This is really simplified. Greatly appreciated. Much better than those university professors who get obsessed in the math without showing the audience the big picture !!
Beautiful, being a professor myself I feel often people fail to focus on the big picture ; it is like we are thinking the local minima as a global minima 😅
I am new to deep learning. The content you provide helps me understand the concept from the very basic and this clarity i could not find in any videos. Keep up the good work!!!!!You are doing a great job.
Great Video Sir. I haven't seen such a simple explanation of such a brainstorming concept. Many people just take days to understand Backpropagation and here you have cleared my concepts withing 10 minutes.
Sir I don't know why but whenever I watch your videos I get motivated to go in the more depth of that particular topic.Thank u sir sharing such valuable content for free .
Krish sir you are really amazing! You have made difficult concepts simpler! Our teachers are not able to do this, but you have done it very well! Thanks for being with us in this learning journey!
Sir there is one thing for sure to tell u, your way of teaching is so relatable and hence easy to understand, Also very effective, I've went through so many tutorials and lectures but it is all making sense now in your video, I m very thankful to u sir keep teaching,. u r a great teacher.
I think that , during the probleme discussed about pass or fail i.e. binary classification the loss function should not MSE . It should be cross entropy loss i.e. -ylog(y)+(1-y)log(1-y)
Thank you Mr.Krish, you just nailed it. I have a question here, how to find loss at any hidden neuron for backpropagation purpose since we will not be knowing the actual value at any hidden neuron except at the output node, right? then how? Thanks in advance.
We are reducing the weights gradually at the hidden neuron so that indirectly it reduces the loss/ cost function at the output node. We dont calculate loss at hidden neuron
Could you please clarify my following doubts? Thanks in advance. 1. Are you using any affine function and/or activation function in the output layer node in order to calculate y_hat? Reason being, weight W4 is passed as input to the output layer node and no details are mentioned about the usual two step process that take place in every node of the neural network i.e. an affine function (where weights are actually used) and an activation function. 2. Is it or is it not the cost function is average of the loss function for individual training samples? Cost function is defined as summation of loss function in this video and not average? 3. I'm not clear on why propagation helps in tuning the model parameters. Back propagation and Gradient descent work together in tuning the weights. Mathematically and geometrically I'm not convinced with the statement "back propagation is used to train the weights of a neural network".
Actually, I think all of y will pass through all neurons. Therefore, the final output will be calculated over the y_hat = w4*(y_prev). When we include the biases in this calculation, the output from a neuron in a layer will be y_prev = (W1*X + biases) -> y_nextneuron = (W2*y_prev + biases).
Btw how the loss function has defined.like lets suppose u have given a random weight first and depending on that u have get a average lost value for all the train sample.so for a specific set of weight u r getting a specific lost value than how u r getting a function.cz u have just got a specific output for a specific input.it Doesn't make anysense about the costfunction
Sir, thanks for providing wonderful videos. Sir if possible could you provide a basic explanation for construction of neural networks in equation format (in first order differential equations) including state vector, activation functions. Thank you sir
Thanks sir. I found your tutorials interesting. I have a question ... what is global minimum and Gradient descent the words that you have used in your lecture. Can you please elobrate. If any video related to thz is available kindly share the link. Thank you
Hi Krish.... I am a regular learner from your videos, this is great. I have a question, in the forward propagation for the very first iteration where and how do we get the values of weights?
Krish, I think it would have been better if you would have taken , Regression example and explained it , Coz the loss function you are showing over here is squared error, which is not the correct loss function for a classification problem, rather binary cross entropy is . May be you can rectify it. :). Even the cost function , You forgot to divide it by number of samples .
This is really simplified. Greatly appreciated. Much better than those university professors who get obsessed in the math without showing the audience the big picture !!
Yup! They uses lot of symbols... Sometimes even its hard to remember those symbols
can't agree more!
Beautiful, being a professor myself I feel often people fail to focus on the big picture ; it is like we are thinking the local minima as a global minima 😅
I can't agree more !! I was trying to do an expensive course, but got back here. It feels like senior padha rahe hai
you just nailed it. Simplicity of his explination is unmatched anywhere. Thanks you sir.
I am new to deep learning. The content you provide helps me understand the concept from the very basic and this clarity i could not find in any videos. Keep up the good work!!!!!You are doing a great job.
The best explanation on NN I've ever seen so far. Thanks for going ahead step by step and explaining in simple words
I have taken courses in other platforms, but I must say, the simplicity I found in your explanation helps me grab the topic much easily🙏🙏🙏
This channel is a discovery...you have been able to cut through all the jargon and make these theories less abstract. Thank you very much
Sir I had watch more than 10 videos but didn't understand now in 9 mint I had understood very well. Simply awesome!!!!!!!!!!!
Great Video Sir. I haven't seen such a simple explanation of such a brainstorming concept. Many people just take days to understand Backpropagation and here you have cleared my concepts withing 10 minutes.
You are genius!!! No other instructor can teach in 10 minutes this complicated concept👏👏👏 Thank you🙏
Sir I don't know why but whenever I watch your videos I get motivated to go in the more depth of that particular topic.Thank u sir sharing such valuable content for free .
Krish sir you are really amazing! You have made difficult concepts simpler! Our teachers are not able to do this, but you have done it very well! Thanks for being with us in this learning journey!
Finally the stuff I had been looking for. Simple and to the point.
finally found a video, that helped me understand this topic. its relief + satisfaction
Clearly the best DL and ML teacher!
Sir there is one thing for sure to tell u, your way of teaching is so relatable and hence easy to understand, Also very effective, I've went through so many tutorials and lectures but it is all making sense now in your video, I m very thankful to u sir keep teaching,. u r a great teacher.
magnificent explanation the simplicity is perfect hard is made easy with you thanks!
Sir your videos are much more better than coursera courses...........Thank You.
Your are great.
The way you explains, anyone can understand.
Thank you.
Awesome playlist. Thank you for sharing your knowledge 😊🤟
Underrated content!
Keep up the good work! 💯
Mind blowing video sir this is the main difference between you and others other institutions run behind the money
This course is simple and clear than most of the courses out there.
Way you have explain complex topic in such simple way ......
The best video so far. So clear and crisp. Hatsoff sir!!
Your style of explanation is helping me a lot - thanks for these videos!
you have explained it in the simplest way ,Thank you !
Best tutorial about deep learning ever!! Thank you so much for making it easy to understand! You are very much appreciated!
bro all these videos simply tells how deeply you have the knowledge of deep learning
khatrnak sikhata he ye banda...... ek bar dekh lo pure dhyan se fir nind mai bhi koi puch le no issue ....chhap jata h dimag mai ......
so far you are the best teacher.
this guy explained optimzation,backpropogation and learning rates all in few mins
Thank you Sir. Amazing.!!! Derivative: Ratio of change of dependent variable w.r.t independent variable
thank you for making this video! hope I can get a good grade in next week exam on AI
hats off to you sir,Your explanation is top level, THnak you so much for guiding us...
Wow man, you’re a blessing. Thank you for this great teaching, you simplified everything.
I came here after watching Coursera course and I think it's more clear and magnificent 🤤
Exactly
Really
hi krish , At 3:30 you said we can do square to make it +vs , we can also use |mod| to make it +ve.
How is everything makes sense? Wow, so inspiring! Amazing.
thank you so much sir, your way to simplify this problem is very nice.👍👍👍👍👍👍thanks again sir.
I have paid fortunes to do masters in data science and ended up watching your videos. You deserve a nobel prize for all these videos.
Amazing work, sir gi, love it, And Thank you for such a concise and understandable explanation.
perfectly explained !!! Simple and to the point !... Kudos ....
best trainer i have ever seen....
You are what I needed. Thank You soooo much :)
Thank you sir for clearing all the doubts by this video
@Krish Naik
How methodical your work is! Brilliant! Keep it up. Your videos clarify things lucidly.
Krish thanks for work.
Tutorial 4 supposed to be Activation function part 2.
Nice Explanation, I like the way you teach.
This was amazing. Thankyou Kris. Thankyou for existing.
Kris 🤡
Excellent, you are our confidence!
Hi Krish, What are the ideal Learning rates that need to be used? How do we decide which Learning rate value is ideal for a Neural network?
Yes it will be mentioned in the upcoming videos
The things taught are well understood.thank you sar🥰🥰
I think that , during the probleme discussed about pass or fail i.e. binary classification the loss function should not MSE . It should be cross entropy loss i.e. -ylog(y)+(1-y)log(1-y)
Damn good explanation .... One question, how we choose learning rate??
Amazing....great job with lot of thanks....
Well done bro, simple and precise
Awesome explanation sir. Thank you for sharing your knowledge
(1):why we use loss function
(2): how should we know that the loss value is minimal or increased
(3): where should we find the learning rate?
Thank you, Sir, for your sharing, with perfect explanations.
you make this topic seem so easy!!!!! Thank you!
what an Explanation !!!! amazing
the man, the myth, the legend
Excellent explanation !!! One question : what happens to biases in backward propagation ?
Simple et concis, je vous remercie pour l'explication :) :)
You deserve 1M like
Simply great 🙏
Thank you Sir for your great efforts
That was a superb video.But now things are getting tougher and tougher.Need to cope up with.
Mahn. You are too good
Wonderful video. Just one question, why we are not taking the mod value of the Loss function and going for squared values?
Awesome explanation 👌
@Krish Naik As this is a classification problem, how can it use squared error as a loss function?
Thank you, Sir, May you please make videos on Unsupervised ML algorithms ?
Thank you Mr.Krish, you just nailed it.
I have a question here, how to find loss at any hidden neuron for backpropagation purpose since we will not be knowing the actual value at any hidden neuron except at the output node, right? then how?
Thanks in advance.
We are reducing the weights gradually at the hidden neuron so that indirectly it reduces the loss/ cost function at the output node. We dont calculate loss at hidden neuron
Could you please clarify my following doubts? Thanks in advance.
1. Are you using any affine function and/or activation function in the output layer node in order to calculate y_hat? Reason being, weight W4 is passed as input to the output layer node and no details are mentioned about the usual two step process that take place in every node of the neural network i.e. an affine function (where weights are actually used) and an activation function.
2. Is it or is it not the cost function is average of the loss function for individual training samples? Cost function is defined as summation of loss function in this video and not average?
3. I'm not clear on why propagation helps in tuning the model parameters. Back propagation and Gradient descent work together in tuning the weights. Mathematically and geometrically I'm not convinced with the statement "back propagation is used to train the weights of a neural network".
Actually, I think all of y will pass through all neurons. Therefore, the final output will be calculated over the y_hat = w4*(y_prev). When we include the biases in this calculation, the output from a neuron in a layer will be y_prev = (W1*X + biases) -> y_nextneuron = (W2*y_prev + biases).
Literally great explanation brother.
thank you very much sir!! u gained a subscriber.
Thank you for this sir ,tmrw I am having an interview
maza aa gaya...awesome
Nice video. Thank you sir.
Btw how the loss function has defined.like lets suppose u have given a random weight first and depending on that u have get a average lost value for all the train sample.so for a specific set
of weight u r getting a specific lost value than how u r getting a function.cz u have just got a specific output for a specific input.it Doesn't make anysense about the costfunction
Does the bias associated with the neuron in the hidden layer also need to be updated during backpropagation?
Sir, thanks for providing wonderful videos. Sir if possible could you provide a basic explanation for construction of neural networks in equation format (in first order differential equations) including state vector, activation functions.
Thank you sir
Thanks sir. I found your tutorials interesting. I have a question ... what is global minimum and Gradient descent the words that you have used in your lecture. Can you please elobrate. If any video related to thz is available kindly share the link. Thank you
Watch next 2 videos you will get to know.
Sir could you please explain back propagation with an example, It would be of great help as that's the part where most of us make mistake
Thanks for turning off the fan! :D
Your lectures is amazing and very helpful but you looks so serious in every video
thanks sir but I have one question how can I get w1,w2,w3.... from the beginning?
Well explained... So back prop and fwd prop both happens in 1 epoch at the same time?
Sir do we need to update bais when we do backward propagation
pretty great explanation
simply the best
Hi Krish.... I am a regular learner from your videos, this is great.
I have a question, in the forward propagation for the very first iteration where and how do we get the values of weights?
There are some weight initialization techniques. Just go ahead with the videos u will be see the video
@@krishnaik06 thanks for the prompt response... I will follow the videos.
In classification the loss function should be different like log loss here sir you use regression loss function please correct if I am wrong
GD vs SGD vs PSO vs GA ?
please give the efficiencies of these optimizers?
Krish, I think it would have been better if you would have taken , Regression example and explained it , Coz the loss function you are showing over here is squared error, which is not the correct loss function for a classification problem, rather binary cross entropy is . May be you can rectify it. :).
Even the cost function , You forgot to divide it by number of samples .
Kish can you explain bias as well because I believe, we can adjust bias as well as a measure of back propagation. Pls guide around it
Yes bias is also a parameter that undergoes back propagation
@@MegaAntimason do you have any references around it
Sorry. . Forgot to say hi.. .
@@hokapokas datascience.stackexchange.com/questions/20139/gradients-for-bias-terms-in-backpropagation
@@MegaAntimason thanks 🙏🙇🙏💕
excellent sir.sir what is learning rate
How do I get access to Tutorial 2 of Tutorial 1- Introduction to Neural Network and Deep Learning
Hi sir, what is the reason behind derivating loss with respect to weights. with that what we will get?