Your videos are amazing! They remind me of the quote "anyone who tries to make a distinction between education and entertainment doesn't know the first thing about either"
Siraj, I jumped into Deep Learning about a year ago and am a huge fan of your videos; this would have been a fantastic resource when I first started out and I think it's so cool that you're taking the time to get through the foundational stuff so everyone can enjoy your other, awesome videos! Thanks :)
Damn the way you've moved from Scalar-to-Vector by Algebra-to-Linear algebra, so Array-to-Matrix. You're a professional Educator, Siraj. Thank you very much.
Loved this video. Very interesting subject. It reminds me of stuff I have been thinking about myself independently for years. Thank you Siraj, this video was great.
SIRAJ, i finished the Spanish subs in the video of the Intro, maybe i'll make the next tomorrow or today at the night. Please, accept it and update the Google Drive Document with the script of this video when you can (: Awesome vid bro!
Gracias Pinkie, i've accepted your subtitles. I've also added this script to that same document. I will continue adding every weeks script to that document and review any subtitle submissions you make. :)
Awesome vid Siraj. One question - at 8:56 when you show the 4 types of norms, is there a norm that you recommend developers should use most often? Or does it depend on the type of ML model you're building?
Hey Siraj Raval, how probable is it for abstract machine learning libraries to reduce demand for machine learning expertise to the point that the average person could not reasonably expect to gain employment in machine learning after investing the considerable effort it takes to understand the underlining math of machine learning? For example in a previous video you showed a abstraction of tensorflow that allowed you to simply take a labeled data set f(Vi) = Yi. All training/learning was then abstracted by the library and and you simply input a new vector that was not labeled and and the library output its best guess of y's value or label based upon the data set you feed it. Will machine learning EXPERTISE only be necessary for companies that develop machine learning libraries? How long is the window for wide demand, not counting grandiose entrepreneurial endeavors?
So how does matrix multiplication work programmaticly in numpy? I thought matrix multiplication is done the algebraic way in a loop like you showed in the algebra example and numpy is just faster because its doing its looping calculations in C. (4:52)
let me also add, the more I go thru his mathematics videos the more I despise my past mathematics teacher. I flunked highschool algebra, yet now I find myself using linear algebra to optimize learning programs. Thank You for the teaching, Siraj.
WOAH. I didn't expect you to understand Einstein Field Equations! Dammit you gotta do a series on them too :3. You inspire me Siraj. You really really inspire me to keep learning. Thank you for existing.
Please help, I don't understand what is tensor. Vector is invariant under coordinate system. I can use no coordinate system at all e.g. for moving particle under the known curve - I can measure the path length. Q1: Is tensor depend on coordinate system? Q2: I have rather good example of tensor of second order like intertia tensor of rigid body. I can use it to "exchange" non-free rotation vector "w" which I can move along a line into momemt of impulse(kinetic moment) via L=IW via using using usual rules of mulitplication. But what does it mean to be a more high dimensional tensor like 3? In video it have been said that this just an array of numbers but to which ojects I can apply this imagine tensor of degree 3 and how I should do it? How this numbers which are represented a tensor changes if use other coordinate system to describe my objects? Thanks
Hey Siraj, please consider making a video on Object detection, the deep learning approach perharps with Fast-RCNN, YOLO, SSD or The tensorflow object detection API..please....
Zero norm at 08:59 is not a norm. Never was be, never will be. Some people which work in convex optimization called it caridnality. At lease because card([1 0]) = 1, but card(2*[1 0]) is also 1. It is not 2.
Thank you for the explanation, but one thing I'm still not clear on is how does Word2Vec finds similar words from sentences. For eg. 1. Mercedes is a German legacy 2. Honda on the other hand is pretty new 3. But nothing beats Toyota reliability In this context, how will Word2Vec get to cluster {Mercedes, Honda, Toyota} as one group - can you please help explain this. I do get calculating distance, vectorizing the position of each word and selecting the dimension randomly but how does Word2Vec works. Thank you again Siraj from a software developer.
hi can you answer this for me sir? 😶 " given the line x+3y+5=0 and a circle x^2+y^2-6x-4y+3=0 find the equation of the tangent to the circle which are parallel."
Would be great to see someone actually point out the mistakes in this video. I begin by pointing out 5:31. By reshaping your input you don't take it to a new vector space. An mXn dimensional vector space and a 1Xmn vector space are homeomorphic. Anyone want to add to this list?
Adding on as I watch, 6:00 word2vec was not a single neural net. It was an ensemble of models. And no it can't learn the embedding for a new word. There's no training/testing split there. There's only an embedding.
I think the example compares an iteration pattern using algebra with processing arrays in say, Python. The %%time magic command will proved that list computation is faster.
HE IS MY 'RIVAL' WE HAVE EQUAL POPULARITY AND MY ADS ARE LIKE HIS WITH SIMILAR COVERAGE AND MONETARY RECOMPENSE, LIKE SCROUGE AND MCQUACK, IT'S A RACE TO THE FINISH WHERE WINNER TAKES ALL, THANK GOD
Crap you are convincing me to take the two vector calculus courses at my college. In addition to the multivariable course. Okay so which foundational papers in the field should I read and which major books for class, for amateurs and professionals.
Hey, I find your memes to be distracting. The one with spongbob at :50 didn't really carry the point you were making. Don't get me wrong, I love that you use memes, but I feel they should be included only to solidify the point rather than a quick joke - or after your point is made so it's not distracting.
09:54 - I think it false description in that sense that you're overfit and etc. what is real going on you solve bicreterion optimization problem. For more details what is regularization please take a look into convex optimization books, such as this web.stanford.edu/~boyd/cvxbook/ p.s. Thanks for video! It is still great!
Your videos are amazing! They remind me of the quote "anyone who tries to make a distinction between education and entertainment doesn't know the first thing about either"
thanks Jonathon! Reminds me of Disney, one of my main inspirations
no matter what people say about your fast paced teaching , i am in love with your style .
"Relationships are like Algebra, you look at your X and Wonder Y"... damn man 😂😂😂
;)))
:D
Yeah literally, it was dope !
Siraj, I jumped into Deep Learning about a year ago and am a huge fan of your videos; this would have been a fantastic resource when I first started out and I think it's so cool that you're taking the time to get through the foundational stuff so everyone can enjoy your other, awesome videos! Thanks :)
awesome!
Damn the way you've moved from Scalar-to-Vector by Algebra-to-Linear algebra, so Array-to-Matrix.
You're a professional Educator, Siraj. Thank you very much.
Great video siraj!your videos aren't just informative but entertaining as well, keep making them. big fan here
I understood a couple of those words.
Izumi Koushiro you need more epochs then..
Epic! Keep watching, concepts will be reinforced over time
Brilliant demo of how vectors can be used in real life. Helps me to put it in the context of my linear algebra course👍
this channel is my safe place.
im honored and will keep it that way
GREAT visuals and production and perfect explanation. you just keep getting better at this.
thanks Aidan for following my progress
Siraj Raval 🅱️🅱️🅱️🅱️🅱️🅱️
Vectors are pretty much the major problem I'm facing with ANNs right now.. thanks for that video!
I have to watch this twice. Once for the math and the other for the humor
Amazing!! The audio was the best on any video yet. Thanks for making this :')
thanks Jake
Best explanation of 'word embedding'! Superb video!!
i've learned soooo much from this series so far.. awesome job.. love you Siraj!
Loved this video. Very interesting subject. It reminds me of stuff I have been thinking about myself independently for years. Thank you Siraj, this video was great.
dope np
SIRAJ, i finished the Spanish subs in the video of the Intro, maybe i'll make the next tomorrow or today at the night.
Please, accept it and update the Google Drive Document with the script of this video when you can (:
Awesome vid bro!
Gracias Pinkie, i've accepted your subtitles. I've also added this script to that same document. I will continue adding every weeks script to that document and review any subtitle submissions you make. :)
Not all heroes were capes
whoa, you said "also useful for preventing overfitting" and I realized I needed to take notes.
I'm glad I stumbled upon your youtube videos sir.
Great Siraj. Keep it up.
I learned to appreciate math thanks to you @Siraj Raval, Thanks so much!
As a high performing Autist, it's very difficult to absorb your Humor. Danke
This video was so amazing! It was well explained, thanks Siraj! :D
Brother your descriptions are really really great!! keep it up.. Congrats
this is an awesome explanation of math we have learned in college thanks Siraj god bless
Did I just witness the best analogy for fit and overfit? Darn!
Liked the video first,then watched it! :)
Awesome vid Siraj. One question - at 8:56 when you show the 4 types of norms, is there a norm that you recommend developers should use most often? Or does it depend on the type of ML model you're building?
Hey Siraj Raval, how probable is it for abstract machine learning libraries to reduce demand for machine learning expertise to the point that the average person could not reasonably expect to gain employment in machine learning after investing the considerable effort it takes to understand the underlining math of machine learning?
For example in a previous video you showed a abstraction of tensorflow that allowed you to simply take a labeled data set f(Vi) = Yi. All training/learning was then abstracted by the library and and you simply input a new vector that was not labeled and and the library output its best guess of y's value or label based upon the data set you feed it.
Will machine learning EXPERTISE only be necessary for companies that develop machine learning libraries?
How long is the window for wide demand, not counting grandiose entrepreneurial endeavors?
This video is really powerful, very well put.
So how does matrix multiplication work programmaticly in numpy? I thought matrix multiplication is done the algebraic way in a loop like you showed in the algebra example and numpy is just faster because its doing its looping calculations in C. (4:52)
will explain this in-depth next week promise
I have to watch Siraj's videos several different times. Not to grasp concepts, but becuz I'm usually laughing the first couple of times thru
let me also add, the more I go thru his mathematics videos the more I despise my past mathematics teacher. I flunked highschool algebra, yet now I find myself using linear algebra to optimize learning programs. Thank You for the teaching, Siraj.
dope haha
Wow I'm early
Hey Siraj could you do a video on quantum computation please please please.
Also ur videos are Boss ur a Boss.
Just the basics of it.
Why quantum computing?
And perhaps an explanation of a simple code.
C.S.Bahushruth you don't write code for a quantum computer yet
thanks! hmm. after this series i will consider it
I really enjoyed the interpretive hand dance tensor music break.
thanks for saying that. i'll do more soon
Hi, I have a general query. Which algorithm is good to detect anomaly neural network or multi variable linear method ?
very good videos Siraj you are the best
WOAH. I didn't expect you to understand Einstein Field Equations! Dammit you gotta do a series on them too :3. You inspire me Siraj. You really really inspire me to keep learning. Thank you for existing.
thats my man keep it up
Siraj Raval Thank you man.
Amazing explanation ! Congratulations
I wonder where you learn you materials from. Research papers? Anyways, thanks for your videos.
Please help, I don't understand what is tensor.
Vector is invariant under coordinate system. I can use no coordinate system at all e.g. for moving particle under the known curve - I can measure the path length.
Q1: Is tensor depend on coordinate system?
Q2: I have rather good example of tensor of second order like intertia tensor of rigid body. I can use it to "exchange" non-free rotation vector "w" which I can move along a line into momemt of impulse(kinetic moment) via L=IW via using using usual rules of mulitplication. But what does it mean to be a more high dimensional tensor like 3?
In video it have been said that this just an array of numbers but to which ojects I can apply this imagine tensor of degree 3 and how I should do it?
How this numbers which are represented a tensor changes if use other coordinate system to describe my objects?
Thanks
you are a wonderful teacher
great video...please make more videos on maths...since it's the essence of machine learning..
u got it
This video is hilarious! Also, wow.
5:00 why do you compare their number of operations, when they aren't even the same thing?
Hey Siraj, please consider making a video on Object detection, the deep learning approach perharps with Fast-RCNN, YOLO, SSD or The tensorflow object detection API..please....
don't forget add this film to your playlist :)
btw... you are great! :D
thx just did
Zero norm at 08:59 is not a norm. Never was be, never will be. Some people which work in convex optimization called it caridnality. At lease because card([1 0]) = 1, but card(2*[1 0]) is also 1. It is not 2.
Thanks for making this video, it is awesome!
Vector is an array of 1-dimension, can pls elaborate this statement, I didn't get it?
it is great keep making videos!!!
Thank you for the explanation, but one thing I'm still not clear on is how does Word2Vec finds similar words from sentences. For eg.
1. Mercedes is a German legacy
2. Honda on the other hand is pretty new
3. But nothing beats Toyota reliability
In this context, how will Word2Vec get to cluster {Mercedes, Honda, Toyota} as one group - can you please help explain this. I do get calculating distance, vectorizing the position of each word and selecting the dimension randomly but how does Word2Vec works.
Thank you again Siraj from a software developer.
thank you for math and intelligence video
can we implement tensor concept in java ?
I am asking because Java includes "Vector class".
Awesome style
vectors into tesla, i like,
you are great.
thanks Tony
So, what is vector exactly?
Although it's just one word, I can see your Chinese mandarin pronunciation is quite good. :)
+Gai Wang Xie xie
2:09 that was epic"stop they get it".
awesome edutainment
I recommend gensim for word2vectors and numpy to make some operations, gensim it's just great for word2vector guys, really it's awesome.
Nice video ! Keep it up
hi can you answer this for me sir? 😶 " given the line x+3y+5=0 and a circle x^2+y^2-6x-4y+3=0 find the equation of the tangent to the circle which are parallel."
awesome is less for this video :D where can i find that matrix multiplication visual tool at 4:37and... you can't divide matrices :P
Would be great to see someone actually point out the mistakes in this video. I begin by pointing out 5:31. By reshaping your input you don't take it to a new vector space. An mXn dimensional vector space and a 1Xmn vector space are homeomorphic. Anyone want to add to this list?
Adding on as I watch, 6:00 word2vec was not a single neural net. It was an ensemble of models. And no it can't learn the embedding for a new word. There's no training/testing split there. There's only an embedding.
so cool your video!
what horoscope are you?
gemini
Just found Welch Labs on youtube as another spot for good youtube videos on introducing ml, you guys should collab
Thank you for the video
7:30 xD Oh, My God, I love your videos
Linear algebra, vectorization, word2vec, cost functions, gradient descent, TPUs, tensorflow, tensorboard, neural networks, vector norms, regularization, and feature vectors in 11 minutes!!
What you explained: 11 minutes of precious and resourceful information
What I got: :O + :P = 69
I feel you Bro... I feel you........
#NeverEnoughSIRAJ . It's almost like you already know what we need to hear.
w00t!
i've just done the exam in linear algebra and i find it here? no thanks (even if your video was much more spicy than my entire course)
AHAHAH fuck, I'm dying, you're so funny 😂😂😂
Right now I'm doing ANN course on Brilliant and you're videos are super helpful, thanks a lot!
thank you so much for this video. :)
great vid
5:00 the linear algebra way is 6 operations, while algebra way is 3 operations.
so Its 2 times SLOWER, not 3 times faster
I think the example compares an iteration pattern using algebra with processing arrays in say, Python. The %%time magic command will proved that list computation is faster.
@@bixbe_sglearn looks like I was wrong. do you by any chance if native python (not numpy) will use AVE and SIMD?
Hi Siraj, appreciate for your precise work. How can we learn machine learning without prior knowledge of Math in details?
Died laughing at 5:56 😂😂😂
thank you so much for this
HE IS MY 'RIVAL' WE HAVE EQUAL POPULARITY AND MY ADS ARE LIKE HIS WITH SIMILAR COVERAGE AND MONETARY RECOMPENSE, LIKE SCROUGE AND MCQUACK, IT'S A RACE TO THE FINISH WHERE WINNER TAKES ALL, THANK GOD
Why Python is used to do this? It isn't slower than other languages?
Luan Brito i think you have to take some time to read more about the topic. there are huge good things about python such as another lenguajes like C.
Eduardo Rasgado yeah, I am in the first period of mechatronic engineering and I only had a introduction on programming.
Siraj, you sound like a human. It's weird. XD Really though your pacing is perfect.
Also, what's that music at 3:02?
ua-cam.com/video/JWnX41TBFF4/v-deo.html
Excellent but maybe if this was broken into 2 vids of 10 mins each lots of concepts would be more clearly explained.
What was the song at 3:10? Great video!
A new error from Moderat
Thank you!
Crap you are convincing me to take the two vector calculus courses at my college. In addition to the multivariable course.
Okay so which foundational papers in the field should I read and which major books for class, for amateurs and professionals.
Off to math on Reddit, Quara, and mathexchange forums.
Awesome man.., (y)
i came here to feed my hunger , my head is full , And i'm laughing without sound
I died laughing at I see vectors
3:10-3:30 cracks me up
Checkout Siraj's new series of mixtapes:
The Math of Intelligence
The Math of Intelligence 2
The Math of Intelligence 3
u know what it is, my courses are like albums
What is the source of your memes?
I literally just googled using non-numerical inputs in a neural network last night
i read ur mind
Hey, I find your memes to be distracting. The one with spongbob at :50 didn't really carry the point you were making. Don't get me wrong, I love that you use memes, but I feel they should be included only to solidify the point rather than a quick joke - or after your point is made so it's not distracting.
Hey SIraj what's your secret, how are you learning so fast?
he has a video explaining how to speed learning.
09:54 - I think it false description in that sense that you're overfit and etc. what is real going on you solve bicreterion optimization problem. For more details what is regularization please take a look into convex optimization books, such as this web.stanford.edu/~boyd/cvxbook/ p.s. Thanks for video! It is still great!
How do you cross a mountain climber and a mosquito? Not at all, because it doesn't work with scalars and vectors.
you speak mandarin?
I like it
1:20 If a hearth rate of 70 is consider healthy, I'm not sure how I'm alive. I have neither a fireplace nor a hearth.
You pronounce Beijing perfectly