An example of a singular value decomposition

Поделиться
HTML-код
  • Опубликовано: 1 фев 2025

Комментарии • 83

  • @alexandermread
    @alexandermread 2 года назад +17

    Just want to say thank you; sat here for hours trying to figure out some of the points you explained in a few minutes. Lifesaver

    • @eevibessite
      @eevibessite 2 года назад

      ruclips.net/video/0Ahj8SLDgig/видео.html

  • @SK-qn5ry
    @SK-qn5ry 11 месяцев назад

    till today , no one have made such good video like you SIR.
    really hatsoff , u covered all corner cases in SVD

  • @wailandaloussi9111
    @wailandaloussi9111 3 года назад +3

    thank you bro , you really helped me . I almost went crazy trying to find something that is clear

  • @twittyparker4531
    @twittyparker4531 4 года назад +8

    i actually have no idea why there are 5 dislikes, these are great SVD decomposition examples. Thank you!

    • @eevibessite
      @eevibessite 2 года назад

      ruclips.net/video/0Ahj8SLDgig/видео.html

  • @dsgarden
    @dsgarden 7 месяцев назад +3

    You should have mentioned around 1:00 that the singular values are the square roots of the non zero eigenvalues, otherwise it appears like you computed it out of thin air

    • @MatheMagician
      @MatheMagician  7 месяцев назад +4

      Hi dsgarden, thank you for your remarkt. You are right, I now added a flag to another video that explains how you can find the singular values; hope this helps!

  • @zuggrr
    @zuggrr 3 года назад +3

    Perfect, exactly what I needed. Thank you sir !

    • @eevibessite
      @eevibessite 2 года назад

      ruclips.net/video/0Ahj8SLDgig/видео.html

  • @AbdulBasitpawarmath
    @AbdulBasitpawarmath Год назад +2

    The complete setup looks like a news channel. also the accent of speaker is like a news reporter.

  • @rahul_r_007
    @rahul_r_007 2 месяца назад +1

    How did you get U3 as [0,0,1]?

  • @ShortWouldYouRather
    @ShortWouldYouRather 3 года назад +2

    Hi I have a question. In the B example, could you also determine u3 by taking the cross product of u1 and u2

    • @MatheMagician
      @MatheMagician  3 года назад +3

      Hi scarface08, thank you for your question. You can indeed take the cross product of u1 and u2 here to determine u3. I use this method here, because it is more general (and also works for bigger matrices), whereas the cross product only works in R3.

    • @ShortWouldYouRather
      @ShortWouldYouRather 3 года назад +1

      @@MatheMagician Alrigth, thanks for the fast response! Greetz uit eindhoven

  • @yassinal-nuaimee1204
    @yassinal-nuaimee1204 2 года назад +1

    how did you know that 1 was an eigenvalue in the first example? thanks in advance!

    • @MatheMagician
      @MatheMagician  2 года назад +2

      Hi Yassin, If you compute det(A^TA -lamda I) you can expand to the last column and you get (1-lambda) times something else. So this determinant is certainly zero if lambda=1. Hope this helps, good luck!

    • @yassinal-nuaimee1204
      @yassinal-nuaimee1204 2 года назад

      Thanks a lot🙏🏼

  • @mpfmax0
    @mpfmax0 Год назад

    Isn’t the first eigen vector w1 supposed to be [-1,1,0]? 1:18

    • @MatheMagician
      @MatheMagician  Год назад +1

      Hi mpfmax0, thank you for your question. We have a choice here; in this case we chose eigenvalues from big to small, i.e. lambda1=2 and lambda3=0. In this v1 corresponds to lambda=2 and equals [1,1,0] - you can check this by computating A^TA w1 and lambda1 w1. And we have w3=[-1,1,0] the eigenvector corresponding to lambda=0. So with this choice w1=[1,1,0] as in the video. If you would choose your eigenvalues from small to large, then lambda1=0 and w1=[-1,1,0], but that is not the choice we made here. Hope this helps, good luck!

    • @mpfmax0
      @mpfmax0 Год назад

      @@MatheMagicianI see, thanks for the clarification. Is there any particular reason to sort the eigenvalues in descending order rather than ascending ?

    • @MatheMagician
      @MatheMagician  Год назад

      @@mpfmax0 It is customary to have the zeros last; it makes some formulas slight easier (see Wikipedia on the singular value decomposition). For the others, it is always possible to arrange in descending order and this how it is usually done (customary).

  • @scottharman3331
    @scottharman3331 3 года назад +2

    how does a 2x3 matrix times its transpose get a 3x3

    • @Student-js4qy
      @Student-js4qy 9 месяцев назад +3

      Notice it's A^T(A) and not A(A^T) therefore, (3x2)(2x3)---->(3x3) inner product the same so outer is the result. Hope I didn't respond to late!

  • @Lute-I3s
    @Lute-I3s 9 месяцев назад

    Why did we find u3 when finding U in B? A bit confused. but didnt find a u3 when we were solving for A?

    • @MatheMagician
      @MatheMagician  9 месяцев назад

      Hi, thank you for your question. I am not sure I understand exactly what you mean, but let me try to help. In this video we are actually solving two problems. First the SVD for A, which requires us to find U. And then we determine the SVD of another matrix B, which requires us to find another matrix U. Was this the problem?

    • @alibkhash
      @alibkhash 8 месяцев назад

      @@MatheMagician I did not understand why we needed to find u3 for The Matrix B.
      I tried to test my answer with U(2x2) and it was perfectly true.

    • @MatheMagician
      @MatheMagician  8 месяцев назад

      @@alibkhash Thank you for your question. Then you might be doing some other type of decomposition. In the SVD we explicitly impose the sizes of the matrices: we want U and V to be orthogonal matrices (hence square matrices), which means that we need U 3 by 3, V 2 by 2 and Sigma 3 by 2 in order to get B=USigmaV^T. You can generalize this idea to non-square U and V, but that is not what we did in this example. Hope this helps!

  • @Rahul-q2k4f
    @Rahul-q2k4f Год назад

    With characteristic equation I got the eigan values as only 1. Then how to obtain 2,0 as eigan values from it.

    • @MatheMagician
      @MatheMagician  Год назад

      Hi, thank you for your question. We have det (1-lambda 1 0 ||| 1 1-lambda 0 ||| 0 0 1-lambda). Expanding along the last row (or column) yields (1-lambda) det(1-lambda 1 ||| 1 1-lambda). And computing the 2 by 2 determinant we have (1-lambda)((1-lambda)^2-1) for the characteristic polynomial. Setting this to zero we find either 1-lambda=0 or (1-lambda)^-1=0, so lambda =1 or 1-lambda = 1 or 1 - lambda =-1, so lambda=1 or lambda=0 or lambda =2. Hope this helps!

    • @Rahul-q2k4f
      @Rahul-q2k4f Год назад

      @@MatheMagician Thankyou. I identified my mistake.

  • @priyasivakumar3607
    @priyasivakumar3607 Год назад

    Clean Explanation Thank you sir 🙏

  • @ShahbazKhan-kh4eu
    @ShahbazKhan-kh4eu Год назад

    Thanks sir You are great!

  • @muhamedmuslibegovic
    @muhamedmuslibegovic Год назад

    How to know how many vectors will U matrix have?

    • @MatheMagician
      @MatheMagician  Год назад +3

      Hi muhamedmuslibegovic, thank you for your question. The sizes of the matrices U, V and Sigma are determined by the size of A. This decomposition is such that, if A is a m times n matrix, than Sigma is as well. Furthermore we want U and V to be orthogonal matrices, which means that they are square matrices. So if A is m times n, so is Sigma and U has to be m times m (otherwise USigma is not defined) and V^T has to be n times n and so does V. Hope this helps, good luck!

    • @muhamedmuslibegovic
      @muhamedmuslibegovic Год назад

      @@MatheMagician Thank you so much

  • @roastytoasty8815
    @roastytoasty8815 4 года назад +1

    This is exactly what I was looking for. Thank you so much!!!!

    • @eevibessite
      @eevibessite 2 года назад

      ruclips.net/video/0Ahj8SLDgig/видео.html

  • @sebastianquintero2628
    @sebastianquintero2628 13 дней назад

    Nais vidio! Sanc u beri moch. (Zorry phor bad inglis)

  • @adedotunowojori7296
    @adedotunowojori7296 2 года назад

    Pls check eigen vectors of w2; 1:19/6:14, I got at lambda 2 ; w2T [0 0 0], pls help clarify

    • @MatheMagician
      @MatheMagician  2 года назад +1

      Hi Adedotun, thank you for your question. I am not sure whether you are asking about the first or second eigenvalue so I will do both of them. For lambda=2 we get A-lambda I = (-1 1 0; 1 -1 0 ; 0 0 -1) which reduces to (1 -1 0; 0 0 1; 0 0 0) so solving (A-lambda I)x=0 yields x3=0, x2 is free and x1=x2, so x2 (1 1 0), which yields w1 if we choose x2=1. For lambda =1 we get A-lambda I = (0 1 0; 1 0 0; 0 0 0) which reduces to (1 0 0; 0 1 0; 0 0 0) and we find x1=x2=0 and x3 is free, so x3(0 0 1). If we choose x3=1 we find w2. Hope this helps!

    • @adedotunowojori7296
      @adedotunowojori7296 2 года назад

      @@MatheMagician thanks, i got it to the point....For lambda=2 and A-lambda I =[-1 1 0;1 -1 0; 0 0 -1]*******(1). i don't understand it reducing to (1 -1 0;0 0 1; 0 0 0). but from (1) i got x3=0, and x1=x2 to yield (1 1 0) eigen vectors. but my earlier question was based on lambda =1, A-lambda I = [0 1 0; 1 0 0; 0 0 0] was obtained, but i still don't understand how x3=1 while x1=0 and x2=0. the expression of eigen vector states (A-lambda *I)(x) =(0), hence x3 should be 0. pls i need explanation in that area, thanks

    • @MatheMagician
      @MatheMagician  2 года назад +1

      Hi Adedotun, for lambda=2 you need the following reduction steps: 1) add first row to second row first, then (2) interchange second and third row, then (3) first and second times -1. For the lambda=1 when you solve you have x3 is free (no pivot), first row yields 0x1+1x2+0x3=0 so x2=0 and second row yields 1x1+0x2+0x3=0, so x1=0. As eigenvector we find (0, 0, x3) where x3 can have any value. Choosing x3=1 we find (0,0,1) as eigenvector. Hope this helps.

    • @adedotunowojori7296
      @adedotunowojori7296 2 года назад

      @@MatheMagician tnks alot, pls kindly suggest other reference materials and similar examples to get good grasp of d concept...

  • @Rahalbro
    @Rahalbro Год назад

    Just what i wad searching everywhere, u3

  • @zhenhuawei7939
    @zhenhuawei7939 3 года назад +1

    It's very helpful, thanks

  • @stevenli6743
    @stevenli6743 4 года назад

    Hello sir, if I only have one none zero eigenvalue, and the other two are zero. What should my u2 be? Thank you! (Similar case with question a)

    • @eevibessite
      @eevibessite 2 года назад

      kindly see, I have solved it here:
      ruclips.net/video/0Ahj8SLDgig/видео.html

    • @大雞雞妹子-t4d
      @大雞雞妹子-t4d Год назад

      gram-schmidt

    • @priyavarshini27
      @priyavarshini27 Год назад

      What was the answer for this

    • @justmarijn9122
      @justmarijn9122 7 месяцев назад

      Sorry I am late, but you can take the vector U1 and make it orthogonal to it. So for example (1,1) becomes (1,-1).

  • @HenryLentham
    @HenryLentham Год назад

    Hi, what if the sigma is 0?

  • @MegaArti2000
    @MegaArti2000 2 года назад

    Nice simple example for fast verification!

  • @nisalbimsara4105
    @nisalbimsara4105 Год назад

    In secind question what is e_3 @4.57

    • @MatheMagician
      @MatheMagician  Год назад +1

      Hi, nisalbimsara, thank you for your question. Here the vector e_3 is the vector (0,0,1).

  • @scottharman3331
    @scottharman3331 3 года назад

    How doe A transpose A magiclaly become a 3x3??

    • @MatheMagician
      @MatheMagician  3 года назад +2

      Hi Scott, A is a 2 by 3 matrix, which means that A^T is a 3 by 2 matrix. An 3 by 2 times a 2 by 3 yields a 3 by 3 matrix. If you want to see more details about matrix multiplication, you can take a look at the playlist "matrix operations", video "the product of two matrices" . Good luck!

  • @tharunkumaran19
    @tharunkumaran19 3 года назад +1

    Thank you sooo much buddy 🤟🤟

  • @habibullahmezbah9406
    @habibullahmezbah9406 2 года назад

    How will i find e3.. I have no idea.. Please give me suggestion

    • @MatheMagician
      @MatheMagician  2 года назад +1

      Hi Habibullah Mezbah,thanks for your question. We are looking for a third vector orthogonal to u1 and u2. In order to find one we can start with any vector in R3 that is not in Span{u1,u2}. So I just picked an easy vector (e3). Alternatively you could take (in this case) the cross product of u1 and u2; that will immediately give you a third vector orthogonal to u1 and u2. Hope this helps.

    • @habibullahmezbah9406
      @habibullahmezbah9406 2 года назад

      @@MatheMagician when i do cross product u1 & u2 matrix, e3 isn’t similar 😪😪,,please send me this manual process... Just e3 matrix how you will obtain 😪😪

    • @zeferreira8805
      @zeferreira8805 2 года назад +1

      @@habibullahmezbah9406 It doesn't have to... + by doing the cross product you'll only have to divide by the norm of the obtained vector, but since the previous vectores are already normalized and orthogonalized, the vector will have already length 1 and orthogonalized too.

    • @zeferreira8805
      @zeferreira8805 2 года назад +1

      The only thing that might differ wether it's v or -v

    • @habibullahmezbah9406
      @habibullahmezbah9406 2 года назад

      @@zeferreira8805 where are you form brother?

  • @alexispadilla7848
    @alexispadilla7848 11 месяцев назад

    Thank you so much, sir.

  • @ammar_maamoun
    @ammar_maamoun 8 месяцев назад +2

    U had all the time to went to calculation but yet you are just assuming the one who will watch this is a professor and wouldn't mind high level small talk with his pal

  • @adithkirankumar1263
    @adithkirankumar1263 8 месяцев назад

    how did you get e3 ?

    • @MatheMagician
      @MatheMagician  8 месяцев назад

      Hi adithkiranmumar, thanks for your question. We are looking for a third vector orthogonal to u1 and u2. In order to find one we can start with any vector in R3 that is not in Span{u1,u2}. So I just picked an easy vector (e3). Hope this helps.

  • @ninamamou7299
    @ninamamou7299 3 года назад +1

    thank you so so much!!!!

  • @sharmabu
    @sharmabu 5 месяцев назад

    Thanks man!

  • @johnnysasquatch3003
    @johnnysasquatch3003 3 года назад +1

    Thanks a lot!!

  • @Recordingization
    @Recordingization 2 года назад

    Nice lecture of SVD.

  • @yxoxz
    @yxoxz Год назад

    woox?

  • @leonardobosco3916
    @leonardobosco3916 4 года назад

    sorry how did you choose e3?

    • @MatheMagician
      @MatheMagician  4 года назад +4

      Hi Leonardo, thanks for your question. We are looking for a third vector orthogonal to u1 and u2. In order to find one we can start with any vector in R3 that is not in Span{u1,u2}. So I just picked an easy vector (e3). Hope this helps.

    • @leonardobosco3916
      @leonardobosco3916 4 года назад

      @@MatheMagician ah alright. Thank you very much, this video helped me a lot!!

  • @_alekss
    @_alekss 2 года назад +2

    I got here looking for real life examples and where to use it in physical world, and didn't get any further

  • @ryanzhu1656
    @ryanzhu1656 5 месяцев назад

    Are you really speaking English in this video? 😂

  • @ahmetsemihkarakas1597
    @ahmetsemihkarakas1597 10 месяцев назад

    Thanks a lot!