I ever seen for such complex topic. Thankyou sir👏
The main goal of SVM is to find the equation of a hyperplane with the maximum margin between positive and negative hyperplanes.
04:22
In machine learning, a decision rule can be derived using the equation w * u + b > 0.
08:44
The equation of the hyperplanes is assumed to be W transpose x + b = 1 for the positive hyperplane and W transpose x + b = -1 for the negative hyperplane.
13:06
The distance calculation involves simplifying equations to x squared and creating positive and negative lines for convenience.
17:28
Multiplying by a factor greater than 1 shrinks the margin, while multiplying by a factor less than 1 expands the margin.
21:50
Maximize distance between positive and negative hyperplanes with constraints
26:12
The distance between two support vectors in SVM can be calculated using the dot product of the vectors and the unit vector
30:34
The expression for distance calculation and the optimization function in SVM
Click to expand
34:54
Don't have words to thank you sir for this more than amazing explanation ! Tried multiple sources but it became more of rote learning but your video actually made me understand how SVM works ! Thanks a Ton Sir ! :)
India need teachers like you.
world*
That explanation with the help of Graph tool is just fabulous.... beginners can easily grasp....Thank you so much 👏
Just watched the same lecture from MIT and you covered every aspect of what they taught in way more better way
Good to know Indians are competing nation wide
I am having very good learning experience from you
Thanks for your efforts.
i can bet, you can't have a better playlist than this🤲
This guy is damn good
What , an , explaination ... What an explaination !!! Hat's off boss🙌
This is the best explanation of SVM....Keep doing good work
Ammmazing guy i just wnated to know about maths of svm to write in my sem exam( b.e) and now i know much more than just that please keep doing tye great work
Oh Wow Wow wow I Found This Amazing Channel 🔥 🔥
Hey thanks for saving my family from that fire, and for teaching me SVM's too.
What a explained bro God bless you
Thanks for giving such a good explaination on this topic on I was struggling to get clearity.It really helped me to get good picture of mathematical concept behind svm.
Best explanation of SVM I have ever seen till now...Amazing man...keep working
Loved the explanation! Thank you so much
Loved your Explanation i am Mtech cse from NIT
Thank you mere ❤️ bhai 😘😘😘😘😘😘
Sach kahu kai videos me muh maar lis lekin samjh yahi aya i love you
great explanation Sir
Same as mit video but the intuition made it clearer. Thanks for this
data science aspirant needs a teacher like you
Best explanation of SVM I ever seen ❤🙌 #NitishSir
Next Level , Itne detail meein Kaun batata haiii... you are
while(1) GEM :)
It was really very helpful lesson. Thanks for your support.
superr bro ......Amazing explanation..
Nice explanation keep going bro...
hats off man. this video is amazing
It's amazing one 🤩👏... you are requested to recreate one dedicated video for SVM in which you will explain SVM theory as well as one small example
Bro your video is extremely good kudos to you. Just improve the sound quality as every thing is hazy while listening that is manageable. But overall awesome❤❤❤
Coming here after MIT openware SVM video and bro let me tell you one thing, you are amazinnnggg
take a bow brother.....brilliant explanation
hello sir why we taking projection of one vector on another??
and how we get that equation 03:54
To determine the distance of that point using that perpendicular line ... And the later whole video is based on that ... What happens if that distance is greater and what happens if it's less...
If I'm wrong,feel free to correct me 🎉
Simply explained. ❤
Great explaination
fantastic math u explicated.
You are theeeee best 😊😊❤
Maja aa gya
Thanks for such nice efforts
Thanks a lot for this easy explanation :)
lovely explanation. Thanks alot.
Hey isnt the projection formula of w on u is dot prod of w and v divided by magnitude of u?
👍👍👍👍 Best
w bar is what? u bar is what? Cn you please tell? I understood c as the linear distance from origin to the hyperplane.
this playlist saved my life
maje aa gaye yaar
You are a legend bro🎉🎉
I have a doubt, pi+ and pi- are basically hyperplanes passing through the closet points from the hyperplane wx + b = 0(let's say hyperplane p), what if the distance between pi+ and p is not the same as pi- and p, there can be a case in which the distance between the nearest point of one class and the hyperplane p and the distance between the nearest point of the other class and hyperplane p may not be equal.....what to do in this kind of a case? And how do we optimize the distance function?
No, the distances between the hyperplane and margin planes are always equal as the hyperplane is in a way the middle plane of both margin planes
Best video for svm
please tell me how the perpendicular w and the coffiecient line vector w are same ??? how
thanks sir 🙏
best best best!!!!
Thank you so much sir 🙏🙏🙏
great
GOATED Explanation
amazing video!
d distance nikalne ke liye W ko unit vector mein kyun convert kiya?
finished watching
how that -c changed to +b at 4:40 ? I really didn't got it. Can anybody help?
Just suppose b = -c, because wx+b = 0 looks more aesthetically pleasing than wx-c =0 ( I oversimplified it yes ...but it's the gist)
for x1, the equation y(wx+b) should be -1, you have kept it +1, am I missing something ?
But how -c was replaced by +b in the equation??? at time 5:18 mins.. please someone explain?
I am also confused there
Thank You Sir.
thanks it was amazing
Hi sir, As we were trying to achieve hypothesis line in Logistic regression such that it achieves equilibrium among dataPoints. And in SVM also we are doing same thing . Then what is the difference/ how it is solving logistic regression's problem/ or what is the problem in Logistic regression when we have achieved Equilibrium hypothesis line ?
@CampusX
The optimization problem to obtain the plane is different in each
. In svm the loss function has two terms: margin and misclassification components. In logitreg, it's the binary cross entropy. So the method in which we obtain the hyper plane is different
😍😍😍👍👍👍
Sir can you please make another svm lecture including mathematics of primal and dual formulation and why dual is important
godly loved it
Great video! One question, at 5:00 you made the decision rule vector(w).vector(u) + b >= 0. How did you put 'b' here which is a characteristic of the hyperplane? Thanks for clarifying in advance.
Bismillah
Obviously, you need to know the intercept of the line. Only knowing the vector w will not help because without b, you don't know where the line is clearly situated. With vector w, you know the direction of the line but not it's exact postition. The line could have a gradient of 3, for e.g., but if you don't know where the line cuts the y axis, how can you make your decision?
did you find the answer?
@@nikipatel7602 its similiar to theta0 in linear regression ...exactly what the justcook commented above
👏👏👏👏👏👏
Feel like clicking the like button million times🙏
Samaj gaya. Ye video pehele kyu nahi dekhi maine
why we consider y=wTx+c over y=mx+c
you very casually said that projection is dot product, can you please provide any source where its written that dot product is actually projection of one vector over another
when the hyperplane is passing through the origin then only w vector become perpendicular to the decision boundary
but what your teaching???
💕❤️
when replace c with b why to change sign ( - ) to ( + ) ? :(
same question
Please let me know the answer
mjhy khud nhi filhal abi tk pta .
he stored -c in b
But why
@4:42 You have replaced -C with +b but did not explain the reason. Pls explain it.
perhaps there was a slight mistake it should be
w.u - c = 0.. such that when any point is greater than 0 than it must lie in positive region else negative region... b and c are just contant for positive n negative hyperplane respectively.
♥
Thanks a lot SIR
pls anyone answer these question
1) the planes pie(+) and pie(-) are decided by the support vectors then how come their equation can be guaranteed to be wx+b=1 and wx+b=-1
2)if the equation are wx+b=1 and wx+b=-1 , the margin(d) we aiming to maximize is basically the distance bw these two planes only, and if we see the distance bw these two planes is 2 which is a constant then how come we can maximize a constant
2.) you're confused about the chronological order mate, FIRST we find the hyperplane ( pie neutral) and THEN we find ( pie + and pie -) AND THEN calculate the distance between the parallels ( pie + and pie -)
1.) the hyperplane SHOULD be equidistant from pie + and pie - , so the magnitude SHOULD be opposite to each other....and about why +1 and -1 and why not our custom favourite numbers like +69 and -69 ?..
It's because essentially we're concerned about the relationship between the numbers and not the numbers themselves because they're arbitrary ( the numbers) .. this is the fundamental essence of feature engineering...
For 100 and 1000 , on log scale are 2 and 3 .. 100 and 1000 , and 2 and 3 they are arbitrary, but essentially represent the same thing (in its magnitude essence)
Girls :- Let me do makeup properly i have to take online class
Le boys:- .............🤣
btw very good lecture sir
Thank you sir
How can we repalce minus c by positive b
Same Q.
Thank you
detailed intuition & derivation
How +b come in place of -c ?
its just a notational change to align with the standard SVM formulation
But c is distance from the origin to the π. where the +b is the interceptor of the line
You got a sub :)
sir your voice is very low at starting of this video , remaining was very helpful
Hazaron krish naik bhi aa jain to ap k 2 seconds main convery kiye gaye content ko bhi defeat nhi kar sktay XD
The so-called real math exaplanation.