Which class exactly did you learn about neural networks? Did you also learn multi-variable calculus (fundamental to even the simplest neural network) in your high school class? I would love to attend!@@randomguy4738
Most educational videos give viewers the impression that they are learning something, while in reality, they cannot reliably explain any of the important points of the video later, so they haven't really learned anything. But your videos give me the impression that I haven't learned anything, because all the points you make are sort of obvious in isolation, while in reality, after watching them I find myself much better able to explain some of the concepts in simple, accurate terms. I hope more channels follow this pattern of excellent conceptual learning.
Being able to explain it at a conceptual level isn't good enough. You can only understand it by practicing (i.e., build neural nets by yourself and play with it)
busi magen, I don't have a story nearly as touching as that of you and your Grandmother, but I think I would cite my dad as teaching this by example when I was growing up, in that the way that he would describe things centered on what they're actually doing in simple terms, rather than on learning the appropriate jargon.
Particle accelerator is used for creating/verifying hypothesis. Your analogy is terrible. Regarding learning a new skill, one needs to practice rather than just passively absorb information. This is why homework exists. Regarding neural nets, anyone think they can "explain" NN after watching this video is frankly laughable. (not saying the content of this video is bad)
I study mathematics, physics and architecture. By definition this man is an ORACLE in the strict meaning of the word. With all honesty I never imagined someone explaining complex topics with the dexterity this man has. He is literally an institution and an outstanding teacher. The computer graphics and the illustrations are simply perplexing. This guy never evades complexity. He never evades complex arguments. He illustrate the complexity and dive into the exhaustive explanation of the details. It's extremely rare to see a professor and a dedicated user to put a lot of effort explaining, animating and describing mathematics the way he does.
@@viharcontractor1679 When did I say that? Please read my comment again. I have no issues with the tutorial, I have objection on the comment to which I have replied. One should always make an appropriate comments. As it is incorrect to say something rude, it is always wrong to do false praising. Have you read the comment? of kummer45? Calling the tutor of the video as " Oracle"? Really? This kind of words should be used for someone like Swami Vivekanda and not for some ordinary tutorial. It almost hurts to see such misuse of words.
@@sdc8591 Come on man, what is wrong with / about that comment? The video is fantastic in every way, It is dense enough that I've had to watch it several times over, yet is able to communicate the concept of a neural network in such a way that even my pea brain can grasp this topic, please think before commenting and make a proper comparison.
@@JCake First of all, don't use the word 'man' , I am a girl. I never said video is bad.It is fine. Why everyone is coming over here and defending the video? Is is so difficult to understand what I am saying? The comment from kummer 45 is an exaggeration and I stick to it because it is. If the video is good enough , one does not need to watch it second time to understand the concept. I had seen one video by Mathew Renze on the same topic. That long tutorial, was first time I came across neural network. It was more that 1.30 hours of series of videos. I never watched it again and still remember every single concept. Now if this man is oracle what you will call him?
secret scientist are much further than this, like they can wirelessly from satellite I think give dream images and change dreams you make yourself... for me they always try to make it ugly,...
It's a statement I don't agree with. At university, we are taught things in a formal and abstract way, not just for the sake of overcomplicating things. I don't think professors, which are primarily researchers should be considered "fools" because they fail to teach their subject in a more intuitive manner.
from what i understand (i am also an dummy i just tell you what i think) the inputs are the pixels the weights are the pixels 's whiteness or blackness it is like lets say we need first pixel to be white so we need the computer to know there is a pixel there (hence it's an input) we need the computer to change how white or black it is (hence the computer's ability to change weights)
This is how you taught Deep Learning, people. I've seen lectures that either be categorized into 2 groups: too hard or too shallow/general. You have balanced between them. Thanks you so much!
@@randompersondfgb yes, you're not understanding. Uncommon Sense is saying "you" nailed it. The you is Cliff Rosen, the original commenter. He's saying that Cliff Rosen nailed it when he wrote the comment "Brilliantly explained".
this is my first introduction to machine learning and I watched this only twice to get it, really goes to show how good of a teacher this guy is, the effort he puts in is nothing short of amazing !
I'm currently taking a computer science math course where the professor strongly advised everyone to watch this exact video series to get an intuition about what all the math is actually used for.
not only is your content extremely helpful in terms of learning complex concepts, but your thoughtful creation of every element - from the overall progression of ideas down to the smallest jokes or teaching points embedded in graphics - is deeply inspiring. Thank you, 3b1b
Around 2 years ago I was a sophomore statistics student and had no idea what deep learning is, until I met this video and 3b1b channel. His clear explanation of neural network and animations blew my mind. Since then I started my journey in machine learning. For a random reason I clicked onto this video again, and realized how long my journey in this field have been. This video really changed my life and I am really grateful about it.
@@yashrathi6862 The linear algebra series that was recommended in the video is a good start, other than that you should keep watching this video and you will start to understand it better the more you do. I am also in class 11 and that is what helped me
One year ago I met this video. I couldn't understand any single word in it. A year later, I am back and I still cannot understand it. I am fucking stupid.......
@@yashrathi6862 To be honest there are no real "prerequisites" for learning neural networks, in the end it just gets down to how familiar you are with the concepts of basic graph theory. However, I admit that it can be pretty overwhelming for someone to try and comprehend all the stuff at once, which is why being savvy with the use of linear algebra is a must. Apart from that you should try your hand at programming once, perhaps the algorithmic mode of thinking would help you deveop an intuition for neural networks. And yes, of course try to explore graph theory, for neural networks will resonate much better with you once you do, imo.
I don't understand notifications either. What, we're supposed to do things other than watch all the remaining 3b1b videos we haven't yet seen between notifications? Who would be so wasteful with their lives???
This is the 80s generation we were listening rock music and looking how to get things done better we grew without mobile phones just sitting front of a computer or playing basketball outside in the park. We grew without rap, hip-hop, either thinking that the gang is a cool guy! this is what now generations require badly!
And our generation is unlucky that we had no such mentors and internet to deliver their videos. Taking this into account, we demand results, youngsters! We had, at least, an excuse for being dumb :)
@@omarz5009 The main downside of python is the fact it's a high-level language and hence kinda slow. But for ML and NN it has several powerful libraries (pandas, numpy, tensorflow) which make up for that. Given Python supports the implementation of C-Code, those libraries could be optimized like heck to the point bothering with the stuff in C++ is just wasted time. Plus Python is much easier to learn, hence more people use it and develope for it.
This video kickstarted my journey in ML a year back. Trust me, back then I watched this video three times to finally understand. It might be challenging for few to get it but when you get it, it just feels amazing
I am just astounded. I spent so much time trying to understand this concept. Everywhere I looked people would show the similar neural network animation, but no one ever really explained and exemplified every single step, layer, term and mathematics behind it. The video is really well structured and with amazing animations. Extremely well done. My mind is so blown I can barely write this comment.
🎯 Key Takeaways for quick navigation: 00:04 🧠 *Introduction to the recognition of digits* - Your brain's ability to effortlessly recognize digits, - Challenges in programming a computer to recognize digits, - Introduction to the relevance and importance of neural networks in machine learning. 01:06 🤖 *Devoting the video to the structure of neural networks* - Creating a neural network for recognizing handwritten digits, - Exploring the simplest form of neural network with no added complexity, - The network's ability to recognize handwritten digits. 02:37 🧩 *The inspiration and components of neural networks* - Explanation of neurons in neural networks, - Neurons as holders of numbers between 0 and 1, - Introduction to layers in a neural network. 03:36 🧠 *How neurons and layers are connected* - Description of connections between neurons and layers, - Activation values and grayscale representation in the input layer, - Exploring the concept of hidden layers and their purpose. 05:06 📊 *Understanding neural network learning and training* - How a trained neural network recognizes digits, - The role of activations and weighted sums in making predictions, - The complexity of learning involving thousands of weights and biases. 07:31 ⚙️ *The role of weights, biases, and patterns in neurons* - Explaining the significance of weights and biases in neurons, - How weights contribute to recognizing specific pixel patterns, - The role of biases in determining neuron activation thresholds. 10:32 🧮 *Compact notation for neural network connections* - Representing neural network connections using matrices and vectors, - Simplifying the expression for transitioning activations between layers, - Highlighting the importance of linear algebra in machine learning. 13:11 🧩 *Neurons as functions and the complexity of networks* - Viewing neurons as functions that take inputs from the previous layer, - Recognizing the network as a highly complex mathematical function, - Discussing the challenges of setting weights and biases manually. 16:20 🔄 *Learning and the future of the neural network* - The promise of neural networks in recognizing complex patterns, - How neural networks learn from data and adapt their parameters, - A brief mention of ReLU activation functions and their advantages. Made with HARPA AI
@alexandrebour7494 Chess and Go are not complicated in terms of rules, especially Go, but indeed quite complex to master. You can ask AI to "invent" a game today, and it will try to mimic existing games and mix and match board-game elements e.g. deck building, work placement, etc. Today, you could build an agent that writes game rules, the other that writes code for the game sandbox, another one that tries to master it, and yet another one that "evaluates" how good the game is. If you run this agent chain over a sufficient computing power, you may get something very interesting
@@HarpaAI I don’t have a computer that can handle that computing power. On the other hand, I have no doubt that advanced technology that can do this work more easily can happen in the near future. This could help me improve the complexity of a pre-existing game that I particularly like, the Hnefatafl.
Finally, a video that does more than just present some neurons and layers and say, “here’s an activation function.” Your video describes how the model is developed and why the algorithmic approach is appropriate for the problems neural networks try to solve. Thanks!
3Blue1Brown "Sigmoid Squishification Function": 11:23 Most brilliantly named function I have ever heard named. Absolutely brilliant. The merger of the technical with the simple with a double alliteration for easy memory.
i took a deeep learning lecture in my last semester and my professor couldnt explain in 4 frickin months what u explained in 20 mins much much appreciated man you're doing awesome work hope to learn a lot from you
I am a Data Scientist and I would like to tell you THANKS. I have NEVER met anyone with the ability to teach complex things in this way. A M A Z I N G. Please continue like this, for example with other statistics videos. You can substitute many of the University courses.
@@duykhanh7746 A bit late but if your question hasn't been answered yet: It doesn't really matter if you have a value >1. Basically anything above 0 is an activation and you can also view it as the size of "a" being the intensity of the activation. Biological neurons can also be more active by firing in fast succession (up until they reach the maximum possible firing rate of like 250-1000Hz depending on the source), but you don't want to introduce things like loops in artificial neurons to not slow down your network. So to simulate this kind of behavior, you just let your output get bigger. You can compensate for the lack of an upper limit in the following neurons by adjusting the weights and the biases. TL;DR: No. :D
how is it possible that I can lie in my bed on a Sunday and am presented with mind-boggling cutting edge knowledge told by an incredibly soothing voice in a world class manner on a 2K screen of a pocket supercomputer basically for free
@@Charge11 And software engineering advancements, thousands of years of intellectual history, biological evolution of conscious brains and so forth. point is, it's miraculous if you step back far enough.
This is the best intro to neural networks I have ever seen. The presentation is excellent! The animations are very very very helpful especially in understanding the formulas and matrices and how they came to be. Thanks a million. Looking forward for the next one.
Written some notes from the video to read quickly. Hope it helps somebody. l Neural Networks can recognize hand written digits, letters, words ( in general, tokens ) l What are Neurons? ○ Something that holds a number [ 0, 1] ○ The higher the number, the higher the "activation" rate l Consider a 28*28 table in which each unit is represented by a value between 0 to 1 ( activation number ) ○ Let us divide each row into a "layer", such that, if we were to divide all the layers, the last layer would contain 10 "cells" ( units ). ○ Values are passed from the previous cells to the last layer ( 10 unit layer ), again, between 0 and 1. The higher or closer the value is to 1, the more probability exists that the image scanned represents that unit cell. So, a unit cell that contains the highest value is indication that the index of the unit cell is the value of the image scanned. ○ 16 cells in the second and third last cells are arbitrary. ○ Each cell is linked ( causes activation ) to some ( not all ) other cells in the next layer which further cause more activation. ○ Each 'cell' corresponds to some sort of identification about how much a certain region 'lights up', and then sends a value to another node which reacts based on the received value. ○ To find whether a certain cell with light us, like each cell be represented by 'a Cell_Number ', and let each cell be 'assigned' a certain weight 'w'. The sum of all the products of each cells 'a' and 'w' will be: w1*a1 + w2*a2 + w3*a3 + w4*a4 + … + wn*an ○ Let these weighted sums represent some 'grid cell'. Each cell is either 'on' or 'off' with respect to being positive or negative. In this case, 'green' represents on, and 'red' represents off. ○ Let us concern ourselves to a certain region where the cells are mostly on. Ergo, we would be basically summing up the weightages of those grid cells. ○ Then, if you suppose a region where there are brighter grid cells in some part which are surrounded by dark grid cells, then that area is the main edge we're looking for. ○ Of course the sum of weightages gives us very different value. In order to 'squish' that number line into 0 and 1 , we use the function: Sigma(x) = 1/(1 + e^-x) Which is a sigmoid function or a Logistic Curve. Our equation now becomes: Sigmoid(w1*a1 + w2*a2 + w3*a3 + w4*a4 + … + wn*an) ○ But what if you don't always want to light up when it's a positive value, and rather want it to light up when the weighted sum of that grid cell full fills some condition, such as > 10. This is called 'Bias For Inactivity'. Using this example, our equation becomes, Sigmoid(w1*a1 + w2*a2 + w3*a3 + w4*a4 + … + wn*an - 10) Here, 10 is the "bias". ○ The possibilities of the different knobs and dials open us to the term of "Learning", which just means to find the correct relation of values which perform the expected behavior. ○ The complete expression above can be adjusted in the formula: a(1) = Sigma(W*a(0) + b ) ( (1) and (0) are superscript here ) Where W = k*n matrix whose elements are weights corresponding to a cell. a(0) = n*1 matrix whose elements are the 'a' of each cell. b= n*1 matrix whose elements are the biases of each cell ○ NOTE: Sigmoid function is not used very often now, instead it is replaced by ReLU ( Rectified Linear Unity ), which is defined as: ReLU(a) = max(0, a), a linear function where f(a) = a for a>= 0, which for a < 0, f(a) = 0.
I know you read this all the time, but I must say it. You videos are simply incredible! Your work reshapes education. You deserve every cent that this platform puts in your pocket.
Behind this material is an extreme shot of giftedness. Explaining something is not easy. You first need a solid physical model for the topic in your brain and then you need to translate this model into a mental model that can be faithfully exported into others' brains. I congratulate you for this excellent job and I hope that you appreciate what you are and what you are doing. This is much more important than how much money this business brings.
watching this for a second time and i can't believe how illuminating is to come back to the basics and get a renewed understanding -- grant, you're a treasure
This is the first time I'm commenting on a RUclips video and honestly, I'm so thankful people like you exist! I wish only the best for you in whatever you do!
I totally agree, my friend. Today is a very important day in the history of youtube mathematics. And since I am the 100th person who liked your comment, I would like to give a little inspirational speech: To all mathematicians, physicists, engineers, computer scientists or people who want to become one of those in the future, today is a very important day. The best youtube mathematician, 3Blue1Brown, has made a video about neural networks and plans to make others about it in the future. I think it's not necessary to explain the inherent significance this topic has concerning the future of our technology and our understanding of the universe and the processes going on in it. These videos will help the new scientific generations to cope with the structures still to be found and to bring on a new and deeper understanding of the things that have been found and examinated before. Humanity is reaching a point, where the wish to understand the world is higher than it has ever been before. You, dear future scientists, can all be a part of the progress we are just going through, you just have to have the Will and the Strength for it, never give up if things aren't working properly or as you expected and always remember: At the end, everything will be fine, so if it isn't fine, it's not the end. Actually, I have reached the end of my little inspirational speech (and it is fine ;) ), and to complement it well, I want to quote a famous poem which plays an important role in a very good and famous science fiction movie.... "Do not go gentle into that good night, Old age should burn and rave at close of day; Rage, rage against the dying of the light. Though wise men at their end know dark is right, Because their words had forked no lightning they Do not go gentle into that good night. Good men, the last wave by, crying how bright Their frail deeds might have danced in a green bay, Rage, rage against the dying of the light." Thank you.
One “like” is not enough for the work that has gone into making one such video. This video should be part of the curriculum and he should get the royalty for this. Awesome work!
I'm currently taking a computer science math course where the professor strongly advised everyone to watch this exact video series to get an intuition about what all the math is actually used for.
My goodness, I’ve watched nearly 20 videos on neural networks, and none of them come close to this one in terms of visual representation and clarity. Thank you very much.
I am not really from a math background but I am hugely interested in programming, and I must say this video has made it easy for me to understand the math behind neural networks! I loved it , thank you!!!
This channel is so damn good. Other channels give some terrible analogies and some other explain it in extreme technical detail. This strikes the perfect balance and provides a foundation to understand the more technical details
Another reason to be mentioned on why ReLU is used instead of Sigmoid is simply the fact that it calculates a lot simpler (obviously cutting negative values vs. exponential operations). Plus another important issue of the σ function is it's gradient which is always below .25. Since modern networks tend to have multiple layers and because multiplying multiple values < 1 quickly becom really small (vanish) networks with a larger number of layers won't train when using Sigmoid. And as always, amazing video, animation and explaination!
This channel and the visualizations it produces to teach subjects like this one is the best advance in the history of communicating mathematical ideas. It's extraordinarily inspiring that one person can have such a large impact on the world today (and for generations to come). Thank you, Grant Sanderson.
To think that someone would make a video of neural network and explain it in a way so simple yet insightful is such a bless especially for people who want to dig deep into machine learning/ deep learning. Thanks 3Blue1Brown!
Neural networks is a topic I've wanted an intuitive understanding of for a while. 3b1b has the most intuitive explanations on RUclips. This video could not be any better.
N·J Media - Intuitive understanding is understanding that in a triangle, for example, the side across from a given angle has to increase or decrease in length relative to its opposite angle, without a mathematical proof.
@3Blue1Brown - A quick suggestion: Red-green color deficiency is the most common form of colorblindness. When trying to represent information via a color spectrum, could you please choose colors other than red and green for this reason? Red and blue are good choices because they are distinguishable by both red-green color deficient people as well as blue-yellow color deficient people, which is the second-most common form of colorblindness. I was completely unable to tell which pixels have positive weights and which ones had negative weights in your example due to my colorblindness. Thanks, and keep up the fantastic videos :)
Upper row of this white zone had negative weights, central part had positive, and bottom row had negative weigths.This means that if you have horizontal line this neuron will have high values, but if vertical line or any other patern then it will have value that is closer to 0.
3Blue1Brown is the go-to channel that explains complex math concepts with the highest clarity without any loss of complexity of the topic. Simply brilliant!
Wow a lot of things that i've learned on this first year of system engineering are captured on this video, but previously I didn't understand the real essence of it. Thank you for these amazing vids! Greetings from Argentina :)
I work in a company developing just this kind of stuff. I’m still baffled how incredibly intelligent people are and I have no idea how they can repeatedly accept me as worthy enough to be with them.
I didn't realize that, I started to understand to the Neural Network which was nothing but a black box for me. I must say if every teacher teaches like you the world will produce quality Engineer and Scientist. you really don't need to ask us for subscription, your work is so admiring, we can stop ourselves without subscribing. You have redefined the Phrase " Simplicity is the best way to Handle Complexity", Thank you very Much sir , I wish you stay healthy, wealthy and wise.
It's still refreshing to watch this video, even after so many years. I used to watch this video when I had started my DS journey and used to grasp these intriguing concepts. Such a remarkable video!
Thank you 3b1b. This video certainly gave me a deep enough understanding to allow my neural networks to retain the information. EDIT: seems like I'm not the only one making lame puns about the title.
For the first argument in the video: "You can recognize that all of these images are 3's, even though the pixels are very different." is complete bullshit. Handwriting varies *_EXTREMELY_* person by person and so humans are very used to looking at different ways to write the same thing, especially with things like cursive. It's not a surprise that we can identify the images, please don't talk like it is a surprise, makes me feel like you're less intelligent than you really are.
Calm down a little... Everything what's been said in this video is in context of machine learning, computers, mathematics, algebra etc. So if we want to treat brain as a complex computer than it's function to recognize letters from pixels is amazing and give food for thought how human's brain really works.
ML grad student here and hands down Grant covered an entire chapter concisely and very clearly in this video. I don’t think reading any academic books will give you this amount of intuition on this subject within a few minutes. Still mesmerized by the effort!
It took me one week to understand this when I was reading a university lecture. You explained it to me in 20 mins. You are such a savior. Thanks 3Blue1Brown!
Just imagining how our brains can do subtle background task without realizing it is just mind blowing. It seems like our brains are like computer and our consciousness is the realization of our brain output.
Even after 6 years from making of this video, when we already have something so advanced like GPT4, as a humble beginner in this domain, this video is so so valuable in understanding the very basics! Huge thank you and kudos sir!
I'm in my first year of engineering, looking to go into CS, and this video makes me extremely excited for my coming education. I've already watched so many of your videos, and they've all had a similar effect. Thank you so much!
You're kind of a genius man! I don't care how much you deny it. Your ability to distill these complex concepts into very simple ones and across so many fields in math is amazing. Also, the way you connect different fields of math to explain solutions REALLY shows a different type of mastery. Thank you for all these videos.
I just love the way the concepts of neural networks are explained in this video. After watching it, you feel like you have an idea about the "building blocks" of a neural network. Since I'm new to the topic, it's hard to judge whether crucial things are left out or over-simplified, but I feel it's a great introduction to the topic. Thanks a lot for sharing this!
Thank you from the bottom of my heart :) am not a developer but was trying to understand LLM for quite some time, I stumbled across your video your visual explanation helped me to really understand what is what
This was one of the best tutorials on the fundamentals of neural networks. Formerly, I was a dentist and now a neuroscience research fellow working on computer vision applications in behavioral neuroscience and have never encountered a tutorial explaining so simple and concise, Thanks for that :-)
3blue1brown should probably start their degree now..What visuals and teaching they are providing..is just awestruckingly simplified and easy to understand...Hats off to you..!!
14:50 You can also drop the bias and use a output of 1. Thus add "1" to your a list. a0 = 1 and (w0,0 -> w0,n) are the bias-es. Keeps your program and math much cleaner.
I have a question for clarification. Does your suggestion infer that each bias will update according to the change in weight? I think I’m misunderstanding something, because my intuition (which isn’t much for MLPs) tells me that your suggestion makes the bias dependent on the weight, which loses valuable information.
I'm still in a shock that none of my friend recommended me about this channel. Since the day I've started watching videos of 3Blue 1Brown, my life has been most productive than ever! Never thought that a lockdown would result in so much productivity!
Cant wait to see the videos you release for generative models! I've seen your probability videos and they were so great! Personally, I want to say thank you for making all these videos. It really solidifies everything I've learned. The way you use visuals to describe certain concepts is amazing! Keep it up!
Your videos are singlehandedly keeping my PhD research on track. Thank you for your time and effort!
This is what you're studying for your PhD? This is what I learned in highschool...
Which class exactly did you learn about neural networks? Did you also learn multi-variable calculus (fundamental to even the simplest neural network) in your high school class? I would love to attend!@@randomguy4738
@@randomguy4738 learning is not about phd or high school it's about need
Whenever you need you learn
@@P_Stark_3786obviously there's a reason they are separated, you won't be awarded a PhD if what you "need" to learn isn't at PhD level
@@randomguy4738 What you learn, what he study are on the different level. Just shut your mouth son
I am blown away by the visual clarity of this description of otherwise a complex technology! More please, I am willing to pay!
thanks fellow indian bro.
= 2 USD😂😂😂😂😂😂😂😂😂😂😂😂
@@andrewl2787 Don't be rude bro, appreciate his good intention, 2$ might mean a lot from where he's from.
@@Dtomper Well said. Peace and love worldwide.
@@andrewl2787 Okay so how much did you pay? I will match you, common now
Humanity has benefited a lot from your work Grant. Eternally thankful for your extraordinary work 🙏🏼
Bhai, you're also here
There i watch your videos and learn 😂😂😂
You're also coming here
Nice 👍
I'm literally experiencing the future of education right now, and this was posted 6 years ago
Most educational videos give viewers the impression that they are learning something, while in reality, they cannot reliably explain any of the important points of the video later, so they haven't really learned anything. But your videos give me the impression that I haven't learned anything, because all the points you make are sort of obvious in isolation, while in reality, after watching them I find myself much better able to explain some of the concepts in simple, accurate terms. I hope more channels follow this pattern of excellent conceptual learning.
Huh, I never thought about it this way, but that's a nice way to phrase what I'm shooting for.
Being able to explain it at a conceptual level isn't good enough. You can only understand it by practicing (i.e., build neural nets by yourself and play with it)
busi magen, I don't have a story nearly as touching as that of you and your Grandmother, but I think I would cite my dad as teaching this by example when I was growing up, in that the way that he would describe things centered on what they're actually doing in simple terms, rather than on learning the appropriate jargon.
:-D . Eebsterthegreat: not so obvious insightful complement in reality, as long as you don't read the "I haven't learned anything" part in isolation.
Particle accelerator is used for creating/verifying hypothesis. Your analogy is terrible.
Regarding learning a new skill, one needs to practice rather than just passively absorb information. This is why homework exists.
Regarding neural nets, anyone think they can "explain" NN after watching this video is frankly laughable. (not saying the content of this video is bad)
I study mathematics, physics and architecture. By definition this man is an ORACLE in the strict meaning of the word.
With all honesty I never imagined someone explaining complex topics with the dexterity this man has. He is literally an institution and an outstanding teacher.
The computer graphics and the illustrations are simply perplexing. This guy never evades complexity. He never evades complex arguments. He illustrate the complexity and dive into the exhaustive explanation of the details.
It's extremely rare to see a professor and a dedicated user to put a lot of effort explaining, animating and describing mathematics the way he does.
Well said sir. 🙌
@@sdc8591 This video never claimed to be an expert level tutorial so stop comparing it to those type of tutorials.
@@viharcontractor1679 When did I say that? Please read my comment again. I have no issues with the tutorial, I have objection on the comment to which I have replied. One should always make an appropriate comments. As it is incorrect to say something rude, it is always wrong to do false praising. Have you read the comment? of kummer45? Calling the tutor of the video as " Oracle"? Really? This kind of words should be used for someone like Swami Vivekanda and not for some ordinary tutorial. It almost hurts to see such misuse of words.
@@sdc8591 Come on man, what is wrong with / about that comment? The video is fantastic in every way, It is dense enough that I've had to watch it several times over, yet is able to communicate the concept of a neural network in such a way that even my pea brain can grasp this topic, please think before commenting and make a proper comparison.
@@JCake First of all, don't use the word 'man' , I am a girl. I never said video is bad.It is fine. Why everyone is coming over here and defending the video? Is is so difficult to understand what I am saying? The comment from kummer 45 is an exaggeration and I stick to it because it is. If the video is good enough , one does not need to watch it second time to understand the concept. I had seen one video by Mathew Renze on the same topic. That long tutorial, was first time I came across neural network. It was more that 1.30 hours of series of videos. I never watched it again and still remember every single concept. Now if this man is oracle what you will call him?
I am Programming for more than ten years and I never saw anyone explain a complex idea by such a clean and clear terms. Well done.
yeah thats power of manim !
can you explain the animation at @9:30
Because you only program without mathematics
secret scientist are much further than this, like they can wirelessly from satellite I think give dream images and change dreams you make yourself... for me they always try to make it ugly,...
Markov Chains ? how simple is that ?
Fantastic visualized learning!
Quote: “Any fool can make something complicated. It takes a genius to make it simple.”…..nailed.
"What one fool can do, another can"
Some guy who wrote a really popular calc textbook
It's a statement I don't agree with. At university, we are taught things in a formal and abstract way, not just for the sake of overcomplicating things. I don't think professors, which are primarily researchers should be considered "fools" because they fail to teach their subject in a more intuitive manner.
@@genericperson8238 a good researcher is not necessarily a good teacher.
@@genericperson8238 Yes, they're a fool in pedagogy
A genius doesn’t really make it simply more they make it concise.
I can't wait for neural networking to be able to recognize my doctor's prescription.
They need to study pharmacists to figure that out.
That would be magnificent!
from what i understand (i am also an dummy i just tell you what i think)
the inputs are the pixels
the weights are the pixels 's whiteness or blackness it is
like lets say we need first pixel to be white
so we need the computer to know there is a pixel there (hence it's an input)
we need the computer to change how white or black it is (hence the computer's ability to change weights)
its actually impossible, that level of calligraphy is indecipherable
Doctors writing make Strings Theory a piece of cake for humans, AIs and aliens.
This is how you taught Deep Learning, people. I've seen lectures that either be categorized into 2 groups: too hard or too shallow/general. You have balanced between them. Thanks you so much!
Brilliantly explained
I had learned about neural networks and knew the mechanics of it. But this is way better explained - you nailed it - brilliantly.
@@uncommonsense9973 Sorry to disappoint you but the commentor isn't the creator of the video lol
@@randompersondfgb he was agreeing with the commenter bro
@@wnyduchess To quote the reply itself; “this is way better explained - *you* nailed it - brilliantly”
@@randompersondfgb yes, you're not understanding. Uncommon Sense is saying "you" nailed it. The you is Cliff Rosen, the original commenter. He's saying that Cliff Rosen nailed it when he wrote the comment "Brilliantly explained".
this is my first introduction to machine learning and I watched this only twice to get it, really goes to show how good of a teacher this guy is, the effort he puts in is nothing short of amazing !
Definitely am gonna have to watch it again. Got half way through and it started to get pretty heavy
I'm currently taking a computer science math course where the professor strongly advised everyone to watch this exact video series to get an intuition about what all the math is actually used for.
Bro, which college you studying in now?
@@vgdevi5167 Aarhus University, Denmark
@@vgdevi5167 1st semester :)
Good to learn from, and also, entertaining to watch. double win.
Linear Algebra? That’s what I’m following in about 6 weeks, which is basically the math behind Neural Networks
not only is your content extremely helpful in terms of learning complex concepts, but your thoughtful creation of every element - from the overall progression of ideas down to the smallest jokes or teaching points embedded in graphics - is deeply inspiring. Thank you, 3b1b
Around 2 years ago I was a sophomore statistics student and had no idea what deep learning is, until I met this video and 3b1b channel. His clear explanation of neural network and animations blew my mind. Since then I started my journey in machine learning. For a random reason I clicked onto this video again, and realized how long my journey in this field have been. This video really changed my life and I am really grateful about it.
@3Blue1Brown
Please give a heart .......
I am in class 11 currently and unfortunately I am not able to understand this. Could you point me to some prerequisites?
@@yashrathi6862 The linear algebra series that was recommended in the video is a good start, other than that you should keep watching this video and you will start to understand it better the more you do. I am also in class 11 and that is what helped me
One year ago I met this video. I couldn't understand any single word in it. A year later, I am back and I still cannot understand it.
I am fucking stupid.......
@@yashrathi6862 To be honest there are no real "prerequisites" for learning neural networks, in the end it just gets down to how familiar you are with the concepts of basic graph theory. However, I admit that it can be pretty overwhelming for someone to try and comprehend all the stuff at once, which is why being savvy with the use of linear algebra is a must.
Apart from that you should try your hand at programming once, perhaps the algorithmic mode of thinking would help you deveop an intuition for neural networks. And yes, of course try to explore graph theory, for neural networks will resonate much better with you once you do, imo.
No man, we don't get notifications for your videos. We search for 3b1b. That's how powerful your content is.
I don't understand notifications either. What, we're supposed to do things other than watch all the remaining 3b1b videos we haven't yet seen between notifications? Who would be so wasteful with their lives???
I just type questions into RUclips and always seem to get his videos as answers
@@FlyingSavannahsusually you'd enable notifications only for youtubers who make videos that you're almost always interested in
This is my very first time commenting on a RUclips video, and it's just to say: This is the best explanation of anything ever.
yet, people don't understand
Some Eng congrats man
i couldn't understand anything over a minute in
Wait until you see his video about the Fourier Transform. My GOD that vid is the best thing i've seen in ages.
I am currently doing my Master's in Data Science and this 18 minute video is better than any course I have taken so far
maybe give up ?
maybe ligma
then you should get into a good university
In schools everyone taught us to practice maths but this man teaches us to imagine maths
True🔥🔥
best comment here, period!
Juju
Did you mean 'visualize' maths
They don't teach you maths, they teach you how to solve exam questions. Maths is what 3Blue1Brown teaches.
Our generation is lucky to have mentors like you, thank you so much sir!
Fine Indeed, Refreshing Super Tenacious
This is the 80s generation we were listening rock music and looking how to get things done better we grew without mobile phones just sitting front of a computer or playing basketball outside in the park. We grew without rap, hip-hop, either thinking that the gang is a cool guy! this is what now generations require badly!
@@elgary9074 I hope that someday scientists will be able to understand what you have written.
@@MiguelAngel-fw4sk 🤣
And our generation is unlucky that we had no such mentors and internet to deliver their videos. Taking this into account, we demand results, youngsters! We had, at least, an excuse for being dumb :)
It takes 3000-4000 lines of code to make those graphics possible, he's a freakin legend
Which is best for neuron? python or c++
@@omarz5009 The main downside of python is the fact it's a high-level language and hence kinda slow. But for ML and NN it has several powerful libraries (pandas, numpy, tensorflow) which make up for that. Given Python supports the implementation of C-Code, those libraries could be optimized like heck to the point bothering with the stuff in C++ is just wasted time. Plus Python is much easier to learn, hence more people use it and develope for it.
@@jagaya3662 That makes sense. Thanks for explaining :)
@@anelemlambo497thank you for explaining :)
How these graphics and animations were made actually?
Every few years I come back to watch this series. The most intuitive and understandable explanation of neural networks that exists
It's like coming to look back at art. Pretty much every 3b1b video is a masterpiece!
The fact that I was sent here by my university lecturer is a testament to how good 3Blue1Brown is.
Same here😂
Same here😂😂
Same here 😂😂😂
Same here 😂😂😂😂
Hm......
_Daniel Shiffman?_
One of the few teachers that don't make you feel stupid, but actually help you understand the topic. I appreciate the time you spend on this.
This video kickstarted my journey in ML a year back. Trust me, back then I watched this video three times to finally understand. It might be challenging for few to get it but when you get it, it just feels amazing
@@DawnshieId why do you think it cannot go beyond 1?
Felt like the brain chair meme when this video finally clicked (after the 4th watch)
@@chitranshsrivastav4648 How do you weight?
after 8:38 felt really hard to understand.. I will try again and comment back
Hell yeah. Im literally in your shoes rn
"Even when it works, dig into why" - 3B1B. Your lessons are pure gold sir. I'm here after watching the entire Essence of Linear Algebra. Thank you.
My God! No words to express as to how you made such a complex topic to be understood using visuals so easily! Hats off!!
In class : printf("Hello world");
The exam :
kkkk
Why did introduce us to quantum mechanics. You sucks.
😂😭
@@ThomasJr {i>{
@ゴゴ Joji Joestar ゴゴ Lol that's because physicists couldn't find anything interesting
Is anyone else nominating this series for the "Distill Prize for Clarity" in 2019? I really think he deserves it, excellent visualizations.
Yeah I would, every day of the year.
Totally! Animation and visualization here makes understanding as clear as a crystal!
yesssss
@@benisrood Can it be nominated for anything else?
I am just astounded. I spent so much time trying to understand this concept. Everywhere I looked people would show the similar neural network animation, but no one ever really explained and exemplified every single step, layer, term and mathematics behind it.
The video is really well structured and with amazing animations. Extremely well done. My mind is so blown I can barely write this comment.
"My mind is so blown I can barely write this comment." lmao
🎯 Key Takeaways for quick navigation:
00:04 🧠 *Introduction to the recognition of digits*
- Your brain's ability to effortlessly recognize digits,
- Challenges in programming a computer to recognize digits,
- Introduction to the relevance and importance of neural networks in machine learning.
01:06 🤖 *Devoting the video to the structure of neural networks*
- Creating a neural network for recognizing handwritten digits,
- Exploring the simplest form of neural network with no added complexity,
- The network's ability to recognize handwritten digits.
02:37 🧩 *The inspiration and components of neural networks*
- Explanation of neurons in neural networks,
- Neurons as holders of numbers between 0 and 1,
- Introduction to layers in a neural network.
03:36 🧠 *How neurons and layers are connected*
- Description of connections between neurons and layers,
- Activation values and grayscale representation in the input layer,
- Exploring the concept of hidden layers and their purpose.
05:06 📊 *Understanding neural network learning and training*
- How a trained neural network recognizes digits,
- The role of activations and weighted sums in making predictions,
- The complexity of learning involving thousands of weights and biases.
07:31 ⚙️ *The role of weights, biases, and patterns in neurons*
- Explaining the significance of weights and biases in neurons,
- How weights contribute to recognizing specific pixel patterns,
- The role of biases in determining neuron activation thresholds.
10:32 🧮 *Compact notation for neural network connections*
- Representing neural network connections using matrices and vectors,
- Simplifying the expression for transitioning activations between layers,
- Highlighting the importance of linear algebra in machine learning.
13:11 🧩 *Neurons as functions and the complexity of networks*
- Viewing neurons as functions that take inputs from the previous layer,
- Recognizing the network as a highly complex mathematical function,
- Discussing the challenges of setting weights and biases manually.
16:20 🔄 *Learning and the future of the neural network*
- The promise of neural networks in recognizing complex patterns,
- How neural networks learn from data and adapt their parameters,
- A brief mention of ReLU activation functions and their advantages.
Made with HARPA AI
When will AI be able to invent complex strategy games such as Chess and Go?
@alexandrebour7494 Chess and Go are not complicated in terms of rules, especially Go, but indeed quite complex to master. You can ask AI to "invent" a game today, and it will try to mimic existing games and mix and match board-game elements e.g. deck building, work placement, etc. Today, you could build an agent that writes game rules, the other that writes code for the game sandbox, another one that tries to master it, and yet another one that "evaluates" how good the game is. If you run this agent chain over a sufficient computing power, you may get something very interesting
@@HarpaAI I don’t have a computer that can handle that computing power. On the other hand, I have no doubt that advanced technology that can do this work more easily can happen in the near future. This could help me improve the complexity of a pre-existing game that I particularly like, the Hnefatafl.
Bio teacher: what is a neuron?
Me: a thing that holds a number between 0 and 1
lmao good one.
get out of my class
InSomnia DrEvil Great explanation, but you ruined the joke lmao
@@prabeshpaudel5615 haha
Also known as FUSSY Logic
Finally, a video that does more than just present some neurons and layers and say, “here’s an activation function.” Your video describes how the model is developed and why the algorithmic approach is appropriate for the problems neural networks try to solve. Thanks!
3Blue1Brown
"Sigmoid Squishification Function": 11:23
Most brilliantly named function I have ever heard named. Absolutely brilliant. The merger of the technical with the simple with a double alliteration for easy memory.
Sigma squishy function
i took a deeep learning lecture in my last semester and my professor couldnt explain in 4 frickin months what u explained in 20 mins much much appreciated man you're doing awesome work hope to learn a lot from you
I am a Data Scientist and I would like to tell you THANKS.
I have NEVER met anyone with the ability to teach complex things in this way.
A M A Z I N G.
Please continue like this, for example with other statistics videos. You can substitute many of the University courses.
May I know which type network connection allows neural network
As a person who has self-learned a bit of python and is just trying to learn this stuff, this is exactly the best place to begin.
My thoughts exactly!
At the end of the video, he showed the relu function f(a)=a with a>0, so the value of the neuron doesnt have to be between 0 and 1?
That's me
@@duykhanh7746 A bit late but if your question hasn't been answered yet: It doesn't really matter if you have a value >1. Basically anything above 0 is an activation and you can also view it as the size of "a" being the intensity of the activation. Biological neurons can also be more active by firing in fast succession (up until they reach the maximum possible firing rate of like 250-1000Hz depending on the source), but you don't want to introduce things like loops in artificial neurons to not slow down your network. So to simulate this kind of behavior, you just let your output get bigger. You can compensate for the lack of an upper limit in the following neurons by adjusting the weights and the biases.
TL;DR: No. :D
🙂
how is it possible that I can lie in my bed on a Sunday and am presented with mind-boggling cutting edge knowledge told by an incredibly soothing voice in a world class manner on a 2K screen of a pocket supercomputer basically for free
Welcome to the 21st century
yet 90% of people use that supercomputer to mindlessly scroll feeds.
@@Charge11 And software engineering advancements, thousands of years of intellectual history, biological evolution of conscious brains and so forth.
point is, it's miraculous if you step back far enough.
because it isn't
It's not free, Google's massive network of AI neuron is harvesting terabytes upon terabytes of information about you every time you click on anything.
I just can't believe this came out 7 years ago.
You are the best.
Same😪
This is the best intro to neural networks I have ever seen. The presentation is excellent! The animations are very very very helpful especially in understanding the formulas and matrices and how they came to be. Thanks a million. Looking forward for the next one.
THE TEACHING ASIDE , THOSE GRAPHICS MAN! TAKES LOT OF EFFORT!
Exactly....Lot of effort is required to make this type of video.
why iam seeing Indians everywhere
@@hmm7458 cause u are also an indian...
@@kartikeya9997 that's not an answer lmaooo
I know I can't do better. I'll be referring students in my neural networks class to these videos, lol.
Written some notes from the video to read quickly. Hope it helps somebody.
l Neural Networks can recognize hand written digits, letters, words ( in general, tokens )
l What are Neurons?
○ Something that holds a number [ 0, 1]
○ The higher the number, the higher the "activation" rate
l Consider a 28*28 table in which each unit is represented by a value between 0 to 1 ( activation number )
○ Let us divide each row into a "layer", such that, if we were to divide all the layers, the last layer would contain 10 "cells" ( units ).
○ Values are passed from the previous cells to the last layer ( 10 unit layer ), again, between 0 and 1. The higher or closer the value is to 1, the more probability exists that the image scanned represents that unit cell.
So, a unit cell that contains the highest value is indication that the index of the unit cell is the value of the image scanned.
○ 16 cells in the second and third last cells are arbitrary.
○ Each cell is linked ( causes activation ) to some ( not all ) other cells in the next layer which further cause more activation.
○ Each 'cell' corresponds to some sort of identification about how much a certain region 'lights up', and then sends a value to another node which reacts based on the received value.
○ To find whether a certain cell with light us, like each cell be represented by 'a Cell_Number ', and let each cell be 'assigned' a certain weight 'w'. The sum of all the products of each cells 'a' and 'w' will be:
w1*a1 + w2*a2 + w3*a3 + w4*a4 + … + wn*an
○ Let these weighted sums represent some 'grid cell'. Each cell is either 'on' or 'off' with respect to being positive or negative. In this case, 'green' represents on, and 'red' represents off.
○ Let us concern ourselves to a certain region where the cells are mostly on. Ergo, we would be basically summing up the weightages of those grid cells.
○ Then, if you suppose a region where there are brighter grid cells in some part which are surrounded by dark grid cells, then that area is the main edge we're looking for.
○ Of course the sum of weightages gives us very different value. In order to 'squish' that number line into 0 and 1 , we use the function:
Sigma(x) = 1/(1 + e^-x)
Which is a sigmoid function or a Logistic Curve. Our equation now becomes:
Sigmoid(w1*a1 + w2*a2 + w3*a3 + w4*a4 + … + wn*an)
○ But what if you don't always want to light up when it's a positive value, and rather want it to light up when the weighted sum of that grid cell full fills some condition, such as > 10. This is called 'Bias For Inactivity'. Using this example, our equation becomes,
Sigmoid(w1*a1 + w2*a2 + w3*a3 + w4*a4 + … + wn*an - 10)
Here, 10 is the "bias".
○ The possibilities of the different knobs and dials open us to the term of "Learning", which just means to find the correct relation of values which perform the expected behavior.
○ The complete expression above can be adjusted in the formula:
a(1) = Sigma(W*a(0) + b )
( (1) and (0) are superscript here )
Where W = k*n matrix whose elements are weights corresponding to a cell.
a(0) = n*1 matrix whose elements are the 'a' of each cell.
b= n*1 matrix whose elements are the biases of each cell
○ NOTE: Sigmoid function is not used very often now, instead it is replaced by ReLU ( Rectified Linear Unity ), which is defined as:
ReLU(a) = max(0, a), a linear function where f(a) = a for a>= 0, which for a < 0, f(a) = 0.
thanks..
Good work👏👏👍, thnx
Thank you sooooooo Much!!!!!!
You saved my life!!!!
Thanks!
I know you read this all the time, but I must say it. You videos are simply incredible! Your work reshapes education. You deserve every cent that this platform puts in your pocket.
Behind this material is an extreme shot of giftedness. Explaining something is not easy. You first need a solid physical model for the topic in your brain and then you need to translate this model into a mental model that can be faithfully exported into others' brains. I congratulate you for this excellent job and I hope that you appreciate what you are and what you are doing. This is much more important than how much money this business brings.
I just watched Welch labs machine learning playlist a few weeks ago. It was mind-blowing. I'm glad you're getting into machine learning too! : )
Yes, Welch Labs is truly great.
Thanks for the recommendation!
watching this for a second time and i can't believe how illuminating is to come back to the basics and get a renewed understanding -- grant, you're a treasure
This the most comprehensive and understandable explanation of a neural network. Thank you.
This is the first time I'm commenting on a RUclips video and honestly, I'm so thankful people like you exist! I wish only the best for you in whatever you do!
PART 1? THERE WILL BE MORE? YAS 3BLUE1BROWN IS DOING NEURAL NETWORKS! TODAY IS A GOOD DAY
You will find this series very helpful as well.
ruclips.net/video/bxe2T-V8XRs/видео.html
I totally agree, my friend. Today is a very important day in the history of youtube mathematics. And since I am the 100th person who liked your comment, I would like to give a little inspirational speech:
To all mathematicians, physicists, engineers, computer scientists or people who want to become one of those in the future,
today is a very important day. The best youtube mathematician, 3Blue1Brown, has made a video about neural networks and plans to make others about it in the future. I think it's not necessary to explain the inherent significance this topic has concerning the future of our technology and our understanding of the universe and the processes going on in it. These videos will help the new scientific generations to cope with the structures still to be found and to bring on a new and deeper understanding of the things that have been found and examinated before. Humanity is reaching a point, where the wish to understand the world is higher than it has ever been before. You, dear future scientists, can all be a part of the progress we are just going through, you just have to have the Will and the Strength for it, never give up if things aren't working properly or as you expected and always remember: At the end, everything will be fine, so if it isn't fine, it's not the end.
Actually, I have reached the end of my little inspirational speech (and it is fine ;) ), and to complement it well, I want to quote a famous poem which plays an important role in a very good and famous science fiction movie....
"Do not go gentle into that good night,
Old age should burn and rave at close of day;
Rage, rage against the dying of the light.
Though wise men at their end know dark is right,
Because their words had forked no lightning they
Do not go gentle into that good night.
Good men, the last wave by, crying how bright
Their frail deeds might have danced in a green bay,
Rage, rage against the dying of the light."
Thank you.
This comment is lit!
yas
RNN? LSTM?
One “like” is not enough for the work that has gone into making one such video. This video should be part of the curriculum and he should get the royalty for this. Awesome work!
Yes!
I'm currently taking a computer science math course where the professor strongly advised everyone to watch this exact video series to get an intuition about what all the math is actually used for.
+1
My goodness, I’ve watched nearly 20 videos on neural networks, and none of them come close to this one in terms of visual representation and clarity. Thank you very much.
I'm studying AI for my masters degree and my professor told everyone to watch this video to understand the concept :D
he knows...
amazing. :D
me too!
School is a scam
Where are you studying ai??, I mean what is the name of your college or university????
I am not really from a math background but I am hugely interested in programming, and I must say this video has made it easy for me to understand the math behind neural networks!
I loved it , thank you!!!
Every second of this video is a Pre-requisite to the next second of the video :D
This channel is so damn good. Other channels give some terrible analogies and some other explain it in extreme technical detail. This strikes the perfect balance and provides a foundation to understand the more technical details
I wish this guy was my math teacher back in high school.
Just shows that good teaching skills are very rare.
Understandable animations on the perfect timing with the words, and no holes on the explanations, makes the trick
Another reason to be mentioned on why ReLU is used instead of Sigmoid is simply the fact that it calculates a lot simpler (obviously cutting negative values vs. exponential operations). Plus another important issue of the σ function is it's gradient which is always below .25. Since modern networks tend to have multiple layers and because multiplying multiple values < 1 quickly becom really small (vanish) networks with a larger number of layers won't train when using Sigmoid.
And as always, amazing video, animation and explaination!
This channel and the visualizations it produces to teach subjects like this one is the best advance in the history of communicating mathematical ideas. It's extraordinarily inspiring that one person can have such a large impact on the world today (and for generations to come). Thank you, Grant Sanderson.
Dude, inspiring comment yourself.
To think that someone would make a video of neural network and explain it in a way so simple yet insightful is such a bless especially for people who want to dig deep into machine learning/ deep learning. Thanks 3Blue1Brown!
Neural networks is a topic I've wanted an intuitive understanding of for a while. 3b1b has the most intuitive explanations on RUclips.
This video could not be any better.
MTO Intuitive understanding?
It isn't intuitive understanding if you have been looking for a explanation in a while xd
N·J Media - Intuitive understanding is understanding that in a triangle, for example, the side across from a given angle has to increase or decrease in length relative to its opposite angle, without a mathematical proof.
@3Blue1Brown - A quick suggestion: Red-green color deficiency is the most common form of colorblindness. When trying to represent information via a color spectrum, could you please choose colors other than red and green for this reason? Red and blue are good choices because they are distinguishable by both red-green color deficient people as well as blue-yellow color deficient people, which is the second-most common form of colorblindness. I was completely unable to tell which pixels have positive weights and which ones had negative weights in your example due to my colorblindness. Thanks, and keep up the fantastic videos :)
Upper row of this white zone had negative weights, central part had positive, and bottom row had negative weigths.This means that if you have horizontal line this neuron will have high values, but if vertical line or any other patern then it will have value that is closer to 0.
windows 10 has colour filters that will fix this for you. go to settings, ease of access, and click on 'colour filters'
3Blue1Brown is the go-to channel that explains complex math concepts with the highest clarity without any loss of complexity of the topic. Simply brilliant!
I'm studying ML and just saying, I've understood neural networks and this is my first tutorial. THANK YOU!!
SUBSCRIBED!!!
Wow a lot of things that i've learned on this first year of system engineering are captured on this video, but previously I didn't understand the real essence of it. Thank you for these amazing vids! Greetings from Argentina :)
Boludo
I work in a company developing just this kind of stuff. I’m still baffled how incredibly intelligent people are and I have no idea how they can repeatedly accept me as worthy enough to be with them.
impostor syndrome. There will almost always be someone better than you, but you are probably better than you give yourself credit for
I didn't realize that, I started to understand to the Neural Network which was nothing but a black box for me. I must say if every teacher teaches like you the world will produce quality Engineer and Scientist. you really don't need to ask us for subscription, your work is so admiring, we can stop ourselves without subscribing. You have redefined the Phrase " Simplicity is the best way to Handle Complexity", Thank you very Much sir , I wish you stay healthy, wealthy and wise.
It's still refreshing to watch this video, even after so many years. I used to watch this video when I had started my DS journey and used to grasp these intriguing concepts. Such a remarkable video!
This will go down as one of the best lectures in history. What an amazing and concise explanation of something I thought I would never understand ...
totally agreed
Thank you 3b1b. This video certainly gave me a deep enough understanding to allow my neural networks to retain the information.
EDIT: seems like I'm not the only one making lame puns about the title.
For the first argument in the video: "You can recognize that all of these images are 3's, even though the pixels are very different." is complete bullshit. Handwriting varies *_EXTREMELY_* person by person and so humans are very used to looking at different ways to write the same thing, especially with things like cursive. It's not a surprise that we can identify the images, please don't talk like it is a surprise, makes me feel like you're less intelligent than you really are.
Calm down a little... Everything what's been said in this video is in context of machine learning, computers, mathematics, algebra etc. So if we want to treat brain as a complex computer than it's function to recognize letters from pixels is amazing and give food for thought how human's brain really works.
Peter Njeim it's not a surprise that you can identify images. The surprise is how complicated image recognition actually is if you think about it.
Peter Njeim, do people invite you for parties?
Faculty of Khan M
ML grad student here and hands down Grant covered an entire chapter concisely and very clearly in this video. I don’t think reading any academic books will give you this amount of intuition on this subject within a few minutes. Still mesmerized by the effort!
Commendable beyond words! 7 years later and this video still explains the concept better than anyone today could
and it was GPT o1 that suggested your video 😂
It took me one week to understand this when I was reading a university lecture. You explained it to me in 20 mins. You are such a savior. Thanks 3Blue1Brown!
When your brain is smarter than you
BEST comment I've seen on YT in a loooong time! :)
@@marshalldestro CALM DOWN
Just imagining how our brains can do subtle background task without realizing it is just mind blowing. It seems like our brains are like computer and our consciousness is the realization of our brain output.
@@goustlz Wow. i am sure most people do not think of this way
You comment made me laugh out loud, I take of my hat to you good sir/madame/unicorn.
The best introduction to Neural Net's I've ever seen. Kudos!
Even after 6 years from making of this video, when we already have something so advanced like GPT4, as a humble beginner in this domain, this video is so so valuable in understanding the very basics! Huge thank you and kudos sir!
I'm in my first year of engineering, looking to go into CS, and this video makes me extremely excited for my coming education. I've already watched so many of your videos, and they've all had a similar effect. Thank you so much!
Same energy brother!
How has your degree been going?
You're the first person to explain bias in an intuitive manner. Thank you.
You're kind of a genius man! I don't care how much you deny it. Your ability to distill these complex concepts into very simple ones and across so many fields in math is amazing. Also, the way you connect different fields of math to explain solutions REALLY shows a different type of mastery. Thank you for all these videos.
Activation, weights, bias. I suddenly understood them all. I can't believe it. You're awesome.
His videos are so elegantly illustrated and flow of thought is so clear. Watching his videos is like listening to music of Mozart to me！
This guy: Uses dark screen to illustrate a long concept so that it's easy on the eyes.
Everyone: "Carefully, he's a hero."
Careful, you might poke the boomers who think that black themes are just for edgy people.
Meanwhile schools: LET'S CHOOSE THE BRIGHTEST AND WHITEST COLORS FOR THE LIGHTS, WALLS AND ACTUAL BOARD
They're bound to pay attention then, right?
Me at night 👌
What a time to be alive, with such RUclipsrs around!
Can’t believe this video is made 7 years back. Such a nice explanation. Thanks, man.
I just love the way the concepts of neural networks are explained in this video. After watching it, you feel like you have an idea about the "building blocks" of a neural network. Since I'm new to the topic, it's hard to judge whether crucial things are left out or over-simplified, but I feel it's a great introduction to the topic. Thanks a lot for sharing this!
The most intuitive channel on RUclips...
I will forever be grateful to you for making learning so much fun!
Thank you from the bottom of my heart :) am not a developer but was trying to understand LLM for quite some time, I stumbled across your video your visual explanation helped me to really understand what is what
Can't believe how well explained and intuitive this is. I aspire to become a teacher like you.
This was one of the best tutorials on the fundamentals of neural networks. Formerly, I was a dentist and now a neuroscience research fellow working on computer vision applications in behavioral neuroscience and have never encountered a tutorial explaining so simple and concise, Thanks for that :-)
what ? you never took Markov Chains ?
I don't know how to express my gratitude to you ...
all your videos are just amazing and incredibly informative.
3blue1brown should probably start their degree now..What visuals and teaching they are providing..is just awestruckingly simplified and easy to understand...Hats off to you..!!
14:50 You can also drop the bias and use a output of 1. Thus add "1" to your a list. a0 = 1 and (w0,0 -> w0,n) are the bias-es. Keeps your program and math much cleaner.
I have a question for clarification. Does your suggestion infer that each bias will update according to the change in weight? I think I’m misunderstanding something, because my intuition (which isn’t much for MLPs) tells me that your suggestion makes the bias dependent on the weight, which loses valuable information.
I'm still in a shock that none of my friend recommended me about this channel. Since the day I've started watching videos of 3Blue 1Brown, my life has been most productive than ever! Never thought that a lockdown would result in so much productivity!
they aren't your friend.
@@userwheretogo 😥😥 Hits hard
cool
seriously, this is the first time i find that ML makes sense! you are amazing
Thank you! Such a great video! Please more!
Cant wait to see the videos you release for generative models!
I've seen your probability videos and they were so great!
Personally, I want to say thank you for making all these videos. It really solidifies everything I've learned. The way you use visuals to describe certain concepts is amazing!
Keep it up!
Currently doing my capstone on deep learning and this is among the best, and easiest to understand descriptions I have seen.
I wonder how many knobs and dials in my real neutrons get tweaked how fast to watch this video. Fantastic. Two thumbs up.
Thank you for this amazing content. Pelase keep publishing about physics and CS!
Most fascinating channel on YT, hands down.