I never thought of start learning AI as I was scared or confused. Now, after watching your videos, I think I am at the right direction. Thank you to avail these rich contents for free.🙏
I found myself very lucky that I found your channel. So awesome explantaion and especially the way you are mapping it to the machine learning concepts. Thank you so much. I will share your channel In my network.
@@JonKrohnLearns is your tutorial leading us beginners to be a machine learning engineer? I want to be one. :), Are this foundations gonna help me know more of the making of the machine learning model.
Thank you for these nice explanations I learned a lot from these videos, more than in schools Thank you for the hard work behind presentations and code demos also, love your weightlifting
At time 6:38 secs you talk about unit vectors and the vector x has unit norm of 1. However, the co-ordinates of x are (1,1), so the unit norm in this case would be sqrt(2). Can you explain how the unit norm is 1 ?
Sure! Both of these bullets are easy to understand when contrasted with the L2 Norm. The L2 Norm, because it has a square, is inherently NON-linear (unlike L1) meaning that L2 starts starts to vary a LOT as we move away from zero whereas the L1 Norm continues to vary at the same rate as did near zero.
Thanks Jon for the wonderful videos. I had few queries: 1. You gave few points explaining how L2 norm is computationally cheaper. I didn't get the reasons. You will be explaining them further later or as they are not used much in ML you are gonna skip it? 2. For L1 norm slide you told it is used whenever the difference between 0 and non 0 is key. Any examples where this will be used? Just to get a practical feel of it.
Glad you're enjoying the videos, Abhi! W.r.t. Q1, do you mean that **Squared** L2 Norm is computationally cheaper? If so, I explain why at 9:17. W.r.t. Q2, this is beyond the scope of my "ML Foundations" series because it's a big ML-specific topic in its own right. If you'd like to learn more, here's a good Stack Exchange post: stats.stackexchange.com/questions/45643/why-l1-norm-for-sparse-models
from the preceding video, from my POV, array is a 1-D space if it is just ( [ ] ), but if it has an extra bracket / ( [ [ ] ] ) then it becomes a 2-D space
thanks for the great course John. Is there any additional good references on Norms you can recommend. I understand at high level what they are, however still wondering why they are derived in first place and more details on how the different types of norms are used. Cheers
Great question! Now that you understand what norms are from this video, you can apply them whenever you encounter them in a machine learning textbook, e.g., this classic on the "deep learning" subfield of ML: deeplearningbook.org
aha i found out your channel through one of my utube subscription known as "ishan sharama" than i jst hoped into your channel Jon Sir ! and i am so gald that u done this complete course for free in utube !! I Love the way u Explain Jon Krohn Sir so simple and easy to understand !! and for instance i am making 🗒 notes also with it ! Thanku for this knowledge 😄🤝!! i have subscribed to ur channel and aslo goona complete the course 🎯. and also i am goona teach it to my friends !! , lots of love from India Jon Krohn sir.
Hi Jon. Thanks for the videos. I'm really enjoying the way you are presenting the material - very clear, and unrushed. I'm a little confused though in this particular one with the diagram of the unit vector. Wouldn't the vector be [1,1] based on how it's in the diagram? If that's the case, then how do we get to ||x|| = 1? I must be misinterpreting something along the way. Thank you.
Ah, the vector does NOT stretch all the way to [1,1], but I should have made that more obvious in the figure. The point truly is downward and to the left relative to [1,1], but it's difficult to see. I'll ensure this is more clear in the ML Foundations book I'm writing - thank you for the feedback, Alex!
since you are demonstrating both PyTorch and TensorFlow, can they be used for different types of machine learning problems? like dealing with images, text, sound , tabular data etc. Is one better than the other for certain types of applications?
They're not the "same" exactly but they are related to regularization. We need to calculate the L1 norm in order to implement L1 regularization, which is itself a part of Lasso regression (as well as being a part of other optimization approaches). Likewise, we need to calculate the L2 norm in order to implement L2 regularization, which is itself a part of Ridge regression (as well as being a part of other optimization approaches).
Wish I had taken these courses before enrolling to an ML class, where I was so lost, lol. Now I will have to do a recovery exam! But I will ace it in God's will!!
Yes! Which is a fun way to describe L1 distance. If you want to travel from Point A to Point B on much of Manhattan, you can only move North-South or East-West along the avenues and streets, respectively, so the distance you drive or walk from Point A to Point B is the L1 distance :)
Kinda, but not exactly. The Squared L2 Norm of some vector x is a dot product of the x vector with itself. If we take the dot product of two different vectors, that is not a Squared L2 Norm.
@@JonKrohnLearns Thanks for clarification. Is L1 norm also known as Manhattan Distance. If yes, why so. I understand L2 is also known as Euclidian distance because of Mathematician Euclid. Why we some many types of Norms?
@@elp09bm1 I appreciate all of the comments, but I feel like some of your questions could be answered by a quick Google search :) Here's a blog post on Manhattan (L1) vs Euclidean (L2) distance: towardsdatascience.com/3-distances-that-every-data-scientist-should-know-59d864e5030a Norms come in handy in lots of places in machine learning, including for model regularization: www.analyticssteps.com/blogs/l2-and-l1-regularization-machine-learning
Coming from non-math/cs background, can't even tell how these videos are helping me.. i was so hopeless and lost then i found this, am so grateful
I never thought of start learning AI as I was scared or confused. Now, after watching your videos, I think I am at the right direction. Thank you to avail these rich contents for free.🙏
I found myself very lucky that I found your channel.
So awesome explantaion and especially the way you are mapping it to the machine learning concepts.
Thank you so much.
I will share your channel In my network.
I'm lucky you found my channel too, Aashish! Glad you're enjoying the content :)
I love how you explain in an intuitive way and then showcase examples.
Love the explanations that are immediately followed by code examples
Heck yes! I'm glad you do because that is definitely my modus operandi for video tutorials :)
@@JonKrohnLearns is your tutorial leading us beginners to be a machine learning engineer? I want to be one. :), Are this foundations gonna help me know more of the making of the machine learning model.
Thank you for these nice explanations
I learned a lot from these videos, more than in schools
Thank you for the hard work behind presentations and code demos
also, love your weightlifting
haha glad you're enjoying the weightlifting videos while also finding value in my tutorials! It means a lot to hear this feedback, thank you :)
Thanks for one more excellent video. I am practising them as I watch.
At time 6:38 secs you talk about unit vectors and the vector x has unit norm of 1. However, the co-ordinates of x are (1,1), so the unit norm in this case would be sqrt(2). Can you explain how the unit norm is 1 ?
7:41 Could you explain more about the second and third bullet point??
Sure! Both of these bullets are easy to understand when contrasted with the L2 Norm. The L2 Norm, because it has a square, is inherently NON-linear (unlike L1) meaning that L2 starts starts to vary a LOT as we move away from zero whereas the L1 Norm continues to vary at the same rate as did near zero.
Thanks Jon for the wonderful videos.
I had few queries:
1. You gave few points explaining how L2 norm is computationally cheaper. I didn't get the reasons. You will be explaining them further later or as they are not used much in ML you are gonna skip it?
2. For L1 norm slide you told it is used whenever the difference between 0 and non 0 is key. Any examples where this will be used? Just to get a practical feel of it.
Glad you're enjoying the videos, Abhi!
W.r.t. Q1, do you mean that **Squared** L2 Norm is computationally cheaper? If so, I explain why at 9:17.
W.r.t. Q2, this is beyond the scope of my "ML Foundations" series because it's a big ML-specific topic in its own right. If you'd like to learn more, here's a good Stack Exchange post: stats.stackexchange.com/questions/45643/why-l1-norm-for-sparse-models
Thanks @@JonKrohnLearns
@@JonKrohnLearns How can I continue after I finished the foundation
Hi,earlier in video,you said that, array are 1 dimensional,but in video at 5:16 ,you said the array x is in 3 dimensional space.
Can you explain that?
from the preceding video, from my POV, array is a 1-D space if it is just ( [ ] ), but if it has an extra bracket / ( [ [ ] ] ) then it becomes a 2-D space
as I've watched he also gave the notes about
scalar = 0-D = magnitude only
vector = 1-D = array
matrix = 2-D = flat table, e.g., square
3-tensor = 3-D = 3D table, e.g., cube
n-tensor = n-D = higher dimensional
Hope these help, :)
thank you so much for your help, can we get those slides through which you are teaching us. thank you
Certainly! Head to jonkrohn.com/talks and search for "linear algebra" on the page
@@JonKrohnLearns thank you so much sir.
@@EjazAhmed-pf5tz you're welcome!
Sir can I plz get the pdf notes for your lectures 🙏
....... This lectures are extremely helpful to me🔥🔥
Thanks a lot for this amazing videos! Grettings ts from Argentina
I really like your video , I learn a lot from them
thanks for the great course John. Is there any additional good references on Norms you can recommend.
I understand at high level what they are, however still wondering why they are derived in first place and more details on how the different types of norms are used. Cheers
Great question! Now that you understand what norms are from this video, you can apply them whenever you encounter them in a machine learning textbook, e.g., this classic on the "deep learning" subfield of ML: deeplearningbook.org
Dude, you're AWESOME! 🔥
Thanks, Kareem! YOU'RE awesome :D
aha i found out your channel through one of my utube subscription known as "ishan sharama" than i jst hoped into your channel Jon Sir ! and i am so gald that u done this complete course for free in utube !! I Love the way u Explain Jon Krohn Sir so simple and easy to understand !! and for instance i am making 🗒 notes also with it ! Thanku for this knowledge 😄🤝!! i have subscribed to ur channel and aslo goona complete the course 🎯. and also i am goona teach it to my friends !! , lots of love from India Jon Krohn sir.
Just wonderful, thank you so much for these videos, great explanation
You're very welcome!
Jon, can we get that presentation file?
loved your videos❤❤❤❤❤❤
I love that you love 'em!
thanks for the knowledge!
Hi Jon. Thanks for the videos. I'm really enjoying the way you are presenting the material - very clear, and unrushed.
I'm a little confused though in this particular one with the diagram of the unit vector. Wouldn't the vector be [1,1] based on how it's in the diagram? If that's the case, then how do we get to ||x|| = 1? I must be misinterpreting something along the way. Thank you.
Ah, the vector does NOT stretch all the way to [1,1], but I should have made that more obvious in the figure. The point truly is downward and to the left relative to [1,1], but it's difficult to see. I'll ensure this is more clear in the ML Foundations book I'm writing - thank you for the feedback, Alex!
@@JonKrohnLearns Thank you for the clarification.
since you are demonstrating both PyTorch and TensorFlow, can they be used for different types of machine learning problems? like dealing with images, text, sound , tabular data etc. Is one better than the other for certain types of applications?
I summarize the key relative strengths of PyTorch and TensorFlow here: ua-cam.com/video/9hT5AKAI924/v-deo.html
@@JonKrohnLearns Thanks so much! :)
Great job
Is L1 and L2 norm are same as ridge and lasso regularization?
They're not the "same" exactly but they are related to regularization.
We need to calculate the L1 norm in order to implement L1 regularization, which is itself a part of Lasso regression (as well as being a part of other optimization approaches).
Likewise, we need to calculate the L2 norm in order to implement L2 regularization, which is itself a part of Ridge regression (as well as being a part of other optimization approaches).
I hope I will complete the series
Cheering for you, Manajit!
superb
:D
Wish I had taken these courses before enrolling to an ML class, where I was so lost, lol. Now I will have to do a recovery exam! But I will ace it in God's will!!
Damn, well hope these videos help!! Glad you've found them :)
thanks
You're welcome, Shamil!
@@JonKrohnLearns when will there be a sequel?
@@Soul.game. haha, a sequel to what?
@@JonKrohnLearns
this course ua-cam.com/video/Qc19jQWHdL0/v-deo.html)))
I learnt that L1 norm is also called Manhattan distance😅
Yes! Which is a fun way to describe L1 distance. If you want to travel from Point A to Point B on much of Manhattan, you can only move North-South or East-West along the avenues and streets, respectively, so the distance you drive or walk from Point A to Point B is the L1 distance :)
So we can say Squared L2 norm is the dot product of two vectors.
Kinda, but not exactly. The Squared L2 Norm of some vector x is a dot product of the x vector with itself. If we take the dot product of two different vectors, that is not a Squared L2 Norm.
@@JonKrohnLearns Thanks for clarification. Is L1 norm also known as Manhattan Distance. If yes, why so. I understand L2 is also known as Euclidian distance because of Mathematician Euclid. Why we some many types of Norms?
@@elp09bm1 I appreciate all of the comments, but I feel like some of your questions could be answered by a quick Google search :)
Here's a blog post on Manhattan (L1) vs Euclidean (L2) distance: towardsdatascience.com/3-distances-that-every-data-scientist-should-know-59d864e5030a
Norms come in handy in lots of places in machine learning, including for model regularization: www.analyticssteps.com/blogs/l2-and-l1-regularization-machine-learning
So. L2 norm is just Pythagoras theorem
haha yes the arithmetic is similar
thanks