Backpropagation explained | Part 1 - The intuition

Поділитися
Вставка
  • Опубліковано 1 січ 2025

КОМЕНТАРІ • 125

  • @danluba
    @danluba 6 років тому +84

    This video deserves to be on television at dinner time.

  • @husseinwassouf2743
    @husseinwassouf2743 4 роки тому +20

    this course is so nice you are simplifying complex ideas in a really nice way, I think this series should be binge watched like a netflix series to refresh your memory on all these concepts XD

  • @GirlKnowsTech
    @GirlKnowsTech 3 роки тому +4

    00:00 Intro
    01:23 Recap stochastic gradient descent (SGD)
    04:42 What is backpropagation, the intuition
    08:28 Summary

    • @deeplizard
      @deeplizard  3 роки тому

      Added to the description. Thanks so much!

  • @parthasarathybalaraman1800
    @parthasarathybalaraman1800 6 років тому +18

    I have just completed watching all the 27 videos in the list. All are excellent videos!! Thanks for uploading. Please continue the good work!! Thanks!

    • @deeplizard
      @deeplizard  6 років тому +3

      Hey Parthasarathy - That's great to hear that you've gone through the entire playlist! I'm glad you're enjoying them!

  • @deeplizard
    @deeplizard  6 років тому +1

    Backpropagation explained | Part 1 - The intuition
    ua-cam.com/video/XE3krf3CQls/v-deo.html
    Backpropagation explained | Part 2 - The mathematical notation
    ua-cam.com/video/2mSysRx-1c0/v-deo.html
    Backpropagation explained | Part 3 - Mathematical observations
    ua-cam.com/video/G5b4jRBKNxw/v-deo.html
    Backpropagation explained | Part 4 - Calculating the gradient
    ua-cam.com/video/Zr5viAZGndE/v-deo.html
    Backpropagation explained | Part 5 - What puts the “back” in backprop?
    ua-cam.com/video/xClK__CqZnQ/v-deo.html
    Note, at 7:44, I misspoke when I stated that the updated values we get for the weights are the the corresponding derivatives of the loss function with respect to each weight.
    Actually, the updated values themselves are *not* the derivatives. Rather, after calculating the derivatives, the weights are updated to their new values, which are calculated *using* the derivatives we obtain. This process of updating the weights is covered in more detail in the following video. This particular detail is mentioned at 1:26: ua-cam.com/video/_N5kpSMDf4o/v-deo.htmlm26s
    Machine Learning / Deep Learning Fundamentals playlist: ua-cam.com/play/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU.html
    Keras Machine Learning / Deep Learning Tutorial playlist: ua-cam.com/play/PLZbbT5o_s2xrwRnXk_yCPtnqqo4_u2YGL.html

  • @RaghavendraBoralli
    @RaghavendraBoralli 6 років тому +47

    Yes i want to know on the math behind backprop

  • @maedehzarvandi3773
    @maedehzarvandi3773 4 роки тому +10

    your videos are actually great in expressing the concepts in possibly quickest time . i havnt seen your other playlists yet. but this one really helped me so much.

  • @justchill99902
    @justchill99902 6 років тому +3

    There is no limit to the awesomeness of her explanation!

    • @luna-kr3zc
      @luna-kr3zc 5 років тому

      there's a whole team behind this
      she is just the voice (but she does a great job at it)

  • @zw9423
    @zw9423 5 років тому +3

    Omg Thank You!
    All these books make it so complex and include a bunch of unnecessary equations before a noob like me can actually understand the process.
    I am interested in the math as well but first I have to know the concept, otherwise I often find myself understanding these complex details and be like, why exactly are we doing this?
    This video actually makes sense and now I can dive deeper into it. Nice animations too. Props to you guys

    • @deeplizard
      @deeplizard  5 років тому +2

      You're welcome, Zi! Glad you found it helpful. The math explanations are in the following videos of this playlist.

  • @nicholasscotto3712
    @nicholasscotto3712 2 роки тому

    after watching sooooooo many videos on this topic yours was EASILY the most helpful by far. It is accurate, simple, and just all around perfect. the image is very simple. I love your voice and the explanation is so good.

  • @ThePRINCEBARPAGA
    @ThePRINCEBARPAGA 5 років тому +3

    I cannot thank you enough. I don't know how many videos I watched to clear this concept, nothing helped but your video. You explained it in a very detailed and clear way. Thank You so much!

  • @hkrishnan91
    @hkrishnan91 3 роки тому +1

    Bless you for making the world a better place. Keep up the good work!

  • @yzyangliang
    @yzyangliang 5 років тому +6

    This is an amazing video and I understand the deep learning instantly!

  • @RohanPaul-AI
    @RohanPaul-AI 3 роки тому

    Just can not THANK YOU enough for this Greatest of videos on the topic. Just Brilliant.

  • @ismailelabbassi7150
    @ismailelabbassi7150 3 роки тому +1

    YES YES YES YES YES YES WE NEED MATH TOPICS

  • @tymothylim6550
    @tymothylim6550 3 роки тому +1

    Thank you very much for this video! It was fantastic that I could finally understand how SGD uses back propagation to calculate the gradients to minus off!

  • @cedricrogers8102
    @cedricrogers8102 6 років тому +1

    This was a very clear explanation of what happens during back propagation. The next step is to provide the math. Thank you for your effort. Great Job.

    • @deeplizard
      @deeplizard  6 років тому +1

      Thanks, cedric! The full math for backprop is covered in the videos following this one! :)
      They are currently #28 - 31 in this playlist:
      ua-cam.com/play/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU.html

    • @cedricrogers8102
      @cedricrogers8102 6 років тому

      deeplizard I will watch them thanks again

  • @mohamedmahdy969
    @mohamedmahdy969 6 років тому +4

    I already know the math behind back propagation; however, I will watch your videos in order to see how you are going to present it. Your way of giving the information is awesome. I want to see how it will work with the math complexity.

    • @deeplizard
      @deeplizard  6 років тому +1

      Thanks, Mohamed! Let me know your thoughts after you finish the following math videos.

    • @mohamedmahdy969
      @mohamedmahdy969 6 років тому +2

      @@deeplizard To be honest, as I mentioned earlier, I already know the math behind the back propagation, yet your videos were a good refresh to me. You used the same mathematical notation and the same methodology of my teacher. he started with the last hidden layer; after that, he generalized to any hidden layer.
      Thanks a lot for your videos.I already finished the math videos and I will finish this list today. In the future, I am going to watch tensorFlow.js series

    • @deeplizard
      @deeplizard  6 років тому +1

      Great to hear, Mohamed! Thanks for letting me know. Would love to hear how your progression is going in the TensorFlow.js series as well once you start!

  • @rajuthapa9005
    @rajuthapa9005 6 років тому +21

    yes math too please.

  • @javierCi
    @javierCi 6 років тому +13

    Thanks you very much for making this great videos. And I want to know more about maths

  • @deepaksingh9318
    @deepaksingh9318 6 років тому +4

    Yess.. Adding a specific Playlist with Maths behind all the functionality with an example (Like uh showed for max pooling which was visible to see hoe things are working with input) would be really great.. Bur still appreciate your efforts in helping students by such a great videos..

    • @deeplizard
      @deeplizard  6 років тому +2

      Thanks, deepak. The next four videos in this playlist show the full math behind backpropagation. Let me know how it goes as you progress through them!

  • @qusayhamad7243
    @qusayhamad7243 4 роки тому +1

    thank you very much for this clear and helpful explanation.
    Words fail to express my gratitude.

  • @TheMeltone1
    @TheMeltone1 2 роки тому +1

    You are truly a fantastic teacher :)

  • @paragjp
    @paragjp 4 роки тому

    Hi, Thanks for super clean way to explaining basic concepts. Keep it up. Request you for following video series 1. Complete Maths and Stats for ML/Deep Learning. 2. Pl have additional series on calculus on backpropogation. 3. Make complete new series on ML/Deep Learning Practical projects rather than Housing Price Prediction, Titanic, iris, Hand written digits ....etc

  • @travel7517
    @travel7517 6 років тому +2

    great..explaination...simple and effective

  • @MsStoCa
    @MsStoCa 4 роки тому +4

    Hey! What's going on everyone? :D
    Great content!

  • @robertoooooooooo
    @robertoooooooooo 6 років тому +2

    Amazing, love your videos, will recommend it to all my friends

  • @richarda1630
    @richarda1630 3 роки тому +1

    such an awesome super simple explanation! and how you bring up the math, it's almost like a teaser :) now I want to see the Math! =)

  • @DM-py7pj
    @DM-py7pj Рік тому

    These are excellent videos. I do worry about biases not being discussed in later videos as these are also being updated.

  • @1sankey2
    @1sankey2 4 роки тому +3

    Great video I wanna see the math. Thank you for uploading this one.

    • @deeplizard
      @deeplizard  4 роки тому +2

      Great :D It's in the following episodes. Starting with this one:
      deeplizard.com/learn/video/2mSysRx-1c0

  • @samaryadav7208
    @samaryadav7208 6 років тому +3

    Wow that was exciting Hi. Thanks for sharing these videos. And Also we want to see the maths.

  • @fahimmahmud3115
    @fahimmahmud3115 4 роки тому +1

    {
    "question": "Since the derivative of the loss function is calculated with respect to the weights of the model during backpropagation, what characteristics should the loss function have?",
    "choices": [
    "Continuous",
    "Discontinuous",
    "Constant",
    "Zero"
    ],
    "answer": "Continuous",
    "creator": "Fahim Mahmud",
    "creationDate": "2020-02-12T04:10:35.701Z"
    }

    • @deeplizard
      @deeplizard  4 роки тому +1

      Thanks, Fahim! Just added your question to deeplizard.com/learn/video/XE3krf3CQls :)

  • @flamboyanz
    @flamboyanz 6 років тому +2

    Thank you so much for this! Its concise and well explained.

    • @deeplizard
      @deeplizard  6 років тому

      You're very welcome, Parvez!

  • @dan7582
    @dan7582 2 роки тому

    At 8:00 you say that the update values of the weights are equal to their derivatives (gradient).
    But since we are talking about a loss function, they should be equal to de NEGATIVE gradient instead?

  • @loganmay2105
    @loganmay2105 4 роки тому +5

    Never thought I'd hear the words "chain rule" outside of my old high school AP calc class

  • @krishnaik06
    @krishnaik06 6 років тому +4

    Math too please

  • @faraazmohammed3693
    @faraazmohammed3693 6 років тому +3

    please make a series on videos explaining math behind machine learning and deep learning in particular. Thanks for Amazing videos.

    • @deeplizard
      @deeplizard  6 років тому

      Hey Faraaz - You're welcome! Did you see the next four videos in the playlist that cover the full math for backprop?

  • @vantongerent
    @vantongerent 3 роки тому

    great video - yes I want to see the math! 🙂

  • @mjain2172
    @mjain2172 4 роки тому +1

    Yes, please explain math behind of this calculation, thanks for explaining deep neural network concepts

    • @deeplizard
      @deeplizard  4 роки тому

      Great :D The math is included in the following episodes. Starting with this one:
      deeplizard.com/learn/video/2mSysRx-1c0

  • @krishnasaibiradar
    @krishnasaibiradar 4 роки тому +1

    Thanks for the awesome video on backpropagation Yes i want to know the math behind backpropagation as the whole logic lies in and around math.So kindly make a video on the math behind the Backpropagation.

    • @deeplizard
      @deeplizard  4 роки тому

      You're welcome, krishnasai! There are 4 episodes following this one that explain the math :) And they all have corresponding blog articles along with the video!
      The next one starts at the link below, and the following 3 are directly after.
      deeplizard.com/learn/video/2mSysRx-1c0

  • @fritz-c
    @fritz-c 4 роки тому +2

    I noticed a broken link in the article for this video, near the words:
    "shown in an [earlier post]."
    It looks like part of the url got duplicated.

  • @omargonalfa
    @omargonalfa 5 років тому +1

    Awesome video. please a brief example of the math behind.

    • @deeplizard
      @deeplizard  5 років тому

      The math is shown in the following four videos of the series :D

  • @rohtashbeniwal9202
    @rohtashbeniwal9202 4 роки тому +1

    great videos,,saveed,subscribed ,

  • @VISWESWARAN1998
    @VISWESWARAN1998 5 років тому +1

    Thank you so much

  • @nourelislam8565
    @nourelislam8565 2 роки тому

    Amazing...
    Before asking about the math courses for the backpropagation. Is it essential to study the math behind deep learning in order to work with Keras or any other APIs?
    Thanks for these fruitful videos!

  • @zrmsraggot
    @zrmsraggot 5 років тому +1

    Why cant we just change the weights from pink to blue nodes since we just want activation in blue nodes to change to be efficient

    • @ojasvinnagpal9472
      @ojasvinnagpal9472 4 роки тому

      I have a similar question: Why can't we simply change the weights between the blue and the yellow nodes and arrive at the optimal values for the hidden layer units?

    • @ojasvinnagpal9472
      @ojasvinnagpal9472 4 роки тому

      Is it the case that in order to change the nodes in each layer, we need to change BOTH the nodes in the previous layer as well as the weights between the previous layer and the current layer?

  • @Arcaerus
    @Arcaerus 2 роки тому

    I would greatly appreciate if you could make a series on how to take the derivative of a bunch of layers! Thank you for these videos!

  • @jesilmohammed7926
    @jesilmohammed7926 5 років тому

    What is the default height and width of a conv_2d filter ?

  • @generalzeedot
    @generalzeedot 4 роки тому

    would this backpropagation technique, therefore, only be usable in supervised learning?

  • @sanketkachole9639
    @sanketkachole9639 2 роки тому

    yes we need it

    • @deeplizard
      @deeplizard  2 роки тому

      You got it - Math starts in the next episode :D
      ua-cam.com/video/2mSysRx-1c0/v-deo.html

  • @richardme9928
    @richardme9928 5 років тому +1

    Any links to videos that do the complete math calculations ?

    • @deeplizard
      @deeplizard  5 років тому

      Hey richard - They are the following videos in this playlist. Check out the full series in order here:
      deeplizard.com/learn/playlist/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU

  • @abail7010
    @abail7010 5 років тому +1

    Something isn't clear tome.
    Do we update the bias for each neuron or is this example without using bias?
    Apart from that, I love your videos and the way you explain things! Please keep up since this is a highly interesting thing. :)

    • @deeplizard
      @deeplizard  5 років тому

      Thank you, Xi! I've left bias out of this example, but they get updated at the same time and in the same fashion as the weights.

    • @abail7010
      @abail7010 5 років тому +1

      Thank you for the reply! I've not seen your video to the bias which cleared my question when I commented. :)

  • @jabrams3643
    @jabrams3643 6 років тому

    i'm confused about the difference between the gradient and derivative? Is it called a gradient of the loss function when you average out all of the derivatives of the loss wrt to a certain weight? Meaning that its called a derivative when its just dloss/dweight for a single sample, and its called a gradient when its the average dloss/dweight for all of the samples? thank you for this amazing series btw it is the best on the internet

  • @GelsYT
    @GelsYT 4 роки тому

    Hello :D Is the output of the activation equal to the weight ? i mean is the activation the weight? of every neuron firing? right? cause i think that's what I remember about your video on activation function
    so basically the output of the activation function is going to be the weight? thank you so much great tutorial !
    you explain it at the simplest way :))

  • @GelsYT
    @GelsYT 4 роки тому

    sooo gradient descent is the one who's updating the weights? thank you soo much

  • @stydras3380
    @stydras3380 4 роки тому

    I can only agree with the other comments, knowing the math is neat :D Is there some sort of follow up video of yours?

    • @stydras3380
      @stydras3380 4 роки тому

      Ok, I have found the videos discussing the math :D

  • @xeroxSoldier
    @xeroxSoldier 5 років тому

    Love your videos, really helping me with my master thesis! :)
    You're talking about SGD but there are also other optimizers. I'm interested especially in Adam, and how it differs from SGD, maybe you got some paper or article recommondations?
    - In my understanding Adam is doing the exact same thing as SGD just using another algorithm, is that correct?
    - You're talking about backpropagation referring to SGD, is Adam also using backpropagation? In my understanding, backpropagation ist just the general term of changing the weights after every forward propagation.

  • @meetayan15
    @meetayan15 6 років тому +1

    hi, do you have any lecture on SGD ?

    • @deeplizard
      @deeplizard  6 років тому

      Hey Ayan - The following two videos (in the order listed) talk about SGD:
      1. ua-cam.com/video/sZAlS3_dnk0/v-deo.html
      2. ua-cam.com/video/_N5kpSMDf4o/v-deo.html
      Note that SGD uses backpropagation during training, which is where most of the "grunt work" comes into play.
      So after generally understanding what SGD is doing from the above videos, this 5-part backprop series, starting with the video we're currently commenting on, gives all the details for what backprop is doing during the training process.

  • @herohari27
    @herohari27 6 років тому

    In which video you were talking about SGD?

    • @deeplizard
      @deeplizard  6 років тому

      Hey Hari - These two:
      deeplizard.com/learn/video/sZAlS3_dnk0
      deeplizard.com/learn/video/_N5kpSMDf4o

  • @rajuthapa9005
    @rajuthapa9005 6 років тому +1

    r u bringing RNN tut too?

    • @deeplizard
      @deeplizard  6 років тому

      Hey Raju - Yes, I have RNNs on my list to cover in future videos.

  • @torbjornstorli2880
    @torbjornstorli2880 6 років тому

    Is the backpropagation being applied only once per epoch? So, if you have 20 epochs you will perform backpropagation 20 times, once per epoch ?

    • @deeplizard
      @deeplizard  6 років тому

      Hey Torbjorn - It occurs at each batch. The details for this implementation are covered in the backprop videos that come after this one in the playlist.

  • @sprajapati2011
    @sprajapati2011 4 роки тому +1

    7:17
    'ie' is pronounced as 'that is'
    not i e itself

  • @aliasgarzakir4779
    @aliasgarzakir4779 5 років тому

    Yes please, math is fun.

    • @deeplizard
      @deeplizard  5 років тому

      All the math is in the episodes following this one :D

  • @torbjornstorli2880
    @torbjornstorli2880 6 років тому

    That is per batch per epoch

  • @aamir122a
    @aamir122a 6 років тому +1

    same here.

  • @ahdm1319
    @ahdm1319 6 років тому

    please explain the maths behind this in seperate videos

    • @deeplizard
      @deeplizard  5 років тому

      The math is shown in the next four videos after this one. Let me know what you think!

  • @saranshtayal2526
    @saranshtayal2526 4 роки тому

    i want to know about the math

    • @deeplizard
      @deeplizard  4 роки тому

      It's in the following episodes after this one :)

  • @josegregorioperezmagallane3211
    @josegregorioperezmagallane3211 5 років тому +2

    We want to Sep the math

    • @deeplizard
      @deeplizard  5 років тому +1

      The math explanation is in the following four videos of the series :D

  • @urvashidang6083
    @urvashidang6083 4 роки тому

    i want to know the maths behind

    • @deeplizard
      @deeplizard  4 роки тому

      Great! It's in the following episodes. Starting with this one:
      deeplizard.com/learn/video/2mSysRx-1c0

  • @danluba
    @danluba 6 років тому +1

    And yes, I could go for some math.

    • @deeplizard
      @deeplizard  6 років тому

      The math for backprop starts in the next video in the playlist! Here are the full details for the backprop series:
      Backpropagation explained | Part 1 - The intuition (this video)
      ua-cam.com/video/XE3krf3CQls/v-deo.html
      Backpropagation explained | Part 2 - The mathematical notation
      ua-cam.com/video/2mSysRx-1c0/v-deo.html
      Backpropagation explained | Part 3 - Mathematical observations
      ua-cam.com/video/G5b4jRBKNxw/v-deo.html
      Backpropagation explained | Part 4 - Calculating the gradient
      ua-cam.com/video/Zr5viAZGndE/v-deo.html
      Backpropagation explained | Part 5 - What puts the “back” in backprop?
      ua-cam.com/video/xClK__CqZnQ/v-deo.html

    • @danluba
      @danluba 6 років тому

      Yeah - I found it. Awesome stuff. Thank you!

    • @deeplizard
      @deeplizard  6 років тому

      Awesome, you're welcome!

  • @yvesa6959
    @yvesa6959 3 роки тому +1

    give maths pls

    • @deeplizard
      @deeplizard  3 роки тому

      The math comes in the episodes that follow this one :D

  • @lando6583
    @lando6583 6 років тому +1

    i don't want to know about the math.

    • @deeplizard
      @deeplizard  6 років тому +2

      Warning: Don't watch parts 2 - 5 of the backprop videos 😜

  • @wesleyrademaker5167
    @wesleyrademaker5167 5 років тому +1

    math please!

    • @deeplizard
      @deeplizard  5 років тому

      Hey Wesley - The math is in the following four videos after this one in the playlist!

  • @akshatsahu2637
    @akshatsahu2637 5 років тому

    The background is so bad. The image and background blend

    • @deeplizard
      @deeplizard  5 років тому

      I agree the image and the background don't have enough contrast with each other. The background has changed in later videos.

    • @akshatsahu2637
      @akshatsahu2637 5 років тому

      @@deeplizard rest all the videos are very very good. The way everything is explained is awesome!. Better that course era.
      One last things, are there any videos in which the programming part is explained or the dimensions of the error or Delta, and the dimensions of theta?

  • @MrFranciscoooooo
    @MrFranciscoooooo 4 роки тому

    A video for this a video for that!!
    Just teach or do a brief sum of what this is, because if I want to search more I will do it from other videos not just yours.

  • @jesilmohammed7926
    @jesilmohammed7926 5 років тому

    What is the default height and width of a conv_2d filter ?