155 - How many hidden layers and neurons do you need in your artificial neural network?

Поделиться
HTML-код
  • Опубликовано: 26 янв 2025

Комментарии • 87

  • @dinhh4248
    @dinhh4248 3 года назад +34

    16:36 is the answer for how many layers do you need

    • @NM-el6ps
      @NM-el6ps 3 года назад +3

      Thank you so much

  • @JoaoPedro-px5sx
    @JoaoPedro-px5sx 3 года назад +25

    Those were some of the best explanations about hidden layers and numbers of neurons I could find, also making it very easy to see in your python plots. Keep up the good work!

  • @nate4511
    @nate4511 3 года назад +13

    Finally, someone who speaks math. Thank you sir

  • @poffer31
    @poffer31 3 года назад +3

    The approach to explaining it trough linear regression was very useful for me, thank you!

  • @PatrickBateman12420
    @PatrickBateman12420 Год назад

    Finally, an explanation that goes straight into code. Awesome!

  • @andrewkicha1628
    @andrewkicha1628 2 года назад +1

    Such a wonderful explanation to the really fundamental question. I wonder why there is so little accessible information for beginners on this topic.
    Thanks a lot for the video.

  • @WahranRai
    @WahranRai 3 года назад +3

    22:55 Rule 3 contradiction with rule 2 : input = 50, output = 50
    . rule 2 gives hidden

  • @Samb12331
    @Samb12331 6 месяцев назад

    You are an amazing teacher, I hope this comment reaches you well! This is some top class free content! Thank you!

  • @wuzark
    @wuzark Год назад

    This is such an intuitive and helpful video. I can see that there is a lot of hard work behind this video. Great job!

  • @abhilashchaturvedi1479
    @abhilashchaturvedi1479 2 года назад

    I really loved your approach. You are explaining the technicalities and discussing the various possibilites while staying on the subject. It's thorough. With other youtubers, I felt like they were too basic and missing the crucial implementation part. Thankyou!

  • @AdobadoFantastico
    @AdobadoFantastico 3 года назад +1

    This is a fantastic explanation. I really appreciate how you involved the math as you walked through your implementation. A lot of people hand wave the math.

  • @3DComputing
    @3DComputing 3 года назад

    Strange isnt it, so many people asking THIS question, and so few people can answer it, THANK YOU

    • @DigitalSreeni
      @DigitalSreeni  3 года назад

      Well, I try to answer it but in reality it is difficult to definitively answer this question as so much depends on the nature of input data.

    • @HK-jw2et
      @HK-jw2et 2 года назад

      @@DigitalSreeni ruclips.net/video/pDXdlXlaCco/видео.html
      Hey. Can you pls help me in understanding how many nodes he used in this project. It's a project based on recognising sign language

  • @vaibhavs2522
    @vaibhavs2522 4 года назад +1

    I don't know why there are too few likes on such an awesome video..you are really great sir.

  • @shipengxu
    @shipengxu 2 года назад

    Thanks!

    • @DigitalSreeni
      @DigitalSreeni  2 года назад

      Thank you for your kind contribution. Keep watching.

  • @devishaarunadevitiwari3988
    @devishaarunadevitiwari3988 2 года назад

    Thanks for giving clarity on such an important notion. worth it.

  • @lakeguy65616
    @lakeguy65616 2 года назад +1

    adding hidden layers without activation functions is essentially linear regression. If the problem is linearly separable, you can find a solution. complex problems with non-linear solutions require hidden layers with activation functions. A more complex solution requires a higher number of hidden layers and activation functions. The "magic" is in the activation function.

  • @ahmedmoayadalhasani
    @ahmedmoayadalhasani 2 года назад +1

    Hi Sreeni, I had a significant mistake and training and test data differences. This, in my opinion, is due to the huge values of the output response numbers, which have increased from 64 to over a thousand. Please, how can I resolve this issue? Can I divide them by their maximum value to fix the issue? What do you prefer, please?

  • @tso8761
    @tso8761 3 года назад +1

    17:53 we can use dropout technique to reduce overfitting btw

  • @shivamsingh-fn8vz
    @shivamsingh-fn8vz 3 года назад +1

    ok so my doubt is i read on stack exchange and also ur 3 rd point in node section that neuron size should be 2/3 of input size so here the input size is equals to number of unique features or length of features input (len of dataset) and also 2/3 neuron = all the neurons in all the layers or only in single layer

  • @grdev3066
    @grdev3066 4 года назад +2

    "If your problem can be solved with linear fitting..."
    Me: trying to survive 2020 ...

  • @dragnar4743
    @dragnar4743 4 месяца назад

    So, basically there is no thumb rule. Ofc, it's understandable as it depends on data. So, we have to do hit & trial and observe the loss & accuracy from train-test set.

  • @surajshah4317
    @surajshah4317 4 года назад +3

    wow, thank you so much for the great video. Sir can you make videos on segmentation using GAN and UNet ??

    • @DigitalSreeni
      @DigitalSreeni  4 года назад

      They are already on my channel. Please explore videos on my channel.

  • @yacineyacine2951
    @yacineyacine2951 2 года назад

    this video is like finding gold ... thannnk youuu

    • @HK-jw2et
      @HK-jw2et 2 года назад

      ruclips.net/video/pDXdlXlaCco/видео.html
      Hey. Can you pls help me in understanding how many nodes he used in this project. It's a project based on recognising sign language

  • @pfever
    @pfever 3 года назад

    10:43 I think that if the learning rate is too small it could get 'stuck' in a local minima, isn't it?

  • @momchi2
    @momchi2 3 года назад

    many good hints and insights here

  • @cyruskavwele5304
    @cyruskavwele5304 2 года назад

    Fantastic explanation.

  • @troupebase2292
    @troupebase2292 2 года назад

    Thank you for this decent explanation

  • @vanithas4336
    @vanithas4336 Год назад

    Amazing..helps me lot for my research work. Thanks

  • @aqibfayyaz1619
    @aqibfayyaz1619 3 года назад

    Awesome video that is what i was looking for.

  • @danielniels22
    @danielniels22 2 года назад

    thanks sir, this is such an enlightenment 😂 ive been using 4 or even 6 layers by thinking that the model could learn very deep, like some unrecognized patterns 🤣🤣 but turns out just use 1 to 2 😭😭 thanks sir, im new to your channel this week btw 🙏

  • @neerav302
    @neerav302 3 года назад

    You are awesome .. you taught me this topic like pro

  • @rfreeman057
    @rfreeman057 2 года назад

    Excellent video!

  • @ad.donielson
    @ad.donielson 2 года назад

    How about neural network without hidden layer for classification?

  • @tesfayeabera4978
    @tesfayeabera4978 3 года назад

    please I need an explanation of how to increase the layer of deep belief network from three-layer to more than 6 and its advantages and disadvantage .

  • @tahseen4790
    @tahseen4790 4 года назад

    Thanks a lot. I regularly watch your Videos.

    • @DigitalSreeni
      @DigitalSreeni  4 года назад +3

      Thanks for watching my videos, I donate all money from RUclips advertisements to charity so please thanks for your contribution by watching part of the advertisements.

  • @RajeshSharma-bd5zo
    @RajeshSharma-bd5zo 4 года назад +1

    Great video and well explained!!

  • @madhumitagiribabu
    @madhumitagiribabu 3 года назад

    thanks much sir found right content after long search

  • @lerneninverschiedenenforme7513
    @lerneninverschiedenenforme7513 3 года назад

    I didn't understand what's happenning, when the number of hidden nodes increase. Does that also lead to overfitting?

    • @DigitalSreeni
      @DigitalSreeni  3 года назад

      Yes, increasing the number of nodes will also lead to overfitting. Anything that increases the nonlinearity in the model and makes it easy for the model to map training data will lead to overfitting.

    • @lerneninverschiedenenforme7513
      @lerneninverschiedenenforme7513 3 года назад

      @@DigitalSreeni Thank you!

  • @pramishprakash
    @pramishprakash Год назад

    Great video sir

  • @SomenathChakraborty
    @SomenathChakraborty 3 года назад

    Could you provide the data source details but it is very small dataset with very limited parameter. But I appreciate your video for clear clarification of the concept.

  • @somewhereinparallelunivers4226
    @somewhereinparallelunivers4226 3 года назад

    I have 7lakhs data.so can you suggest me how many neuron can i use for my neural network.. i am using curve fitting neural network.

  • @dikshitlenka
    @dikshitlenka 3 года назад

    Can't we use Keras Tunner to find the exact number of layers and neurons required in the network?

    • @DigitalSreeni
      @DigitalSreeni  3 года назад +1

      Keras is for hyperparameter tuning and I don't think it is for defining models. I may be wrong as I haven't explored Keras tuner much. If your goal is to find the best model for you problem I recommend AutoKeras.

  • @domenicobezuidenhout1587
    @domenicobezuidenhout1587 4 года назад

    For my thesis I am using weather data to predict future values using the CNN but for my loss and Val loss I get nan values? Do you know of a way I could fix this sir?

    • @DigitalSreeni
      @DigitalSreeni  4 года назад

      There are many reason why you’d get a NaN for loss and the most probably reason is high learning rate. If your learning rate is 0.01 try changing it to 0.001 and see if that helps.

  • @kakshah
    @kakshah 4 года назад +1

    Hi, thank you, may I know which tool is used to make this video?

    • @DigitalSreeni
      @DigitalSreeni  4 года назад

      Not sure what you are asking... can you be a bit specific?

  • @shubhamsongire6712
    @shubhamsongire6712 2 года назад

    Powerful explination

  • @heliyahasani6859
    @heliyahasani6859 2 года назад

    Least Squares Optimizer is same as Analytical Solution.(Wrote this comment to avoid confusion :) )

  • @ssonicmoumed
    @ssonicmoumed 4 года назад

    Thank you, great explanation.

  • @pallavi_4488
    @pallavi_4488 3 года назад

    you are worth listening

  • @tharindukanchana2077
    @tharindukanchana2077 3 года назад

    Thank you sir, very nice explanation

  • @TheConsoleMania
    @TheConsoleMania 2 года назад

    the number of neurons in both hidden layer, should be the same?

    • @DigitalSreeni
      @DigitalSreeni  2 года назад +1

      No, they can be anything.

    • @TheConsoleMania
      @TheConsoleMania 2 года назад

      @@DigitalSreeni yeah, but if I follow the rules in the video, i obtain about 12 neurones. This number should be the same on both hidden layer ? Or maybe the second one should be smaller ?

    • @TheConsoleMania
      @TheConsoleMania 2 года назад

      @@DigitalSreeni I have 12 input and 1 output

  • @aomo5293
    @aomo5293 2 года назад

    error again when calculating mean squared error !
    line 71
    you should use: np.mean((y_test--pred)**2) not np.mean(y_test-y_pred)**2 !!
    Thank you for good content

  • @mojoway9379
    @mojoway9379 3 года назад

    Thanks for the video very helpful

  • @abderrahmaneherbadji5478
    @abderrahmaneherbadji5478 4 года назад +1

    Many thanks

  • @hahavv7058
    @hahavv7058 3 года назад

    excellent vedio that give me great help!think you sir~

  • @tehpson
    @tehpson 3 года назад

    OMG thank you, I finallyu understand

  • @NoamRathaus
    @NoamRathaus 3 года назад

    TLDR; 1 or 2 hidden layers - or just guess because he doesnt know

    • @DigitalSreeni
      @DigitalSreeni  3 года назад

      This is an educational video intended to train the viewer on the implications of number of neurons and hidden layers. In fact, I try to design my content such a way that the viewer gains incremental knowledge on a specific topic. I am sorry if the title set a different expectation to you.

  • @علیشمسی-س7ب
    @علیشمسی-س7ب 2 года назад

    سلام.Hi