This is the best video out there to explain Positive Semidefinite Matrices! I finally understood what Positive Definiteness means visually after struggling with the topic for a long time.
My God, you've done an amazing job! I Hope you'll continue to grow, we Need more channels like this to Give credit to this beautiful topics with so smooth animations!
Quadratic functions f(x) = x'Qx, (x is a vector and Q is a matrix), you can consider a symmetric matrix, because: x'Qx is a scalar, so x'Qx = (x'Qx)^T (transpose) x'Qx = x'Qx /2 + (x'Qx)^T / 2 = x'(Q+Q^T)x A=Q+Q^T A is symmetric because A=A^T, so f(x) = x'Ax, where A is symmetric independently of Q
I ran across positive semidefinite matrices while studying machine learning, but never had a good intuition about them. Thank you for providing some conceptual insight into them.
Awesome. I only encountered this concept from the error message when I am trying to generate some multivariate Gaussian random number with a specified covariance matrix. Nice to see the formal definition!
thank you so so so much; ive been using the definition of positive definite for so many courses but I never understood why/how it is used i just merely did the calculation it's almost like a eureka for me thank you!!!!
At 5:30 you say if one or more of the diagonal entries is zero then it is only positive semi-definite, but don't the other non-zero diagonal entries make the overall dot product of x and Ax greater than zero. Then it meets the criteria of positive definite right? What am i missing here? Thanks in advance.
If you have a matrix A that has a Zero as its i-th diagonal element, the multiplication v^T Av with v as the i-th canonical basis vector turns out to be zero ( it gives a exactly the ith element of the diagonal). So a matrix with a zero on its diagonal can at most be positive semidefinit. Analogously you can argue that a matrix cannot be psd when it has a negative number on its diagonal. Hope that helps
@@tobiherberts3128 thank you for responding to my question! It makes sense now. In the definition i saw 'for all x not equal to zero' so did not think of the canonical vector.
Thank you! It’s hard to point you in the right direction if you don’t explain how you found these eigenvalues, but Here is a sanity check you can do: the sum of the eigenvalues should be equal to the trace (i.e., the sum of the diagonal elements), and the product of the eigenvalues should be equal to the determinant.
Thank you for making this, finally I get a concrete, visual intuition of what positive definiteness means!
You are very welcome. Thanks for the nice comment!
The animations are so beautiful! And the geometric intuition was new to me, I like how it also gives a picture for positive linear functionals.
Thank you so much for the very nice comment, stay tuned for more!
How and why is this channel so underrated. The quality of work is one of the finest.
This is the best video out there to explain Positive Semidefinite Matrices! I finally understood what Positive Definiteness means visually after struggling with the topic for a long time.
Finally I got a geometric intuition of what positive definiteness means. Very well explained! Thank you so much.
Lol you literally uploaded this video the day before my convex optimization exam. Thanks!
Awesome!!!! Good luck for your exam!
Amazing video! After years of studying at uni I finally understood the real intuition behind Positive definite matrices. Thank you so much! :D
😅😮😢🎉
This is a BEAUTIFUL video. I am floored by your animations and how you communicated the relevance of PSD matrices.
What a great effect your videos have on the course of understanding of many people,
Boom! What a great way to start out -with a single element matrix were the element is a positive number. Fantastic!
My God, you've done an amazing job! I Hope you'll continue to grow, we Need more channels like this to Give credit to this beautiful topics with so smooth animations!
Thank you so much 😀 I checked your channel and you are doing such fine work yourself!
Thanks a lot for uploading this. I had been searching for a visual explanation on the topic for a long time. Also, really well explained.
Thank you very much. :)
Such a great explanation! I have never read a more clear definition of positive definite matrix.
Quadratic functions f(x) = x'Qx, (x is a vector and Q is a matrix), you can consider a symmetric matrix, because:
x'Qx is a scalar, so x'Qx = (x'Qx)^T (transpose)
x'Qx = x'Qx /2 + (x'Qx)^T / 2 = x'(Q+Q^T)x
A=Q+Q^T
A is symmetric because A=A^T, so f(x) = x'Ax, where A is symmetric independently of Q
Please keep making videos. So helpful and intuitive!
Thank you! Will do!
Such beautiful and clear explanations. You are a legend in the making!
Thank you Ario!!
I ran across positive semidefinite matrices while studying machine learning, but never had a good intuition about them. Thank you for providing some conceptual insight into them.
This is an amazing series, and I look forward to sharing this with my students!
Please do!
This is a really great video-love the vibe from the music and timing at the end 😂
man you are a legend. you have got a new subscriber
Excellent video. Clear, concise, and illuminating.
Much appreciated!!!
wow, best explanation on positive definite that I can find. Thank you so much
You're very welcome!
Wow this is an amazing video and a great channel!
Thank you so much!!
Explanation and animations are really great. keep up the good work.
what a wonderful explanation! keep going!
Awesome. I only encountered this concept from the error message when I am trying to generate some multivariate Gaussian random number with a specified covariance matrix. Nice to see the formal definition!
Just WOW! What a great explanation.Thank you!
I cannot say anything more than amazing video! Thank you for your huge contribution!
Thanks! Glad this was helpful!
Amazing amazing videos, cannot say how much I appreciate this!
Thank you!
huge fan!
Hooray! :)
thank you so so so much; ive been using the definition of positive definite for so many courses but I never understood why/how it is used i just merely did the calculation it's almost like a eureka for me thank you!!!!
Great to hear!!
This is a fantastic video
This is a fantastic comment!
Nicely explained. Thank you
totally love your work dude!
Brilliant explanation!
This is amazing. I would love to see some "visual" explanation of SDP relaxations a la Lassere.
Absolutely wonderful.
this is amazing. thank you.
I am sleepy and tired but this was still an awesome watch
Tremendously cool explanation, thank you very much!
You're very welcome!
WOW! THAT WAS AWESOME!
Nice video!
Why did you place the z>=0 plane at such an angle in 8:00? I would have assumed it to be paralell to the xy-plane.
Great video, thank you!
Awesome video!
At 5:30 you say if one or more of the diagonal entries is zero then it is only positive semi-definite, but don't the other non-zero diagonal entries make the overall dot product of x and Ax greater than zero. Then it meets the criteria of positive definite right? What am i missing here? Thanks in advance.
If you have a matrix A that has a Zero as its i-th diagonal element, the multiplication v^T Av with v as the i-th canonical basis vector turns out to be zero ( it gives a exactly the ith element of the diagonal). So a matrix with a zero on its diagonal can at most be positive semidefinit.
Analogously you can argue that a matrix cannot be psd when it has a negative number on its diagonal.
Hope that helps
@@tobiherberts3128 thank you for responding to my question! It makes sense now. In the definition i saw 'for all x not equal to zero' so did not think of the canonical vector.
So good, thank you
awesome video
What's the visual meaning of x^T*A*x?
Excellent video illustration.
But can someone tell me how at 8:23 he gets eigenvalues 1+x and 1-x?? I got them as 1-/+ sqrt(2*x)
Pls help!
Thank you!
It’s hard to point you in the right direction if you don’t explain how you found these eigenvalues, but Here is a sanity check you can do: the sum of the eigenvalues should be equal to the trace (i.e., the sum of the diagonal elements), and the product of the eigenvalues should be equal to the determinant.
Great Video! Thank you so much :))))))))))))))))))
Can you make also on backprop ? But please do with matricess
Wow, a light bulb for me turned on with this…thanks very much!
Great to hear!!!
nice animations!
is there a name for "good solution to NP-hard problem" problems?
CSC420 gang
Amazing
Wa 7miiiiiim9
Hadchi zine
Maymknch
You speak to fast
Amazing video. Thanks!