This is what should be explained before teaching anyone about Convolutional neural networks. Those photoshop custom filters & then the entire code thereon gave perfect visual intuition about convolutional layers. Thanks it was such an enlightening video.
4:38 - "Convolution Neural Networks can be applied to lots of scenarios besides images" - Dan, you sir are not only correct, you are fabulous for planting this seed early in the video. CNNs sure can be used for text, speech, music, and so much more than just images!
Thank you very much. I had to take a break in my Tensorflow class to understand convulutional filters better. This was a very good explanation. Thanks once again.
Hi Dan. I'd just like to thank you, you make everything much easier for me to understand. While I mainly focus on Python, your videos are still a massive help and source of entertainment. Thank you!
I have bad english, even i dont understand what he was said, but i just love the way how this amazing man explain every single his video, i love it, this make me wanna hear more and more, im just student, try to learn front end, i love programming, hope i can being like him someday,
Ohh the curious part of me feels so great after I watch any coding train videos even if it's not related to my work or any project that I am working on, It feels great appease my curiosity. I wish there could would be a space train channel where Dan talks about space that would be ultimate. What a guy :)
Thank you very much for making this. I was having trouble finding videos covering convolutional layers which attempted to cover the full concept of a filter.
I understand how to do all the steps. I set up a Jupyter notebook that does these operations on images. Exploring interesting kernels which select edges and all. I just have an issue with how to create a good dataset or even find one for what you are looking for.
If the cnn uses backprop to configure the filters, then what is the target output? I get the cat image is the input, but what is the desired output ( as well as how many output cells) to compare the input to?
Would love to see a series building out the CNN similar to your chapter 10 series for Nature of Code. It really shouldn't be so different as what you have already built (at least from the description that you gave here), isn't it just passing the error from the output => hidden => convolutional layer? and then using gradient descent on all those weights in the new layer's matrix for each node per pixel input? It seems way more computationally intensive than a standard NN, even if you have a 3x3 matrix for the convolutional layer isn't that pixel amount * 3 * 3 inputs into the hidden layer?
Yeah, I'm really interested about the way to make the backpropagation on cnn... I tried to make it 2 years ago from scratch but I missed something and my backpropagation isn't working. It should be nice to understand where I made bullet. :) Great work Dan, as always.
Found you from Code Bullet. Still curious as to why you teach JavaScript and JS-based stuff in specific, but it doesn’t really matter :p Also, nice video :)
I like being able to have all my examples run in the browser and also I'm invested in p5.js as a learning / creative coding platform through my work with processingfoundation.org/. So those are the main reasons for JS right now!
I would love to work this out sometime I am definitely interested also! But sadly it's lower priority for me as I have so many topics I want to get to and CNNs is one where I can make heavy use of existing implementations.
Love the videos. One thing that bothers me while watching these back and seeing you slowly over time adopt more recent ES syntax, you use let for every variable. Variables that are never changed should be defined as const.
You say, "Longwinded", and make it seem like that was too lengthy an explanation... I disagree! It was a great explanation. Altho... js seems to produce much confusing boiler plate code, but... that isnt your fault! hehe great hands-on video. :) *watches next vid*
Great video. But I think, you should always develop the subject from the scratch, without any libraries. I think we learn more about the subject, when it's not already implemented. Thus, people from other language can learn it. It mean there are more people watching you, since you regroup any kind of language for the same subject.
This is what should be explained before teaching anyone about Convolutional neural networks.
Those photoshop custom filters & then the entire code thereon gave perfect visual intuition about convolutional layers.
Thanks it was such an enlightening video.
Thank you sir so much, its individuals like you who actually make all the difference in the world
4:38 - "Convolution Neural Networks can be applied to lots of scenarios besides images" - Dan, you sir are not only correct, you are fabulous for planting this seed early in the video. CNNs sure can be used for text, speech, music, and so much more than just images!
I bought a cnn course in udacity and let me tell you that you explain it way better. You really helped me to understand the whole picture of CNN.
This is a really clear explanation of a convolutional nn. I've heard so many other people describe them, but your explanation is the clearest yet.
OMG, in just two minutes you clearly explained the concept of convolution which I was struggling to understand after reading a few papers 🙏🏾!
i just got what i wanted. i was suffering from google searches about CNN and finally Daniel, you started this series. Thanks, keep going man.
I really needed help with visualizing filters and this video helped me a lot! Thanks a bunch ^^
Still the best explanation of Convolutional Neural Network anywhere on the internet.
Thank you very much. I had to take a break in my Tensorflow class to understand convulutional filters better. This was a very good explanation. Thanks once again.
its been 2 years i'm following the channel, There is always something new to learn :)
Hi Dan. I'd just like to thank you, you make everything much easier for me to understand. While I mainly focus on Python, your videos are still a massive help and source of entertainment. Thank you!
One of the best videos for explaining filters Thank you!
Your videos are really fun to watch, and has real depth in the subject at the same time. Quite unique on TH-cam.
Give this man a medal.
I have bad english, even i dont understand what he was said, but i just love the way how this amazing man explain every single his video, i love it, this make me wanna hear more and more, im just student, try to learn front end, i love programming, hope i can being like him someday,
Ohh the curious part of me feels so great after I watch any coding train videos even if it's not related to my work or any project that I am working on, It feels great appease my curiosity.
I wish there could would be a space train channel where Dan talks about space that would be ultimate. What a guy :)
Thank you very much for making this. I was having trouble finding videos covering convolutional layers which attempted to cover the full concept of a filter.
Love this! Thank you for all that you do! You're an incredible teacher!
One of the best explanations on kernels (y)
I have been using photoshop for 20 years and had never really understood how the custom filter worked until now
1:05 - I think this is the day. Could you please make a series on convolutional networks from scratch.
Yes! I've been hoping you'd do this forever! Love you dan you're our patron saint haha
Thank you so much for making these videos. I am really grateful.
Great explanation.... Looking forward to the rest of the series.
Me: I would really like a new NN video from Daniel
Daniel: here you are
Awesome video. Can't wait to see the next part
I understand how to do all the steps. I set up a Jupyter notebook that does these operations on images. Exploring interesting kernels which select edges and all. I just have an issue with how to create a good dataset or even find one for what you are looking for.
You are just too amazing. Thank you!!
Great video! 👍🏻👍🏻
If the cnn uses backprop to configure the filters, then what is the target output? I get the cat image is the input, but what is the desired output ( as well as how many output cells) to compare the input to?
26:50 lmao the face that the cat is making
this is amazing, thank you so much for posting this
It would be really wonderful if you write your own code for convolutional layer for your Nature of code.
Would love to see a series building out the CNN similar to your chapter 10 series for Nature of Code. It really shouldn't be so different as what you have already built (at least from the description that you gave here), isn't it just passing the error from the output => hidden => convolutional layer? and then using gradient descent on all those weights in the new layer's matrix for each node per pixel input? It seems way more computationally intensive than a standard NN, even if you have a 3x3 matrix for the convolutional layer isn't that pixel amount * 3 * 3 inputs into the hidden layer?
Amazing. Thank you so much
That's great. Thank you!
Hi, nice video, as always! Will you explain how error is back propagated to nudge the filters, if it is? Thanks!
Yeah, I'm really interested about the way to make the backpropagation on cnn... I tried to make it 2 years ago from scratch but I missed something and my backpropagation isn't working. It should be nice to understand where I made bullet. :)
Great work Dan, as always.
With the sumR or G or B, you are adding propably 9 times 255. Why does it work? Am I missing something?
Indeed that is most likely happening. I guess p5 just converts any value above 255 to 255 and shows it as white.
Yeah that was going to be my question too. I’m also assuming it’s capped to 255 somewhere
thx tho first time i'm seeing something usefull without being too complicated on my ai researches. Want to do some now
I have a question please, I don't know if I can create a face recognition application with ml5 js? Thank you
I'd like to hear more about different research papers on deep learning
Could you please talk about building an OCR solution for handwritten text in the wild (not isolated characters)?
Loved this video!
Well done
I am looking forward to get the backpropagation explanation.... I have no idea how it works in convolutionnal network... Pleeasse !!!
Found you from Code Bullet. Still curious as to why you teach JavaScript and JS-based stuff in specific, but it doesn’t really matter :p
Also, nice video :)
I like being able to have all my examples run in the browser and also I'm invested in p5.js as a learning / creative coding platform through my work with processingfoundation.org/. So those are the main reasons for JS right now!
@@TheCodingTrain yeah ok that makes sense :D
Is it possible to make confusion matrix using ml5.js
Thanks a lot, your grate
Can you post videos based on python? Great Work !!!
What are you using to code
Thanks ❤️❤️
Oh great ! Convolutional Neural networks are ou subject at university !!
So this is how keras works right?
Can you do that from scratch? The most challenging part is backpropagation in CNNs for me. Would love to see how you would do that
I would love to work this out sometime I am definitely interested also! But sadly it's lower priority for me as I have so many topics I want to get to and CNNs is one where I can make heavy use of existing implementations.
Sir Plz do a video of experimenting GANs in ml5.js
Thanks!
Love the videos. One thing that bothers me while watching these back and seeing you slowly over time adopt more recent ES syntax, you use let for every variable. Variables that are never changed should be defined as const.
You say, "Longwinded", and make it seem like that was too lengthy an explanation... I disagree! It was a great explanation. Altho... js seems to produce much confusing boiler plate code, but... that isnt your fault! hehe great hands-on video. :) *watches next vid*
i don't really understand what's going on.. but it's fun to watch anyway
Why ml5.js instead of Python?
pet or plant or friend... isn't that the same ? i mean what could a friend be if not a plant or a pet ?
First
I am good with python wanna learn AI hope its not boring in the long run
Second
Great video. But I think, you should always develop the subject from the scratch, without any libraries. I think we learn more about the subject, when it's not already implemented. Thus, people from other language can learn it. It mean there are more people watching you, since you regroup any kind of language for the same subject.
My cousin is epileptic, he sometimes has convolutionals :(
you mean, this wasn't already with a ribbon, and a bow? could have fooled me
Third
Are you carrying the river of knowledge?
At first i read *Corona Virus Neural Network by coding train*
Lol🙏🙏🥴🥴
that was very un-convoluted for being about something so convolutional
convolution(t) = sum(x(u)*y(t-u),u,0,all u), not sum((x(u)*y(t+u),u all u)
Your explanation is a bit convoluted. 🤡