👉 Check out the blog post and other resources for this video: 🔗 deeplizard.com/learn/video/YRhxdVk_sIs 👀 Come say hey to us on OUR VLOG: 🔗 ruclips.net/user/deeplizardvlog
Thanks, Yousuf! However, when I click on the link, it takes me to the playlist without error. Pasting it below for you just in case: ruclips.net/p/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU Additionally, we have corresponding blogs for most videos in this playlist at the link below as well! deeplizard.com/learn/playlist/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU
Great video. One small thing I wanted to point out: when convolving a filter throughout an image, generally, it's more accurate to center the filter on the pixel you want to apply on rather than using the topleft pixel of the filter. This probably doesn't matter a whole lot in CNNs, but it does if you are looking to use a specific filter like edge detection, corner detection, etc.
Great video! I learned a lot from watching it. I just want to point out that at around 5:10, the convolution is not a dot product of two matrices. It's simply A[0,0] * B[0,0] + A[0,1] * B[0,1] + A[0,2]* B[0,2] + ...+ A[2,0]*B[2,0] + A[2,1]*B[2,1] + A[2,2]*B[2,2]. First number in the bracket represents row number, and the second represents coloum number. The dot product of two 3x3 matrices should be another 3x3 matrix.
Hey MakersAll - Thank you! Yes, you're right. What we're doing is summing the element-wise products of each pair of elements in the matrices. I elaborate more on this in the corresponding blog in the section "a note about the usage of the dot product": deeplizard.com/learn/video/YRhxdVk_sIs
Thats the comment i was looking for! for a momment I thought I forgot what a dot product was. Glad to see that im not the only one that noticed that mistake.
In addition to what you mention about the dot product, I would like to point out that what is depicted in the video is not a convolution but actually a correlation. It easier to explain a correlation (because it comes down to a simple element-wise multiplication), but the elements are multiplied differently in a convolution (first element of the image patch is multiplied by the last element of the kernel/filter, second element of the image by second last and so on...)
It was for me too, Sam! I use the same Excel example to illustrate zero padding and max pooling as well, which are tools/techniques commonly implemented in CNNs. If you liked this explanation, then you may be interested in those as well! Zero Padding: ruclips.net/video/qSTv_m-KFk0/видео.html Max Pooling: ruclips.net/video/ZjM_XQa5s6s/видео.html
i'm a college cs student who is taking a class on ML (and I have final exam tmrw lmao). this video was such a huge help. I learned more from this video than my professor's entire three day lecture on CNNs thank you!!
THIS IS IT. Best ever playlist for clearing your prolonging doubts. THE explanations in each video are just so enlightening. so glad I stumbled upon this playlist. ;)
Thanks for such a clear, to-the-point explaination. I like that you directly pointed out the key behind CNN is nothing but layers of filters. Very well done!
I like the way you used illustrations to explain the idea behind. It is usually not very easy to imagine all this without the help of visual material. Good work :)
Thanks! I've been taking baby steps into machine learning and this has been one of the cleanest and most easily understood introductions to convolutional networks that I've ran into.
Thank you!!! This is the best explanation i've found so far. Mainly because you talk about the mathematics of it with the simplest example. Which is exactly what I was looking for
Usually people say that Convolutional Neural Networks are the hardest of all types of neural networks to learn to use, but you made the basic idea of it clear as the sky.
Tbh you won't know till you try. People say the kanji in Japanese is really hard to learn, but when you try it out in ain't so bad. I guess it depends on the individual to whether it is hard or easy.
Previously watched another video on CNNs - which was way longer - and by the end, I had no idea. In just 8 mins of your video, it was like a light bulb moment. Fantastic work.
Hi! We are researchers in human-computer interaction (HCI) looking for people who have taken an initiative to recently learn Machine Learning on their own, for career, course or curiosity. Seems like you are in that place currently. Would you mind telling us here (www.surveymonkey.ca/r/SelfLearning_ML) about your experiences and any difficulties you faced while self-teaching ML and how you overcame them. There is also a chance to win $50 giftcard. You can help this project by taking out 5-10 minutes to participate in our study. For more details, see here: www.surveymonkey.ca/r/SelfLearning_ML Please share this request with your colleagues or friends who fit this description. People from any major/background may participate. The survey will be open until July 31, 2020.
This is BY FAR the best explanation ever given for CNN and makes sense. Thank You. For other explainers: "If you can't explain it simply, you don't understand it well enough." :P
Excellent and clearly explained. Thank you for taking the time to make this video! I've been exceedingly frustrated by other content that doesn't make CNNs as intuitive as you did. Thanks again!
I've been watching explanations on CNNs for over a year now and the part after 4:28 is the best i've seen yet, The excel visualization of images and filter convolving is fantastic thank you very much for this excellent explanation
I feel moved to say some big thing to praise your efforts, but the reality is that I wouldn't be doing computer science if I had a gift for speaking. Big kudos, this video is exactly what I needed to understand convolution.
Have been watching deep learning videos all day since a couple of days, to get the basic knowledge for a big assignment at university (which is due in 4 days :( ). Terribly sleepy, just decided to watch one final video before calling it a night. Decided on this as it was the shortest. Didn't realise it would be this enlightening! Subscribed and look forward to watching your playlists. Thank you!
This is my first learning video on neural networks and you made it so easy to understand. And i come from a hardware background, only knowledge that i had around this was digital signal processing. Kudos to your explaining skills!
This video is amazing, starts slow and builds up on it. The excel sheet visualization was amazing. Subscribed! That being said, I wish you would continue to explain, what we are getting after the convolution filters, pooling etc.
Thank you, Yousuf! Glad to hear you enjoyed the video! In regards to what we're getting after the convolutional filters, we have more videos and blogs in this series that explain everything that comes after convolutional layers (like pooling, padding, etc.), as well as how convolutional filters can be visualized. DL Fundamentals: deeplizard.com/learn/playlist/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU Even more content that pertains to these topics is available in our PyTorch series as well. Pytorch: deeplizard.com/learn/playlist/PLZbbT5o_s2xrfNyHZsM6ufI0iZENK9xgG Enjoy!
I'm a software developer and heard about CNN for a while, but didn't know how it works. Right now I kinda understand it. Computer science fascinated me again. Hooray for we human.
I've been learning CNN from another course but this made me visually realise what I was learning. Illustration like these leave a lasting impression in mind and help in understanding complex concepts in future. Thank You !
Great video. The jargon is always the thing that gets on the way for me. Your walkthrough of the visualisation really helped me build intuition about what convolutional layers add to a network.
Thanks so much for the video! It was always a struggle to understand CNN (cuz ppl explain CNN in a most complicated way😢) but your video really helped in understanding the core of CNN. Keep up the good work 🤗
This is the first time I get introduced to CNN, and you made it interesting :) . Your way of teaching is very engaging. Though i have few questions, hope you will reply soon. Q-1) What kind of filter functions over receptive windows are convolutional layers learning? (a)Softmax (b)Sigmoid (c)tanh (d)Linear functions followed by any activation Q-2) Which of the following are true? (a)CNNs can detect features in any portion of an image (translation invariance) because of strides and maxpooling (b)CNNs lose information on spatial location of features because of max pooling. (c)CNNs with convolutional layers followed by fully connected layers can process images of any size (d)All of the above (e)None of the above Q-3) In a neural network consisting of max pooling layer, what is the multiplication factor for a gradient passing through a neuron which is non-maximal in the pooling window? * (a)-1 (b)0 (c)1 (d)Output of maximal neuron Q-4)Given an all-white image to a well trained CNN based image classification model, are the output probabilities for each class the same? * (a)Yes (b)No (c)Can’t say Q-5)Which of the following patterns can CNNs learn? (a)Spatial patterns in 2-D (b)Spatial patterns in 3-D (c)N-gram based patterns in text (d)All of the above (e)None of the above Q-6) Which of the following layers are useful for up-sampling upto a pixel level, features extracted using convolution in a CNN? (a)Fully connected layers (b)Convolution layers (c)Transposed convolution layers (d)Softmax layers (e)All of the above Q-7)Larger receptive field in a CNN results in_ (a)Better model performance (b)Larger number of model parameters (c)Both (d)Neither
Please point it out if there are wrong points - CNN has Convolutional Layers as hidden layers - Filter can be a pattern detector. - You can set multiple filters for each conv layers, and randomly assign values for each filter. - At the Conv layer, the filter of the layer slide over the entire inputs. (If a filter is 3x3 table, the dot product will be computed with 3x3 table of input matrix) - As the layers be deeper, the filters can become more sophisticated and detect much more complex patterns.
![Seungmin "그렇게, 천천히, 우리(As we are)" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/kAzmhLHePqU/mqdefault.jpg)
👉 Check out the blog post and other resources for this video:
🔗 deeplizard.com/learn/video/YRhxdVk_sIs
👀 Come say hey to us on OUR VLOG:
🔗 ruclips.net/user/deeplizardvlog
Hey! The link to your Machine Learning playlist in the description isn't working.
Thanks, Yousuf! However, when I click on the link, it takes me to the playlist without error. Pasting it below for you just in case: ruclips.net/p/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU
Additionally, we have corresponding blogs for most videos in this playlist at the link below as well!
deeplizard.com/learn/playlist/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU
Great video. One small thing I wanted to point out: when convolving a filter throughout an image, generally, it's more accurate to center the filter on the pixel you want to apply on rather than using the topleft pixel of the filter. This probably doesn't matter a whole lot in CNNs, but it does if you are looking to use a specific filter like edge detection, corner detection, etc.
Hello deeplizard, your videos are so good, can you please share with me what application you use to prepare them
great explaination
I'm a data scientist using ML, DL and CNN's in specific on a daily basis. This series is one of the best I've seen. Well done and thank you!
Hi
Can you send me your WhatsApp number to ask you some questions about DL and CNN
wow I wanna be a real data scientist too
not only in my imagination
One of the best explanations of CNN on RUclips. Great work.
Great video! I learned a lot from watching it. I just want to point out that at around 5:10, the convolution is not a dot product of two matrices. It's simply A[0,0] * B[0,0] + A[0,1] * B[0,1] + A[0,2]* B[0,2] + ...+ A[2,0]*B[2,0] + A[2,1]*B[2,1] + A[2,2]*B[2,2]. First number in the bracket represents row number, and the second represents coloum number. The dot product of two 3x3 matrices should be another 3x3 matrix.
Hey MakersAll - Thank you! Yes, you're right. What we're doing is summing the element-wise products of each pair of elements in the matrices.
I elaborate more on this in the corresponding blog in the section "a note about the usage of the dot product":
deeplizard.com/learn/video/YRhxdVk_sIs
Thats the comment i was looking for! for a momment I thought I forgot what a dot product was. Glad to see that im not the only one that noticed that mistake.
Guillermo Rodriguez yeah, i was confused a little bit too
Thank you for this comment! I was worried because I thought my understanding of dot products was incorrect.
In addition to what you mention about the dot product, I would like to point out that what is depicted in the video is not a convolution but actually a correlation. It easier to explain a correlation (because it comes down to a simple element-wise multiplication), but the elements are multiplied differently in a convolution (first element of the image patch is multiplied by the last element of the kernel/filter, second element of the image by second last and so on...)
The excel example was a light bulb moment for me
It was for me too, Sam! I use the same Excel example to illustrate zero padding and max pooling as well, which are tools/techniques commonly implemented in CNNs. If you liked this explanation, then you may be interested in those as well!
Zero Padding: ruclips.net/video/qSTv_m-KFk0/видео.html
Max Pooling: ruclips.net/video/ZjM_XQa5s6s/видео.html
Thank you!!
Credits to fast.ai for the excel example :)
Same here xDDD
Same here - good job!
You just did in less than 10 minutes what my professor could not in 2 hours. I owe you a drink!
i'm a college cs student who is taking a class on ML (and I have final exam tmrw lmao). this video was such a huge help. I learned more from this video than my professor's entire three day lecture on CNNs thank you!!
Ahnaf Khan haha. It’s weird that college professors usually teach things in the hard ways to follow
I watched several videos on CNN but this one is the best video I ever watched on CNN. This video made me understand CNN better than before. Thank you
THIS IS IT.
Best ever playlist for clearing your prolonging doubts. THE explanations in each video are just so enlightening.
so glad I stumbled upon this playlist. ;)
Thanks for such a clear, to-the-point explaination. I like that you directly pointed out the key behind CNN is nothing but layers of filters. Very well done!
I had never seen such a concise yet well-explained video on CNN before. This was really helpful. Thank you!
I like the way you used illustrations to explain the idea behind. It is usually not very easy to imagine all this without the help of visual material. Good work :)
Thanks for the feedback, Eren!
You are Turkish?? 🤢🤮🤮🤮
@@user-yj4qz5lo6k fucking racist..
Thanks! I've been taking baby steps into machine learning and this has been one of the cleanest and most easily understood introductions to convolutional networks that I've ran into.
Out of all the youtube Videos on CNN this one nailed it! Great video keep it up , cheers
I learnt more from this video than from my entire course dealing with neural networks. The visualization is exactly what I was looking for, Thanks!
your teaching is better than most paid courses, thank you.
Thank you!!! This is the best explanation i've found so far.
Mainly because you talk about the mathematics of it with the simplest example. Which is exactly what I was looking for
Usually people say that Convolutional Neural Networks are the hardest of all types of neural networks to learn to use, but you made the basic idea of it clear as the sky.
Tbh you won't know till you try. People say the kanji in Japanese is really hard to learn, but when you try it out in ain't so bad. I guess it depends on the individual to whether it is hard or easy.
The most clearest and simple explanation I have ever seen for neural network and CNN , Thanks
Previously watched another video on CNNs - which was way longer - and by the end, I had no idea. In just 8 mins of your video, it was like a light bulb moment. Fantastic work.
Hi!
We are researchers in human-computer interaction (HCI) looking for people who have taken an
initiative to recently learn Machine Learning on their own, for career, course or curiosity. Seems like you are in that place currently. Would you mind telling us here (www.surveymonkey.ca/r/SelfLearning_ML) about your experiences and any difficulties you faced while self-teaching ML and how you overcame them. There is also a chance to win $50 giftcard.
You can help this project by taking out 5-10 minutes to participate in our study.
For more details, see here: www.surveymonkey.ca/r/SelfLearning_ML
Please share this request with your colleagues or friends who fit this description. People from any major/background may participate. The survey will be open until July 31, 2020.
THANK YOU. 2 hours of excessive reading and researching and your video finally made me get it!
This is BY FAR the best explanation ever given for CNN and makes sense. Thank You.
For other explainers: "If you can't explain it simply, you don't understand it well enough." :P
This was probably the most enlightening video on the subject for me. Your voice is very calming to listen to also :)
Great video not only to understand CNNs, but convolution as a concept in particular. Seems like everyone agrees, that Excel example was on point!
I needed this for my PhD, you're a savior.
Excellent and clearly explained. Thank you for taking the time to make this video! I've been exceedingly frustrated by other content that doesn't make CNNs as intuitive as you did. Thanks again!
I've been watching explanations on CNNs for over a year now and the part after 4:28 is the best i've seen yet,
The excel visualization of images and filter convolving is fantastic thank you very much for this excellent explanation
Highly underrated. Great Video. Liked it before the video even ended!
Even if you've written down what you were saying, not stuttering or mincing words is pretty amazing!!!!
Nice visualization, adequate explanation of complex topic and pleasant voice. I am sticking here.
I found this after a lecture on CNN's. Now it is much clearer. Thank you
same
I feel moved to say some big thing to praise your efforts, but the reality is that I wouldn't be doing computer science if I had a gift for speaking. Big kudos, this video is exactly what I needed to understand convolution.
You are amazing. This video taught me something in 8 minutes something I've had issues learning for 4 hours. Thank you.
I am so thankful to you for making these videos on deep learning in such easy to understand way
Simple and clear, best explanation I've watched on RUclips, thank you
I'm doing a thesis about machine learning. This video is a really great help! Thanks a lot
Have been watching deep learning videos all day since a couple of days, to get the basic knowledge for a big assignment at university (which is due in 4 days :( ). Terribly sleepy, just decided to watch one final video before calling it a night. Decided on this as it was the shortest. Didn't realise it would be this enlightening! Subscribed and look forward to watching your playlists. Thank you!
Love how you explain how memory works in convolution. This help understand cognitive learning. Long-term, short-term filter.
The excel spreadsheet was the key to understanding CNN for me.
Impressive how you managed to answer all the questions I had about CCNs in just one video.
The best explanation of CNN I've ever seen. Thanks a lot!
This is my first learning video on neural networks and you made it so easy to understand. And i come from a hardware background, only knowledge that i had around this was digital signal processing.
Kudos to your explaining skills!
Straight to the point & Excellent illustration . Great work!
Thank you very much for this video! I am very new to CNNs, and watching this video helps me understand how CNNs work in general!
Before I didn't understand anything about how Neural Networks work. Thank you sooooooo much for explaining it in such an amazing way.
Glad to hear that, Pranjal! You're so welcome!
That's the most awesome example I've come across in a long time.
I have seen many explanations -- yours is the best yet -- thank you
mike
You explained filters like no one did.. Thank you very much
This video is amazing, starts slow and builds up on it. The excel sheet visualization was amazing. Subscribed!
That being said, I wish you would continue to explain, what we are getting after the convolution filters, pooling etc.
Thank you, Yousuf! Glad to hear you enjoyed the video!
In regards to what we're getting after the convolutional filters, we have more videos and blogs in this series that explain everything that comes after convolutional layers (like pooling, padding, etc.), as well as how convolutional filters can be visualized.
DL Fundamentals:
deeplizard.com/learn/playlist/PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU
Even more content that pertains to these topics is available in our PyTorch series as well. Pytorch:
deeplizard.com/learn/playlist/PLZbbT5o_s2xrfNyHZsM6ufI0iZENK9xgG
Enjoy!
@@deeplizard Thank you so much! I will check them out.
I'm a software developer and heard about CNN for a while, but didn't know how it works. Right now I kinda understand it. Computer science fascinated me again. Hooray for we human.
Best explanation about Convolutional Networks i ve ever seen. Awesome 👍👍👍👍
Bestes Video zu dem Thema auf ganz RUclips.
this is the one of the best video to explain CNN n the filter examples are great. thanks
Can't thank enough for this simple and beautiful visual explanation
I've been learning CNN from another course but this made me visually realise what I was learning. Illustration like these leave a lasting impression in mind and help in understanding complex concepts in future. Thank You !
Brilliant! If you can describe this to someone like me to even vaguely understand, that is brilliant. Superb teaching.
great explanation.......I was confused about CNNs but this video cleared my doubt.......thanks deeplizard
Great video. The jargon is always the thing that gets on the way for me. Your walkthrough of the visualisation really helped me build intuition about what convolutional layers add to a network.
simple and precise explanation of a complex topic. Thank u very much.
5 years later and this video is still awesome ♥
Great video. Working on advanced projects regarding rcnn. The basics from this video really helped cover weak points!
Simple visual overview of CNN! Great job!
Best CNN explanation ever seen!
Thanks so much for the video! It was always a struggle to understand CNN (cuz ppl explain CNN in a most complicated way😢) but your video really helped in understanding the core of CNN. Keep up the good work 🤗
Briliantly explained! Super clear and esay to understand!
I really like the way you teach us, it's amazing, everything was explained very precisely, thanks a lot mam!
You guys are the bestttt...
I am following the complete playlist and so far all the concepts are so much clear.....
Continue to make video on ML
Love
There are plenty of doubts that i had got on this topic. And this video gave me answers for each and every doubt i got.
thank you very much.
wow.. best explanation ever to explain what the heck convolution is.
This was one of the best explanations on Conv Nets outs there! Thanks!
I keep finding deeplizard gems like this one in your playlists. Thanks.
Oh wow! Thank you this is a real insight into how this actually works, how exciting!
Excellent visualizations to explain exactly what's happening.
Thank You ! It must have been really hard to put the video together but let me tell you, you just did a great job. Thank you !
very well explained, currently preparing my thesis on skin analysis. CNN greatly helping out
That was a very good introduction to the basics of CNN, thanks
I really like the way you explain. No matter if you're new to this topic or just refreshing from a different POV. Subscribed and thanks!
That excel example though, what an excellent way of teaching!!!
Very helpful. The examples used for explanation made things much clearer.
OMG!! The concept was very well explained. Thank you
The concept has so simplistically and brilliantly been explained. Thanks! 😃
You are a rockstar for making this learning material
You are awesome, your way of teaching is wonderful.
I am new to CNN and this video simplifies it. Thank you for the video.
Explained Convolutional neural network very well. Thanks
Amazing. Explaining such a complex concept in such a concise and clear presentation, just amazing. Thank you.
Amazing video. Really one of the best I have seen about the topic till now.
That excel example is genius way to explain things!
This is the first time I get introduced to CNN, and you made it interesting :) . Your way of teaching is very engaging.
Though i have few questions, hope you will reply soon.
Q-1) What kind of filter functions over receptive windows are convolutional layers learning?
(a)Softmax
(b)Sigmoid
(c)tanh
(d)Linear functions followed by any activation
Q-2) Which of the following are true?
(a)CNNs can detect features in any portion of an image (translation invariance) because of strides and maxpooling
(b)CNNs lose information on spatial location of features because of max pooling.
(c)CNNs with convolutional layers followed by fully connected layers can process images of any size
(d)All of the above
(e)None of the above
Q-3) In a neural network consisting of max pooling layer, what is the multiplication factor for a gradient passing through a neuron which is non-maximal in the pooling window? *
(a)-1
(b)0
(c)1
(d)Output of maximal neuron
Q-4)Given an all-white image to a well trained CNN based image classification model, are the output probabilities for each class the same? *
(a)Yes
(b)No
(c)Can’t say
Q-5)Which of the following patterns can CNNs learn?
(a)Spatial patterns in 2-D
(b)Spatial patterns in 3-D
(c)N-gram based patterns in text
(d)All of the above
(e)None of the above
Q-6) Which of the following layers are useful for up-sampling upto a pixel level, features extracted using convolution in a CNN?
(a)Fully connected layers
(b)Convolution layers
(c)Transposed convolution layers
(d)Softmax layers
(e)All of the above
Q-7)Larger receptive field in a CNN results in_
(a)Better model performance
(b)Larger number of model parameters
(c)Both
(d)Neither
Excellent explanation! So glad for all the free learning content!
Man, what a way to explain :) You should do this with more Neural Network topics.
Well done, really! This video is quite helpful when it comes to understanding what the CNN does in practice. Thanks!
best explanation for CNN. Thank DeepLizard for great work!
That was amazing, absolutely the clearest CNN explanation I have ever seen, make more videos please :) and thank you so much
Deeplearning is really amazing I might get hooked onto this😍
Please point it out if there are wrong points
- CNN has Convolutional Layers as hidden layers
- Filter can be a pattern detector.
- You can set multiple filters for each conv layers, and randomly assign values for each filter.
- At the Conv layer, the filter of the layer slide over the entire inputs. (If a filter is 3x3 table, the dot product will be computed with 3x3 table of input matrix)
- As the layers be deeper, the filters can become more sophisticated and detect much more complex patterns.
OMG CNN is complicated tbh, but you gave a very simple explanation to it! THANK YOU SO MUCH!
AMAZING EXPLANATION!!!! Very well done!!!! I am super impressed!!!
Liked it. Thanks for brief explanation of CNN. I was worried that it is much more complex but you made my life simpler
woew awesom explanation. You deserve more subscribers