Convolutional Networks, but explained for humans

Рет қаралды 7,721

Күн бұрын

I've recently found it really beneficial to start reading research papers. Highly recommend checking out AlexNet and YOLO if you're interested.

Пікірлер: 23

@ACken2 10 ай бұрын

This is probably one of the easiest to understand video on CNN. Surprised it has less than 100 views.

@NITILPODDAR-lt1oq 9 ай бұрын

I am at the foundational level in data science. This video was something. It went over my head for now but I think I will get to it soon. Currently I am studying advanced probability as part of my course.

@addmoreice 10 ай бұрын

The location information is still there, it isn't lost. It's just that it's encoded in the order of the nodes rather than as information directly being input into the system. Worse, this is now information which is locked to the specific size/location of *this specific image resolution*. Convolutional networks allow this 'meta' information to be retained and provides it to the network, but 'costs' extra layers to make it invariant across multiple images. A well worth it trade off since some of that information will be useful for further layers for feature detection anyway.

@brendawilliams8062 9 ай бұрын

Similar to 12.3::: minus .123::::: to you hit the right string.

@sharmaanuj334 10 ай бұрын

Great video! couldn't believe after watching that it only had 10 views, your content is strong - keep progressing and helping the community

@tylerbakeman 9 ай бұрын

You make it sound simple, but we’ve skipped over the details to actually help us solve this very tricky problem. Images and neural networks are definitely the fad today, but it’s easier to learn how to construct a neural network with a model that has very few input nodes. I’d be down for a more in depth description of how you take the simple model (building a model that would work for an image, being able to recognize patterns). Recognize patterns is hard though: You have to know what 2D geometry you’re looking for in an image, probably apply some pixel operations on the image (like greyscale), and then run it through the network. You could identify lines, circles, … etc … I don’t have much experience in AI, but I assume you would check to see if those lines resemble the pattern to object wireframes (ie Facevectors for facial recognition) but that probably isn’t true. So. Hopefully you’ll teach me that someday.

@alexgonzo5508 9 ай бұрын

Awesome explanation; hoping you continue making these type of ML explanation videos. Thanks... subscribed.

@LookingGlassUniverse 10 ай бұрын

This is such a helpful video, thank you!

@MrDesillu 9 ай бұрын

0:39 "and initialize it with random weights" i am already lost here, so i guess i am not human. You made it like people know already NN, we don't. But i will come back to it later; it interests me. I like anyway for encouraging you to do videos

@sanchogodinho 10 ай бұрын

Thank you so much!!!!!

@carrumar 10 ай бұрын

Great and concise intro, thanks

@jorgeromero4680 10 ай бұрын

That's great I would like More of this if that's possible. Thanks

@arpy8 9 ай бұрын

really cool!

@siddeshanm1642 10 ай бұрын

GREAT EXPLAINATION

@taratara2509 9 ай бұрын

Спасибо за видео! Было действительно очень полезно!

@weeb3277 9 ай бұрын

nice video man, but the text fading in and out feels a bit odd.

@mamystrotv4745 9 ай бұрын

Great content❤️ Please, can you tell me what tools did you use to make this wonderful presentation?

@jacobpradels 9 ай бұрын

Thank you! I used Microsoft Powerpoint

@taneliharkonen2463 9 ай бұрын

Awesome dip your toes intro! For people that are just a little more interested, here is an awesome video from computerphile where Mike Pound digs just a bit deeper still, but keeping it still conversational: kzfaq.info/get/bejne/pt9lldyAtK6-fWw.html There is way more videos on relating topics. The video about image convolutions is super interesting and makes Convolutional Neural Networks "CNNs" make more sense.