How Convolutional Neural Networks work
ฝัง
- เผยแพร่เมื่อ 2 มิ.ย. 2024
- Part of the End-to-End Machine Learning School Course 193, How Neural Networks Work at e2eml.school/193
A gentle guided tour of Convolutional Neural Networks. Come lift the curtain and see how the magic is done. For slides and text, check out the accompanying blog post: brohrer.github.io/how_convolut...
Check out • How Convolutional Neur... for better audio and a more detailed account.
Follow me for announcements: / _brohrer_ - วิทยาศาสตร์และเทคโนโลยี
i've watched a good dozen of videos about convolutional neural networks but you've nailed it!!!
thx for the insights
Thanks Ivan :) I'm very happy to hear it.
Really nice presentation for sure. Just one recommendation: please normalize your audio. It's super easy to do and makes people feel much more comfortable as they don't have to crank up the volume where any tiny notification will blow your eardrums to bits.
I also feel like you could easily talk way faster, although adjusting playback speed manually does the trick just fine.
Thanks Min, and I appreciate the feedback. I've actually been speeding up my delivery just a bit in subsequent presentations. I was trying to keep the pace moderate for non-native English speakers, but went overboard. I like the audio normalization tip. I'll make sure to do that next time.
please re-upload this one with the audio made louder. btw, normalizing might not make it loud enough, feel free to make it louder by hand - the peaks will just be cut off, the speech will still sound good most likely. for best results use an audio dynamics compressor.
pace is no big deal, and it might even be better to err on the slow side. people who want it faster can watch at 1.25X or 1.5X speed, but 0.5x speed gets a little bit wonky. gotta love youtube.
Finally, someone who explained it in details, not some bunch of "CNN takes image, makes some stuff and poof it can classify" crap. Thank you, you are the hero wee need, not the hero we deserve.
Thanks Adam! Yours is officially the best compliment ever, narrowly beating out "The Mr. Rogers of Data Science." img0.etsystatic.com/065/1/9143987/il_570xN.761988760_8vsd.jpg
While my comment got your attention I will slip in an extra question for you, it's very clear for me how CNN can classify, but what about image processing? Can it do that (on it's own)? For example like noise removal or image colorization. It's like we would need to skip pooling, but then wouldn't we lose global information that we would need?
Yes. CNN explained in details, step by step. I appreciate it very much!
The example with X and O was a brilliant approach. This is where most machine learning videos miss out.
Hands down one of the greatest educators of the 21st century.
I have a report on neural networks due 3 days into the semester so this helps and makes me not want to die as much as before.
what are you studying?
Damn, dude. Didn't expect to see you here!
What was your report on? Melon classification using NN?
Cyranek meme god
Made my end project in 3 days
Best visualization of convolution I've seen.
Clear description of what's happening behind the scene. Thanks for the great tutorial!
Every slide that is presented here opens up a new dimension of understanding about the CNNs. Awesome job!!!
Great content! Amazing visualization for all the maths! I finally understood what Convolutional Neural Networks are all about. Even after a whole course on ML at uni, I still couldn't get it, but a mere 26 minutes of your video did the wonder. Keep it up! :)
Don't be scared off by the length of this video. Excellent use of simple examples to illustrate core concepts. The author has a finely tuned sense of what to focus on and what to gloss over to impart the gist. Many explanations either get lost in (easily forgotten) jargon or muddied by too many details too fast. The overview of the limitations and application rules of thumb is especially valuable.
Great stuff.
It honestly isn't lengthy for the content that is inside! People can search for years and struggle to find something that they can understand on this topics, and when they do, it takes a very long time to learn. This is only half an hour, shorter than a class that you would take at a university, but it is so worth it!
I would say the clearest introduction to the subject I have ever watched, thank you.
Clear, concise, and very well organized. Props.
By far, the best explanation over the web.
So for a while I had a basic understanding of this algorithm, but I would never have felt comfortable trying to make one on my own. Watching this video felt like all of the pieces were lining up together into a beautiful picture of how this algorithm works. This was really well done.
Thanks +askmiller. You made my day.
@@BrandonRohrer I'm trying to implement my own algorithm as well and I coudn't understand how would I make the algorithm decide which features are relevant. In your explanation, it seams to me that you choose them and not the algorithm itself
I had bookmarked this post, came back after 2 years and found I understand it so much better. Wish I could give it a second thumbs up. Thanks!
I'm glad to hear it :) Welcome back.
You took a subject many others struggle to explain (also here in youtube) and what you made is art.. the are of simplification.
Many thanks buddy!
Please make more videos like this. You helped me understand concept which I was learning for months in just 26 min.... Thanks Sir..
Best Video I ever seen about neural network....thanks for uploading Brandon
I watched series of videos and I was trying to find one that covers the core concepts without going into details and I could not. Finally, I came up with this video. I am sure you knew what other videos are missing and you have covered all here ! Good job
the most comprehensive guided tour ever done about convolutional Neural Networks, and within less than 30min. Simply brilliant !
Thanks olivier! It's also a huge honor to get a nod of approval from Ted.
Best video till now ... which I saw for CNN ... Great..
very good video, finally someone explained it clearly. Thank you !
Thank you Wang :) I'm happy it was clear.
This Guy Deserves a Noble Prize for Teaching!
Best explanation for Convolutional Neural Networks ever seen
I'm a second year engineer from Germany and currently reading "Python Machine Learning" from Sebstian Raschka (newbie to programming as well!). Your videos clearly put all the pieces together. Thank you for being a great teacher :)
Thank you so much for this video, I have an exam in a couple of days and I have been so lost with all these terms but your video is so incredibly clear. More than the recorded lectures of my professor and books I have read. Thank you for taking the time to explain in a way that makes someone with very little knowledge on the topic feel like an expert!
Best video I've found on CNNs. Thanks for taking the time to make this topic more accessible!
beatifull video man, i see you spend a lot of time making it. I enjoyed it very much, thank you :D
The best and the most detailed explanation I have ever heard. good to know you didn't rush, through the convolution, pooling or anything else! The example was great and easy to understand. Thumbs up!!!!
I'm happy it was clear Sahil. That is my number one goal.
The flow of your explanation was neat and point to point thanks a ton!
One of the better videos explaining Convolutional Neural Networks! Thanks!
Brandon, this is really good. Thank you very much!
ty .. this answers some of my questions about cnns regarding pattern recognition (the filtering steps)..
amazing description! Thank you so much Brandon.
You are a lifesaver!!!
I have to elaborate a paper on CNNs for uni (ImageNet Classification with Deep Convolutional Neural Networks) and it really got me confused with all those new terms and general complex language (engl. is not my first language).
Thise video helped me even more than I hoped! Thank you!!
You make complicated things look quite easy to understand
I appreciate it Youssef
Thats really true
I don't know why but I teared up after understanding something so clearly. Like just how beautiful the intuition is, and I feel blessed to have something so amazing as my career, while half of the world is stuck in dull jobs they hate. You are a hero.
Sanket Shah, your comment made me tear up too. I'm very happy the video helped.
This was --hands down-- the best intro video on the topic I have seen yet. Great work. Was a fan of your videos at Microsoft
Beautifully explained, with the perfect flow right from the beginning !!!
Seriously my graduate level class on Image Processing has spent two lectures explaining this and only now it finally clicks!
I'm glad to hear it Joel! Hang in there.
A saying "Today's scientists have substituted mathematics for experiments, and they wander off through equation after equation, and eventually build a structure which has no relation to reality" by Tesla.... it is really true in the case of CNN concepts. Brandon.. you have structured the concepts of CNN so simple and easy to understand in this tutorial. Best video on CNN so far. Looking fwd for more video sessions from you !!!
Thanks for the wisdom Kiran. Tesla is a hero of mine.
actually it has a "relation to reality" as it implements our object classification in our brain.
the convolutions' output "feature maps" are bigger and bigger features - like lines, curves, then bigger and bigger parts of the object - we recognize and that's how we classify the objects too.
pooling helps us recognize them invariant of angular etc. differences, like if I hold up a shoe for you sideways, you'll still know it's a shoe
This is the best and most clear explanation of back propagation and how neural networks works. Thank you for such excellent work.
This is the cleanest explanation of CNN I have never watched on TH-cam! Thanks for sharing!
Great introduction to ConvNets! Thanks a lot.
Please also make similar videos about RNN, LSTM, etc.
Also please if possible do a series on Gradient descent algo's such as momentum, AdaGrad, Adam etc. with mathematics
+1
1) You're cute
2) The graphics are great
3) The video was very informative
4) The sound had a bit of echo
Tres Kewlshoes first part is disrespectful and inappropriate.
@@adamboris7925 How so?
I wish people would say I am cute. Don't you?
Plus I don't think Tres is looking for a partner over youtube lol. I think it was meant more for comedic relief after watching a 30 minute video on machine learning.
Randy Diaz cute sounds gay
@@ChessCat1500 you're gay to find it gay
This is the best tutorial in this topic i've seen so far. Thanks for sharing!
This was better than any Stanford or Harvard lecture on this topic. Well done! Thank you!
Wow, thanks Gabe! I'm glowing.
This video saved my semester.
What are you studying?
Very useful and clear presentation
same
When you used the filter/kernel, why did you average the results?? Everywhere else I've looked there's only the sum
EXCELLENT explanation without maths of the how. Great job.
This is seriously the best video on CNNs!!! I cannot believe the skills you have to make this soo easy to understand!! Thank you so much, everything is so clear to me now : )
You nailed it, best video in CNN so far, at lest for a ML newbie like me.
beautiful explanation! 10/10
Well thank you very much!
This is the best that I heard on convolutional NN. It is easy to describe difficult things in difficult ways. Much harder to do the same in simple words. You have nailed it.
This is one of the best videos I have watched for CNN. I really appreciate your work. Thank you for this ART.
this is the best presentation of cnns i've seen so far. good job.
since it was so good, i wish you explained backpropagation a little bit more because this is something i still dont know how it works in detail. there can be so many layers and weights, how you know which one you have to adjust?
I was thinking the same. How backpropagation, or any other learning algo, works with this?
Thanks Windar, I'm glad it was helpful! And thanks for the +1 on backpropagation. I've added "How Backprop Works" to my to-do list. The really short version is that every weight in every layer gets adjusted a little bit every time. The slightly longer version is that it helps if you randomly select half the weights in every layer to adjust and pick a new set each time. This is called "dropout" and it helps avoid overfitting.
wow, even in 1-2 phrases you can explain more than other guys in 1 hour presentations.
"How Backprop Works" from you would be nice, but dont do it just because of me. i subscribed anyway!
You compute the derivative of each unit from back to front. The derivative is really a slope. The slope then tells you what direction (+ or -) you should adjust the weight in order to improve the output of this particular unit. Do this across the whole network and you've improved the output of the whole network. Check out karpathy's blog: karpathy.github.io/neuralnets/
I appreciate, professor^^ from S.korea
chang yong kang 여기서 뵙네요...하핳😄
@@korean4130 님 저도 치과 진단에 도움이 될까 해서 관심을 ㅎㅎ^^
m.blog.naver.com/nkabcd2/221637820902
Can you create brain-computer interface?
Hello from 청주시 haha
@@austinbecton5341 ^^ welcome ^^ have a good day ^^
THE BEST introduction to convolutional neural networks for BEGINNERS!!! Many many thanks!
The illustration images are totally awesome!
It's inaudible :(
The volume is too low, I can't even hear you!
That's the best and concise video I have seen so far. Thanks!
You sir, are the best minimalist teacher I have found on the internet. For a beginner you are a blessing!!
When I'm looking at pornography, my brain votes 1.00 that this is pornography. ;)
Awesome video on convolution neural networks. Thanks so much Brandon, I was having difficulty to understand the layers but you made it crystal clear.
clear crystal presentation - amongst the best video on CNN I have seen. Thanks!
Astronomical explanation of several concepts and how they work together! Wow!
Brandon,
Man, I have watched numerous videos but nobody explains it as cleary as you do. I beleive any in the IT world, watching your video will understand. Good job making it dumbproof and so within reach.
These 26 minute video clarified a concept that days of research elsewhere hasn't been able to so concisely and intuitively, so it helps me understand rather than just be able to "replicate" a model. Thanks Brandon, this was brilliant!
Concise and understandable explanation, with great rule of thumb at the back. Awesome video!
Fantastic presentation! You provided exactly the information I was looking for along with exceptionally clear explanations. Thank you!!
One of the best explanations I have found..hats off
Man your concept is Crystal Clear !!
I am new to neural networks and found this one of the clearest tutorials ever. Very good.
Your video is the best illustration of CNN I found so far
This is by far the best video about neural networks I've ever seen. Simply awesome! :)
Haven't even finished watching this, just to comment. It's BRILLIANT! Very well explained. Thank you
This is the simplest video i 've seen on CNN, by simplest I mean explaining all the concepts in a simpler manner. Thaks a ton .
Definitely the best video on the subject. Thank you for spending the time making it.
26+ min video but so dense with information, well done sir!
Thank you so much for explaining neural networks in such a easy way. I was really confused when I saw the other videos and people mentioning about gradient descent and back propagation.But the concepts are now clear. Thanks again for this awesome video.
This is one of the best explained videos I've ever watched. Subscribed.
great speed, well chosen words and figures. This is a perfect explanation video
Great explanation Mr. Rohrer. Appreciated.
With just first 9 minutes of video, I was so sure I will understand the rest of the video. Thank you!
Talk about straight forward. Yes! I have been struggling with all the terminology and basic language in this science to find out that its not that hard to understand at all. THANK YOU.
Great introduction to ConvNet! Complicated stuff is clearly explained. Thanks a lot!
The best videos explaining CNNs i've seen. Very good slides also. Thanks for sharing the PPT and video!
Great illustration! Thank you!
Very good for startups and very straight forward. This one deserves my like! Regards from the Philippines.
Great video. I feel like this is one of the most comprehensive especially on how it explains the process step by step (for instance i noticed other explanations don't bother with how stacking layers works)
Have to say this is one of the best videos on the topic, making the key ideas easy to understand.
Thank you for this awesome video. I've spent more than 2 months trying to get ConvNets with no luck, but now I do!!!
That's awesome Yaison. I'm happy to help!
I am glad that I understood CNN finally with the help of this video, You are exceptionally great tutor. Thank you so much.
This is really really really awesone, such a simple explanation any one can follow. I watched many other videos got bits and pieces. This video helps me to put all of them togethor in a much much meaningful way.
Thanks a Ton.
Sir, many people has already said that your video is the best, like many other videos about CNN get the same comments. But with me, i think your video is the best. Your voice, your slide, the video length,... is perfect. Thanks you so much sir.
That was an excellent in depth explanation.
this is the best TH-cam video on the subject so far
Awesome presentation-beautifully demystified the complex subject