How I Understand Flow Matching

Поділитися
Вставка
  • Опубліковано 21 лис 2024

КОМЕНТАРІ • 65

  • @GapLoser42
    @GapLoser42 5 днів тому +1

    Thank you for your excellent video! This is the clearest video I have ever watched explaining Flow Matching in such an interesting way!

    • @jbhuang0604
      @jbhuang0604  4 дні тому

      Thank you so much for your kind words!

  • @plcrodrigues
    @plcrodrigues 5 місяців тому +11

    This is gold. Congrats for doing such a fun, pedagogical, and informative on a topic that can often be quite dry in the literature.

    • @jbhuang0604
      @jbhuang0604  5 місяців тому

      Thanks for your kind words!

  • @manuellecha
    @manuellecha 5 місяців тому +4

    Thank you very much! You did a really nice job! The video is clear, visual, and informative. It is consistent with the timeline and evolution of the field, and it effectively conveys the information along with the motivation for the development of these models.

    • @jbhuang0604
      @jbhuang0604  5 місяців тому

      Glad you liked it! It’s a lot of fun making this video!

  • @pravin1390
    @pravin1390 3 місяці тому +3

    Thanks for the explanation. I feel it makes sense to use the notation p(x_t, t). Then, it is clear to prove ∂/∂t p(x_t, t) = -div(p(x_t, t) u(x_t, t)) is equivalent to
    d/dt p(x_t, t) = -p(x_t, t) div(u(x_t,t)), which is indeed required here. Please let me know if my comments make sense. The first equation is partial derivative and second equation is total derivative is what I sense.

    • @NirGoren-k2k
      @NirGoren-k2k 18 днів тому

      Thank you, was struggling to make sense of this part.

  • @AnujZore-pe9mg
    @AnujZore-pe9mg 5 місяців тому +4

    Thanks once again for easy-to-understand explanation! Gonna miss CSMC 733 lectures :(

  • @catherineyang5199
    @catherineyang5199 5 місяців тому +1

    Thank you for the video! This is the most clear explanation of flow matching on the internet ❤

    • @jbhuang0604
      @jbhuang0604  5 місяців тому

      Thank you so much for your kind words!

  • @因幡の黒うさぎ-i1p
    @因幡の黒うさぎ-i1p Місяць тому

    Your lecture was truly inspiring and made complex concepts so easy to understand. Thank you for your incredible clarity and passion - I’m deeply grateful!

    • @jbhuang0604
      @jbhuang0604  Місяць тому

      You're very welcome! Glad that you like it!

  • @nathan_ca
    @nathan_ca 5 місяців тому +1

    Thanks! This is amazing video to get students, like me, to re-engage with these topics that i haven't had a chance to explore more ❤

    • @jbhuang0604
      @jbhuang0604  5 місяців тому

      Thanks! Glad that this is helpful.

  • @DimitrivonRutte
    @DimitrivonRutte 5 місяців тому +3

    Awesome to see easy-to-understand explanations of current research topics, keep up the great work!

  • @r00t257
    @r00t257 5 місяців тому +1

    Legend comeback 🙇! Your educational video is worth more than gold.💓🙏

    • @jbhuang0604
      @jbhuang0604  5 місяців тому

      Thanks a lot! Glad you like it!

  • @kimchi_taco
    @kimchi_taco 12 днів тому +1

    This is gold mine! Thank you!

  • @JQXU-z3s
    @JQXU-z3s 4 місяці тому +1

    Thank you for your excellent work! Absolutely clear and informative.

  • @amirhosseinraffiee8270
    @amirhosseinraffiee8270 5 місяців тому +1

    Thanks for the video. Great way to explain a complex concept

    • @jbhuang0604
      @jbhuang0604  5 місяців тому +1

      Appreciate your comment! Thanks for watching the video. Hope you enjoyed it.

  • @sayakpaul3152
    @sayakpaul3152 3 місяці тому +1

    Do you have the next video already? It's so good!

    • @jbhuang0604
      @jbhuang0604  3 місяці тому

      Thanks! I am working on it. :-)

  • @kvu207
    @kvu207 5 місяців тому +2

    beautiful! May I ask how you made the animations for this video?

    • @jbhuang0604
      @jbhuang0604  5 місяців тому +1

      Most of the animations are from the “morph transition” in PowerPoint slides. The rest are from Adobe premier pro.

  • @yeon6761
    @yeon6761 5 місяців тому +1

    perfect video! Thank you for your works.

  • @jackshi7613
    @jackshi7613 5 місяців тому +2

    excellent video!

  • @Neo-kx3fe
    @Neo-kx3fe 5 місяців тому +2

    @10:55 with that local outgoingness on the left, why there is one additional term p_t(x_t) inside the d/dx bracket? This term seems to disappear in @11:18. Thanks.

    • @karnikram
      @karnikram Місяць тому

      The p_t(x_t) term on the right means that the temporal change in p_t(x_t) is also proportional to the current likelihood value and not just the vector field. However it gets cancelled out when you take the log-likelihood of p_t(x_t)

    • @NirGoren-k2k
      @NirGoren-k2k 19 днів тому

      @@karnikram Can you explain how exactly does it get canceled out?

    • @karnikram
      @karnikram 18 днів тому

      @@NirGoren-k2k d/dt log(p_t(x_t)) = 1/p_t(x_t) * d/dt p_t(x_t). We know the total derivative d/dt p_t(x_t) = -p_t(x_t) * div (u_t(x_t)) => d/dt log(p_t(x_t)) = -div (u_t(x_t))

  • @ahsentahir4473
    @ahsentahir4473 17 днів тому +1

    That was so good man!

  • @julienblanchon6082
    @julienblanchon6082 5 місяців тому +1

    This is brilliant !

    • @jbhuang0604
      @jbhuang0604  5 місяців тому +1

      Glad that you enjoyed the video!

  • @adrienforbu5165
    @adrienforbu5165 5 місяців тому +1

    nice visuals, good job

  • @tauhidkhan453
    @tauhidkhan453 17 днів тому

    Question:
    How CFM is different from Rectified Flows?

  • @ruoshiliu6024
    @ruoshiliu6024 5 місяців тому +1

    Amazing work Jia-Bin!!
    P.S. you should create a bibtex for this video so it can be cited in literature :P

    • @jbhuang0604
      @jbhuang0604  5 місяців тому +1

      Haha! Thanks! Too bad google scholar don’t include views of UA-cam videos.

  • @morrisfan2004
    @morrisfan2004 2 місяці тому

    I wondering if there are any urgent, potential applications of flow matching in industry?

    • @jbhuang0604
      @jbhuang0604  Місяць тому

      I think many of the recent text-to-image generation models are now trained with flow matching. There are also many other applications beyond image generations.

  • @emilbogomolov5709
    @emilbogomolov5709 29 днів тому

    How did nabla become divergence? at 10:59

  • @xulin730
    @xulin730 Місяць тому

    Nice Job

  • @JingHe-q6p
    @JingHe-q6p 17 днів тому +1

    I can't understant why z_* = u(z_* ), z_{k+1} = x-u(z_k), and x_{k+1} = x_k+\delta u(x_k) after 7:38

  • @rexton123
    @rexton123 3 місяці тому

    What are the advantage of flow matching compared to diffusion models?

    • @jbhuang0604
      @jbhuang0604  Місяць тому

      You can view it as a generalization of diffusion models. The training can converge faster and you won't have the difficulty where you cannot reach pure Gaussian distributions using finite diffusion steps.

  • @rexton123
    @rexton123 3 місяці тому

    Amazing

  • @shuchangzhou
    @shuchangzhou Місяць тому

    At 7:39, by "constrative map", do you really mean "Contraction mapping"?

    • @jbhuang0604
      @jbhuang0604  Місяць тому

      Yes, contraction mapping. en.m.wikipedia.org/wiki/Contraction_mapping

  • @dreadfulbodyguard7288
    @dreadfulbodyguard7288 Місяць тому +2

    Why is everyone saying this is so simple? I can't understand anything after 2:30

    • @jbhuang0604
      @jbhuang0604  Місяць тому

      Sorry about that! Probably should cover a bit more probability basics in the video as well.

    • @dreadfulbodyguard7288
      @dreadfulbodyguard7288 Місяць тому

      @@jbhuang0604 thanks for response. The problem is that I don't have maths background and your ideas are appearing very abstract. So, if you can add some python code to explain various mathematical expressions you are using, I think that would be very helpful for people like me.
      I am using chatgpt to do that currently.

  • @thienthuoan1081
    @thienthuoan1081 Місяць тому

    感觉有点像李宏毅老师的风格哈哈哈