Thank you for the lecture Can anyone explain the reason behind the other factors apart from the loss function in the optimization equation? why do we need to reduce the distance of the line or the hyperplane center from the axis central point? -> ||w||^2 and what is the 'subject to' condition? how did it come by or what is its purpose? why should the points be parallel to the SVM line ( assuming dot product) correct me on this if it is wrong
We have the main line as W.X +b =0 and for the margins we will have the lines as W.X +b = +- k. On normalizing by k we will have W.X + b = +-1. So when y = +1 for green dots and -1 for the magenta dots and to have all points in the corresponding direction we will have the constraint y(W.X+b) >=1 To give some lenience and not have a hard bound we will have W.X +b = 1 - ζ where ζ is Zeta. if ζ = b then W.X = 1. Which makes the constraint y(W.X) >= 1 i.e. the minimum of W.X = 1 since y = 1 Correct me if I am wrong
You are one of the greatest teacher of mankind
What does the constraint imply geometrically in the linear SVM optimization problem?
These videos are treasure.
Thank you for the lecture
Can anyone explain the reason behind the other factors apart from the loss function in the optimization equation?
why do we need to reduce the distance of the line or the hyperplane center from the axis central point? -> ||w||^2
and what is the 'subject to' condition? how did it come by or what is its purpose? why should the points be parallel to the SVM line ( assuming dot product) correct me on this if it is wrong
||w||^2 actually is the inverse of distance(magnitude), which is not mentioned in the lecture. You can find other explanation from internet.
Does anyone can explain the subject to min|xj.w|=1, why it looks like in this formula? 16:03
We have the main line as W.X +b =0 and for the margins we will have the lines as W.X +b = +- k. On normalizing by k we will have W.X + b = +-1. So when y = +1 for green dots and -1 for the magenta dots and to have all points in the corresponding direction we will have the constraint y(W.X+b) >=1
To give some lenience and not have a hard bound we will have W.X +b = 1 - ζ where ζ is Zeta. if ζ = b then W.X = 1. Which makes the constraint y(W.X) >= 1 i.e. the minimum of W.X = 1 since y = 1
Correct me if I am wrong