Accelerate Gradient Descent with Momentum (in 3 minutes)

Поділитися
Вставка
  • Опубліковано 15 гру 2024

КОМЕНТАРІ • 35

  • @jakubjan44
    @jakubjan44 Рік тому +20

    Brilliant videos! If I may suggest one thing, it would be to set the background music a bit quieter.

  • @BrianAmedee
    @BrianAmedee 3 роки тому +2

    I love watching these quick explainers! Well done this one was good

  • @dave1b
    @dave1b 5 місяців тому +5

    Great video, the music though is way to loud.

  • @szymonkaczmarski8477
    @szymonkaczmarski8477 3 роки тому

    Your videos are great! Keep up the good work, we need you!

    • @szymonkaczmarski8477
      @szymonkaczmarski8477 3 роки тому

      Some ideas for future videos: basis expansion, kernel functions.

    • @VisuallyExplained
      @VisuallyExplained  3 роки тому

      Thank you for the video suggestions, it's helpful. :-)

  • @shlokpikle8341
    @shlokpikle8341 3 роки тому +2

    Too good!
    Can you also make a video on Adam optimiser?

  • @zeinmoualla923
    @zeinmoualla923 Рік тому +3

    I love your work man! One note, please stop the music it is distractive.

  • @kristinnkristinsson1369
    @kristinnkristinsson1369 9 місяців тому +1

    I take issue with your analogy to a ball rolling on surface. You add the friction term as being proportional to the velocity of the ball. In fact friction is not proportional to velocity, it is constant in magnitude. So the friction term that appears at 3:00 when you've set the time step to dt=1 should read -F_mu*e_{v_{k-1}}, where e_v is the unit vector in the direction of v. This is not pedantry either, since this means that when you combine the first (old velocity) and third (friction velocity change) terms, the result isn't proportional to the old velocity, and cannot be simplified as alpha*v_{t-1}. This means that the difference of behavior of a ball on a surface and momentum gradient descent is not a matter of tuning parameters, but is rather fundamental.

  • @ekaakbarpermatasari2376
    @ekaakbarpermatasari2376 4 дні тому

    Brilliant videos! I wanna request, please make a video about Hamiltonian Monte Carlo Algorithm. 😃

  • @AJ-et3vf
    @AJ-et3vf Рік тому

    Great video. Thank you

  • @aakashyadav1589
    @aakashyadav1589 4 місяці тому

    0:20 wrong, updates happen in opposite to gradient

  • @antonisstellas741
    @antonisstellas741 Рік тому

    amazing explenation!

  • @pnachtwey
    @pnachtwey 7 місяців тому

    how can v_k be used before it is calculated in the next line? How can one know the 'condition' if this is actual data and not a mathematical formula?

  • @vTutorLive
    @vTutorLive 3 роки тому +1

    I enjoyed this video a lot. I found this channel a few weeks back when you shared it with r/learnMath. We had talked a bit about some audio tuning - I can't quite tell if you changed your settings, but your mic technique sounds a little better. Maybe it's placebo, but I think it sounds a bit more dynamic, though I can still hear a bit of sibilance here and there (I'm no expert myself, but I try to attack it if I can hear it using some plugins).
    Can I ask what your background is as it relates to math visualization - were you passionate about teaching during school and just decided to pick up the visualization software? I'm amazed you found the time and energy to learn Manim and put together such high production visuals, and I think the M.O of this channel is great. I'll be sharing these videos with my students

    • @VisuallyExplained
      @VisuallyExplained  3 роки тому +2

      Hi there, very nice to run into you again. Your audio tips helped a lot actually, thanks! Math visualization has been a hobby of mine since I started giving talks during my PhD. It does take a long time to get right, but I kinda enjoy the process.
      I checked some of your video, and I can see that you are doing a great job making math fun and interactive. (And your sound quality is definitely better than mine.)

    • @vTutorLive
      @vTutorLive 3 роки тому +1

      @@VisuallyExplained Yeah I started a livestream platform this semester to provide math help as a hobby. It's been pretty fun and quite a few students have told me that it's made the difference between them staying with/dropping their calculus courses. It's a pretty good fit for me (Bachelor's in Math/CS) because I can provide a lot of hands-on help with intro/intermediate material for many students, but don't have the experience to do visuals or higher-level talks.
      Actually my audio settings are a little hamfisted right now for most of my videos because they are direct rips of my streams, and I don't get to use all my plugins because the audio processing is through a different workflow. Trying to improve that as I go along!

  • @TheKlaMike
    @TheKlaMike 3 роки тому +1

    do you publish the code behind these videos? (manim stuff in particular) these are awesome!

    • @VisuallyExplained
      @VisuallyExplained  3 роки тому +2

      Thanks ! I did for a few of the earlier videos (e.g., github.com/bachirelkhadir/manim-kalman and github.com/bachirelkhadir/manim-newton). I plan to release the code for the new ones once I get a chance to clean up the code a little bit.

  • @alaanasser3243
    @alaanasser3243 2 роки тому +2

    Amazing content!! May I ask which software do you use for the 3d visualisation?

  • @fabricetshinangi5042
    @fabricetshinangi5042 3 роки тому

    Nice presentation

  • @yli6050
    @yli6050 Рік тому

    Very nice!

  • @ajaya652
    @ajaya652 2 роки тому

    Excellent video😍

  • @chriswendler5464
    @chriswendler5464 2 роки тому

    Amazing!

  • @patakhte6367
    @patakhte6367 3 роки тому

    thank you so mach

  • @fatihburakakcay5026
    @fatihburakakcay5026 3 роки тому

    Awesome

  • @alibastami
    @alibastami 8 місяців тому +1

    music was too loud and noisy

  • @NikolajKuntner
    @NikolajKuntner 3 роки тому

    nice.

  • @AndresGarcia-pv5fe
    @AndresGarcia-pv5fe 7 місяців тому

    good but why the loud ass shopping music