Simple and beautiful lecture...I got reminded of the quote "If you can't explain something in a simple way then you don't understand it well enough"...Kudos to the professor for his deep understanding and simple explanation....
Man you are having way to much fun here... you treat this more like a game show than a lecture. On the other hand... camera is not keeping up... a wide shot beats a close up at any lecture. Sometimes your voice drops to a mumble... auto level in post will easily fix volume swings. I don't care if I have to watch it a couple of times to get it down, it's a fun thing ti watch. SVD... a rotate a stretch and a rotate... simple simple simple... I studied linear algebra via MIT 18.06 Gilbert Strang TH-cam open courseware... a much different perspective. He is much more theoretical.... that's a good thing for an intro. You are much more playful as you show us how to use these tools we worked so hard to get. I'm an old man, I do this for fun... thank you for making it entertaining.
I have an hard time to understand why we are trying to maximize variance(theta) explicitly to obtain U in step 1? Why we don't just compute the PCA of X^TX? To obtain both U and Sigma?
Amazing lecture. I love all lectures from Kutz. Tiny suggestion. Var(θ) looks like variance of θ, rather than "variance as a function of θ". By the way. I think θ is just an eigen vector of the var-covariance matrix of 𝑰 (i.e. the final pixel measurements in two directions). So we don't actually need to go through the derivative steps to find θ, right?
Yo Kutz, I like your video, thanks. Question on how your Var(theta) function is derived. I know you assume mean is zero, which gets it close to looking like normal variance, but I just do not see any identities or properties which get this function you're calling out.
Say you have a point P with coordinates (a, b). Draw a line D with angle θ (w.r.t x axis). Then a * cosθ + b * sinθ is the distance from origin to the projected point (of P) onto line D. Do it by your hand and it is easy to find out.
@Nathan Kutz, Sir the Lecture was very useful in getting an insight on the ICA. I am currently working on Noise separation Project, it would be really helpful if you can give me some idea on how to implement the iCA for my Project. Kindly request you to share your email ID please.
Nathan Kutz for president! Very nice and informative lecture.
Simple and beautiful lecture...I got reminded of the quote "If you can't explain something in a simple way then you don't understand it well enough"...Kudos to the professor for his deep understanding and simple explanation....
Never have i ever seen someone manage to make such dry things fun! You have an awesome personality!
Man you are having way to much fun here... you treat this more like a game show than a lecture.
On the other hand... camera is not keeping up... a wide shot beats a close up at any lecture.
Sometimes your voice drops to a mumble... auto level in post will easily fix volume swings.
I don't care if I have to watch it a couple of times to get it down, it's a fun thing ti watch.
SVD... a rotate a stretch and a rotate... simple simple simple... I studied linear algebra via MIT 18.06 Gilbert Strang TH-cam open courseware... a much different perspective. He is much more theoretical.... that's a good thing for an intro.
You are much more playful as you show us how to use these tools we worked so hard to get.
I'm an old man, I do this for fun... thank you for making it entertaining.
Thank you for this very clear and fun Lecture!
JKU Students leaving traces everywhere cause they are provided with crap lectures
@@d.m.2343 I also hate JKU
45:31 Can anyone explain why minimize the 4th moment (kurtosis) rather than the 2nd moment (variance)?
I have an hard time to understand why we are trying to maximize variance(theta) explicitly to obtain U in step 1? Why we don't just compute the PCA of X^TX? To obtain both U and Sigma?
Amazing lecture. I love all lectures from Kutz. Tiny suggestion. Var(θ) looks like variance of θ, rather than "variance as a function of θ". By the way. I think θ is just an eigen vector of the var-covariance matrix of 𝑰 (i.e. the final pixel measurements in two directions). So we don't actually need to go through the derivative steps to find θ, right?
14:00 kurtosis of gaussian is 3, no? the excess kurtosis is 0.
Well, there you go, you answered it yourself.
Oh, how do I know the sequence of your lectures... PCA is labeled 1, 2 and 3... where do I go for the sequence of all lectures?
I'm not a mathematician, why the kurtosis equal to zero make the variables independent?
Yo Kutz, I like your video, thanks. Question on how your Var(theta) function is derived. I know you assume mean is zero, which gets it close to looking like normal variance, but I just do not see any identities or properties which get this function you're calling out.
Say you have a point P with coordinates (a, b). Draw a line D with angle θ (w.r.t x axis). Then a * cosθ + b * sinθ is the distance from origin to the projected point (of P) onto line D. Do it by your hand and it is easy to find out.
Can anyone provide a derivation of optimizing kurtosis derivation ?
i cant listen, very low sound
@Nathan Kutz, Sir the Lecture was very useful in getting an insight on the ICA. I am currently working on Noise separation Project, it would be really helpful if you can give me some idea on how to implement the iCA for my Project. Kindly request you to share your email ID please.