(ML 14.9) Backward algorithm

แชร์
ฝัง
  • เผยแพร่เมื่อ 7 ม.ค. 2025

ความคิดเห็น • 22

  • @wen7166
    @wen7166 5 ปีที่แล้ว +9

    i was stuck by the derivation starting 3:45. Took me a while to realize that if P(X|Z)=P(X, Z)/P(Z) is used, everthing falls into place - P(Z) will be cancelled out once chain rule is applied. Great series though.

  • @ElizaberthUndEugen
    @ElizaberthUndEugen 5 ปีที่แล้ว +2

    If anyone else tripped over 12:57:
    The assertion is that p(a, b | c) = p(a | b) p(b | c).
    Let p(a | b, c) = p(a | b) (Markov assumption)
    Then
    p(a, b, c) = p(a, b | c) p(c)
    = p(a | b, c) p(b | c) p(c)
    => (cancel p(c) on both sides)
    p(a, b | c) = p(a | b, c) p(b | c)
    But by assumption p(a | b, c) = p(a | b). Thus
    p(a, b | c) = p(a | b) p(b | c)
    q.e.d.

  • @muhibullah2134
    @muhibullah2134 5 ปีที่แล้ว +1

    mind blowing lectures on HMM

  • @siddharthsvnit
    @siddharthsvnit 7 ปีที่แล้ว

    Thanks for these videos, they save lot of time.

  • @seculardharma
    @seculardharma 9 ปีที่แล้ว +4

    I wish I had found this video a month back - HMM put me in a lot of bind over the semester

  • @amrittiwary080689
    @amrittiwary080689 2 ปีที่แล้ว

    cant find any explanation on why beta(n) = 1

  • @runzombies
    @runzombies 13 ปีที่แล้ว +1

    Great video! I'm in a computational genomics course now and feeling a bit overwhelmed - this definitely helped. Subbing now :-). I'll be coming back to this channel this semester a lot. Do you know where I could find a video of an explanation for how to implement (however simply) this algorithm? That would really seal in all of this information.

  • @lancelofjohn6995
    @lancelofjohn6995 2 ปีที่แล้ว

    why is beta(n) equal to 1?

  • @njordanjameson
    @njordanjameson 11 ปีที่แล้ว +1

    Great video! I have a quick question. When you factor \sum p(x_{k+1:n}, z_{k+1} | z_{k}), you write p(x_{k+2:n} | z_{k+1}, z_{k}, x_{k+1}) p(x_{k+1} | z_{k+1}, z_{k}) p(z_{k+1} | z_{k}). Why is there not a p(z_{k}) after this final term? According to Bayes rule, we should write P(A,B) = P(A | B) P(B), so p(z_{k+1}, z_{k}) should factor into p(z_{k+1} | z_{k}) p(z_{k}), right? Thanks!

    • @jairo1412
      @jairo1412 11 ปีที่แล้ว +3

      Hi, I think I can answer that. If you put a p(z_k) in there, you would have (according to the conditional probability formula) \sum p( x_{k+2:n} , z_{k+1}, z_{k}, x_{k+1}) , that is: \sum p(x_{k+1:n}, z_{k+1}, z_{k}) and NOT \sum p(x_{k+1:n}, z_{k+1} | z_{k}). If you multiply the equation on both sides with p(z_k), you can check that it is indeed correct ;-)
      BTW the formula used there is:
      P(A,B|C) = P(A|B,C)*P(B|C)

  • @pratyushtiwary3045
    @pratyushtiwary3045 7 ปีที่แล้ว +4

    Very useful set of lectures. I think it would be nice to clarify that beta(n) can be set to any arbitrary value. See for instance Rabiber, Proc. IEEE (1989) page 263, after Eq. 25
    www.ece.ucsb.edu/Faculty/Rabiner/ece259/Reprints/tutorial%20on%20hmm%20and%20applications.pdf

  • @sarash5909
    @sarash5909 5 ปีที่แล้ว

    Thanks! This was useful! :)

  • @edithwu8912
    @edithwu8912 5 ปีที่แล้ว

    saves my life, god bless you

  • @charliewong1598
    @charliewong1598 11 ปีที่แล้ว +4

    Thanks for your great video. I'm confusing about why Beta(n) should always be 1. Could you please give more explanation about this?

    • @p.z.8355
      @p.z.8355 8 ปีที่แล้ว +3

      Actually its arbitrary as long as its not 0 things will work

    • @arihuntl
      @arihuntl 6 ปีที่แล้ว +1

      Remember what Beta_k is--probability you'll see X_k+1:n given z_n. Since x_n+1:n is an empty list, the question then becomes what is more likely to terminate the sequence, and they are all equally likely, thus we set them all to 1.

  • @uninstead
    @uninstead 12 ปีที่แล้ว +1

    Hi greate videos. But i think you should use matrices A and B in the final formulas instead of theier probabilites meanings, then you can skip that 'known' point.

  • @jkobject
    @jkobject 5 ปีที่แล้ว

    it is sum for k from 1 to n right?

  • @chiragmakhija8142
    @chiragmakhija8142 7 ปีที่แล้ว

    amazing video!

  • @zhengheng1982
    @zhengheng1982 4 ปีที่แล้ว +1

    The notation is extremely confusing ...

  • @alviur
    @alviur 9 ปีที่แล้ว

    Thanks!!