@@derroz3157 thnks to thanking guy to me you thanking him/her thnks for thanking i can't thanking you for too much thankful but thanks to say thanks to a person who is thanking MIT from there side I thank you both to be thanking them,
They're making basics available which r available in courser and udemy.. do u think they're putting vidoes of advanced stuff that their students actually learn (and which distinguishes them)?
By "YouSum Live" 00:00:10 Introduction to MIT course on deep learning 00:00:41 Evolution of AI and deep learning 00:02:56 Realism and virality of AI-generated content 00:04:15 Accessibility and cost-effectiveness of deep learning 00:04:19 Advancements in deep learning applications 00:05:11 Empowering deep learning models to create software 00:06:02 Teaching foundations of deep learning 00:07:30 Importance of understanding intelligence and AI 00:14:00 Transition from hand-engineered features to deep learning 00:16:00 Significance of data, compute power, and software in deep learning 00:17:24 Fundamentals of a neural network: the perceptron 00:19:59 Mathematical representation of a perceptron 00:20:51 Activation function and its role in neural networks 00:20:57 Importance of activation functions in neural networks 00:21:11 Sigmoid function: squashes inputs into probabilities 00:23:01 Need for nonlinearity in neural networks 00:23:32 Linear functions insufficient for handling nonlinear data 00:24:22 Nonlinearities enhance neural network expressiveness 00:26:01 Visualizing neural network's decision-making process 00:27:33 Sigmoid function divides space for classification 00:28:16 Understanding feature space in neural networks 00:29:21 Building neural networks step by step 00:31:41 Perceptron's fundamental equation: dot product, bias, nonlinearity 00:32:49 Defining layers and passing information in neural networks 00:37:19 Cascading layers to create deep neural networks 00:38:18 Applying neural networks to real-world problems 00:40:38 Neural network training process explained 00:40:50 Neural networks learn like babies, need data 00:41:12 Teaching neural network to make correct decisions 00:41:32 Importance of minimizing loss for accurate models 00:41:55 Training neural network with data from multiple students 00:42:21 Finding network that minimizes empirical loss 00:42:40 Using softmax function for binary classification 00:43:27 Loss function for real-valued outputs 00:47:56 Gradient descent for optimizing neural network weights 00:59:51 Introduction to gradient descent algorithms 01:00:11 Stochastic gradient descent (SGD) explained 01:00:45 Importance of mini-batch gradient descent 01:01:37 Faster convergence with mini-batches 01:02:03 Parallelization benefits of mini-batches 01:02:30 Understanding overfitting in machine learning 01:04:41 Regularization techniques: Dropout and early stopping 01:06:56 Monitoring training and testing accuracy 01:08:48 Summary of key points in neural network fundamentals By "YouSum Live"
I am a high school student and I am currently self-studying deep learning and I find it very helpful. I hope one day I can attend your lectures in person. Thank you very much.
@@lakshyajain6765 don't need to since every semester course is self contained unit. This is not created for UA-cam, it's for MIT students and every semester there is new batch.
@@RadixSort3 Thanks a LOT!!! this is my alt. Do you have any idea on some more MIT ML related lectures. I would like to do some research in this field and try to get into a phd program
After being in college for 4 years and dealing with loads of professors, I can hands down say this guy is the best lecturer I've ever seen! Explains tough concepts so well.
I want to take this moment to thank UA-cam, MIT and Alexander Amini for suppling this content 4 a person like me who is studding deep learning but was not fortunate enough to study in MIT🙏🙏
Attended Deep Learning lectures at a topmost college of a country, here he clearly explained all that in a single lecture for which the former took 10s of lectures to explain.
This is not for beginners. Having 3+ years of experience in deep learning i found it interesting on how much information is shoved into 1 single video . Note that each concept is very vast if we dig deeper
I'm in my mid 50s now, and I keep telling my kids this same thing. When I was their age, the Internet wasn't available yet. Information was very hard to come by, let alone quality courses like these. Now, it's all at their/our fingertips! A great time indeed.
Yesterday we started system identification using neural network, I watched your lecture and now I feel quite comfortable using the concept of deep learning. Thank you Sir and love from Pakistan....
I look at these videos every year after the new annual release and it just never gets old. Too bad in my work, I don't get a chance to apply this knowledge. It is still super fun to watch, like a fun show to me
WOW!!!!😍My professor is chinese and I know he knows a alot of things but after watching this teacher teachin, I understood the importance of a good presentation and most importantly, what a good presentation look like.
@@HeyMr.OO7 What do you mean by low level in your definition? It is as low level as you can get in this field that you can perform calculations on an entire network by hands without having to rely on computers, not to mention programming languages or libraries. Some data scientists or self-taught professionals I have talked to who are fluent in machine learning tools which are considered high levels do not quite completely understand this low level fundamental and I doubt if they could hand calculate an entire network from scratch.
I've been attending these classes from IIM-C. You've summarized 6 hours of my professors' class in just an hour. I'm coming here right on the day of exam to revise everything Thanks MIT❤
Sir you are doing a great job, I am student of BSCS, last year from Pakistan. But being a student to learn Deep Learning from last 2 year, I am still a beginner, as the system is not very modern. This lecture seems like a new start for me, which feels very promising. Can you please share the other lectures, so I (students like me) can really advance in this field, and maybe start working at MIT someday. Thanks for teaching in such a beutifull way.
@@webgpu ofline lectures? I guess you meant to say, "What are the advantages of following live online lectures over recorded online lectures? Did I get the question correctly?
@@page002 sorry I most probably was not able to express myself properly. I meant "what are the advantages of the [opposite of live] lectures - so I think that's what you also meant in your past comment 👍
@@webgpu don't worry. So here's my point - I prefer Live Recorded lectures over only recorded lectures because when we follow live I think we can connect more with the instructors. Also it gives us the impression that we are also a part of it which a recorded and already published can never give(at least that's what I think). And last but not the least if we follow the live (recorded) lectures here we will have a clear goal and a Dateline to follow. And I think that's a great thing. So, any day I prefer Live Recorded lectures or Live lectures if possible over recorded lectures specially for technical things and programming. I am a pretty bad communicator so, I hope you got your answer even a little. BTW, if you don't mind, try to follow Live lectures once I think you will be able to see the difference personally. Happy Learning
It's nice having up to date lessons on this stuff considering how fast it moves, even if a good amount of the core content presumably largely stays the same.
I did this stuff 8 years ago in uni - felt like the deep learning stuff was too "stochastic" 😉 and avoided it. Looking to try this out again now without going through all the new research material, and this video has been great - just the right amount of detail!
Both theory and actual implementation in industry code! Perfect! Also, great pacing and depth! After 5 minutes in one episode, and i can already tell this is the best beginner ai lecture series I have seen!
@@mehulnakra2457 this is much harder than DLS for sure, I have studied both. But MLS and DLS from Andrew Ng give u a broad view of ML and DL, so if u are studying these courses, keep studying
How fascinating is it i wanted to learn about neural networks and just searched neural networks mit and found a course thankyou so much youtube and MIT.
I'm a beginner in ml and ai fields and it's amazing to have these lectures online and free. I've a doubt: the neural network showed in 33:44 shouldn't be named 'multi' layer rather than 'single' layer neural network since it has an output layer separated of the hidden layer? Thanks!
Thanks for the lecture, please please make a video or provide a pdf of MATH too, I wanna know the math behind deep learning, svms, pca, ML in general aka grad descent etc, how then that changes when many layers are involved (as in deep learning) so basically normal ML -> i/p -> mat mul -> o/p deep learning -> i/p -> mat mul = linear x matrix . non linear x matrix . linear or non linear x matrix ..... -> o/p etc etc etc I mean try and simplify what goes on mathematically then also give enough formalization that some of us can begin to understand a few of the key ML papers on Arxiv. This has been our biggest challenge truly.
At 22:45 you mention the ReLU function has a discontinuity at '0', IIUC this is not true, ReLU is a continuous function, even at '0'. It is however not differentiable at '0'.
There was a young lady named bright, who could travel much faster than light, she set off one day, in a relative way, and returned the preceding night.
It's wonderful to listen to the lectures of the MIT professors I hope one day i will attend it on person if god allows! I also want to thank UA-cam, MIT and Alexander Amini for this wonderful work, due to their efforts, we students from all over the world can reach the lecutres of worlds most renowned institutions! Thanks for ur efforts again
This video is interesting because,this video helps me understand the current price and prediction of Palantir stock. The analyst explains incredibly. Thank you for sharing this valuable information.
Really like the fact that everything is explained so simply and in a way that is digestible for most people. Personally I found it a great video for revision and brushing up concepts that build ML.
studying AI&Robotics at the new UTN and must say your videos are life-saving! Thanks for sharing your lectures. I really like how well and easy everything is explained. Really learned a lot!
I did this in 2005 in assembler on the development KIT board ADSP-2189. The learning parameter = gradient could be computed after certain epochs. The algorithm was able to compute each weight very easily, so that weach W(i) was tested and the smaller error the better weight vector was taken into the next epoch. For instance the algorithm tested the W(i) within the range -1 to +1, so it started with -0.5, then 0, then 0.5, and divided each range by two on 16 bit numbers.
00:04 Introduction to MIT 6.S191 - Deep Learning Course 03:02 AI's realism in generating hyperrealistic content 07:36 Teaching machines to process data and inform decision-making abilities 10:02 Introduction to foundations of neural networks and upcoming guest lectures 14:22 Introduction to Deep Learning Paradigm Shift 16:38 GPUs and open source tools drive deep learning advancements 20:42 Different types of nonlinear activation functions in neural networks 22:32 ReLU activation function introduces nonlinearity in neural networks 26:30 Sigmoid function divides space based on input value 28:25 Neural networks have millions or billions of parameters, making visualization challenging 32:26 Building a single layer neural network is simple and modern deep learning libraries provide tools to easily implement it. 34:32 Introduction to two-layered neural network with weight matrices 38:22 Building a neural network to predict class performance based on lecture attendance and project hours 40:19 Neural networks need to be trained with data and feedback to make accurate predictions 44:08 Training neural networks involves finding the weights that minimize loss. 46:14 Gradient descent helps find local minimum by updating weights. 50:05 Computing gradients for weights in neural network 52:02 Overview of forward and back propagation in neural networks 55:43 Setting adaptive learning rates to navigate minima and maxima 57:34 Training neural networks involves optimizing weights with billions of dimensions efficiently. 1:01:22 Mini batches offer faster convergence and parallel computation 1:03:24 Overfitting and underfitting in machine learning 1:07:08 Monitor loss during training to prevent overfitting 1:08:57 Fundamental building blocks of neural networks
I really Appreciate you guys taking this possible... Much love and thanks to you... I hope someday I will be able to continue my studies in such a great university such as MIT. ;)
@43:33 - Depending on the loss function we use, it defers what output we get. For example: If we use the Binary-cross-entropy loss function, we get a probability distribution as the output If we use the Mean-squared-error-loss function, we get a real-valued output Have I got it right?
Thanks for the videos and the slides, they are great assets for students and teachers. I wish that you have explained more about back propagation with a numerical example, and the different activation functions we can use in the last layer for the different classification problem, like binary classifications multi-class classification and regression problems
Phenomenal Alexander !! Thanks a ton for your efforts in optimizing & simplifying the concepts. I hv focused on compute gradient went through many but none said the actual pitfall of compute gradient though it plays a crucial role it has some noise & to ovecome that I hope we can use adaptive learning rate algorithms like RMSProp, or Adagrad
This is one the best lecture series for deep learning out there... keep up the good work!!!! Will there be any lecture on the lab assignment - on how do you configure your tensorflow on Google Colab for the assignement/project? I believe that it would be idea/good if there is some lecture video to show how do you configure the Tensorflow on Google Colab. Thank you.
This I got if it may be helpful: Setting up a TensorFlow lab assignment on Google Colab involves a few steps: 1. *Create a new Colab notebook*: Go to Google Colab and create a new notebook by clicking on "New Notebook" or "File" > "New Notebook". 2. *Install TensorFlow*: Run the following command to install TensorFlow: ``` !pip install tensorflow ``` 1. *Import TensorFlow*: Run the following command to import TensorFlow: ``` import tensorflow as tf ``` 1. *Verify TensorFlow version*: Run the following command to verify the TensorFlow version: ``` print(tf.__version__) ``` 1. *Enable GPU acceleration*: If you have a GPU available, run the following command to
Check this out.. 1. _Create a new Colab notebook_: Go to Google Colab and create a new notebook by clicking on "New Notebook" or "File" > "New Notebook". 2. _Install TensorFlow_: Run the following command to install TensorFlow: ``` !pip install tensorflow ``` 1. _Import TensorFlow_: Run the following command to import TensorFlow: ``` import tensorflow as tf ``` 1. _Verify TensorFlow version_: Run the following command to verify the TensorFlow version: ``` print(tf.__version__) ``` 1. _Enable GPU acceleration_: If you have a GPU available, run the following command to enable GPU acceleration: ``` !pip install tensorflow-gpu ``` Then, restart the runtime by clicking "Runtime" > "Factory Reset Runtime" or "Runtime" > "Restart Runtime". 1. _Verify GPU acceleration_: Run the following command to verify GPU acceleration: ``` print(tf.config.experimental.list_devices()) ``` This should list the available devices, including the GPU. 1. _Set up the assignment_: Follow the instructions provided in the assignment or project to set up the environment, load the data, and implement the required tasks. 2. _Load the data_: Use the appropriate library (e.g., Pandas, NumPy) to load the data into Colab. 3. _Implement the tasks_: Write the code to implement the required tasks, such as data preprocessing, model training, and evaluation. 4. _Run the code_: Execute the code cells to run the tasks. 5. _Visualize the results_: Use visualization libraries (e.g., Matplotlib, Seaborn) to visualize the results. 6. _Save the notebook_: Save the notebook regularly to avoid losing your work. Some additional tips: - Make sure to save your notebook regularly to avoid losing your work. - Use the "Cells" menu to insert new cells or delete existing ones. - Use the "Markdown" option to format text and headings. - Use the "Code" option to write and run code. - Use the "Output" option to view the output of your code. - Use the "Restart" option to restart the runtime if needed. By following these steps, you should be able to set up your TensorFlow lab assignment on Google Colab and start working on your project. bessssst!
Do we have more elaborative video regarding how loss function values are calculated and how the gradient is calculated? And how do we figure out this is how our loss function is going to look over particular weights?
some one stated abt the ReLU function has a discontinuity at '0', this is not true, ReLU is a continuous function, even at '0'. It is however not differentiable at '0'. Yes the reason is ,actually, the function has a corner tangent lines which will have a sharp bend , causing a non-differentiable point.
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
It's wonderful to see universities of the calliber of MIT making education accessible to everyone for free. Thanks MIT!!
Thanks for thanking you thanking MIT for thanking for the videos
@@derroz3157 thnks to thanking guy to me you thanking him/her thnks for thanking i can't thanking you for too much thankful but thanks to say thanks to a person who is thanking MIT from there side I thank you both to be thanking them,
They're making basics available which r available in courser and udemy.. do u think they're putting vidoes of advanced stuff that their students actually learn (and which distinguishes them)?
By "YouSum Live"
00:00:10 Introduction to MIT course on deep learning
00:00:41 Evolution of AI and deep learning
00:02:56 Realism and virality of AI-generated content
00:04:15 Accessibility and cost-effectiveness of deep learning
00:04:19 Advancements in deep learning applications
00:05:11 Empowering deep learning models to create software
00:06:02 Teaching foundations of deep learning
00:07:30 Importance of understanding intelligence and AI
00:14:00 Transition from hand-engineered features to deep learning
00:16:00 Significance of data, compute power, and software in deep learning
00:17:24 Fundamentals of a neural network: the perceptron
00:19:59 Mathematical representation of a perceptron
00:20:51 Activation function and its role in neural networks
00:20:57 Importance of activation functions in neural networks
00:21:11 Sigmoid function: squashes inputs into probabilities
00:23:01 Need for nonlinearity in neural networks
00:23:32 Linear functions insufficient for handling nonlinear data
00:24:22 Nonlinearities enhance neural network expressiveness
00:26:01 Visualizing neural network's decision-making process
00:27:33 Sigmoid function divides space for classification
00:28:16 Understanding feature space in neural networks
00:29:21 Building neural networks step by step
00:31:41 Perceptron's fundamental equation: dot product, bias, nonlinearity
00:32:49 Defining layers and passing information in neural networks
00:37:19 Cascading layers to create deep neural networks
00:38:18 Applying neural networks to real-world problems
00:40:38 Neural network training process explained
00:40:50 Neural networks learn like babies, need data
00:41:12 Teaching neural network to make correct decisions
00:41:32 Importance of minimizing loss for accurate models
00:41:55 Training neural network with data from multiple students
00:42:21 Finding network that minimizes empirical loss
00:42:40 Using softmax function for binary classification
00:43:27 Loss function for real-valued outputs
00:47:56 Gradient descent for optimizing neural network weights
00:59:51 Introduction to gradient descent algorithms
01:00:11 Stochastic gradient descent (SGD) explained
01:00:45 Importance of mini-batch gradient descent
01:01:37 Faster convergence with mini-batches
01:02:03 Parallelization benefits of mini-batches
01:02:30 Understanding overfitting in machine learning
01:04:41 Regularization techniques: Dropout and early stopping
01:06:56 Monitoring training and testing accuracy
01:08:48 Summary of key points in neural network fundamentals
By "YouSum Live"
Thanks!
I am a high school student and I am currently self-studying deep learning and I find it very helpful.
I hope one day I can attend your lectures in person.
Thank you very much.
Idem, It would be really cool
I like to lean self study too
Perfect, you are creating a nice future for yourself
@@MehdiAhmadian I'll do my best.
I've been following these MIT Deep Learning lectures since 2019. I've learned so much. Thank you, Alexander and Ava.
So do I need to watch all previous lectures too? Or are the ones in this 2024 course enough?
@@lakshyajain6765 don't need to since every semester course is self contained unit. This is not created for UA-cam, it's for MIT students and every semester there is new batch.
@@RadixSort3 Thanks a lot!!! Do you have any other resources on MIT ML lectures for their students? this is my alt acc
@@RadixSort3 Thanks a LOT!!! this is my alt. Do you have any idea on some more MIT ML related lectures. I would like to do some research in this field and try to get into a phd program
@@RadixSort3Where can I find the next part?
After being in college for 4 years and dealing with loads of professors, I can hands down say this guy is the best lecturer I've ever seen! Explains tough concepts so well.
Maybe 'cause I don't have a strong base, there's a bunch of stuff I just don't get.
Mnn no n no k no no n no nnnnnn. 😅😅mn no nnn no nnnnnnnnn nnnnnn😅nnnnnnnnn no n no 😅 no nnlnn
No nnlnn😅n nn
Nnnnnnnnnnnn no nn
Nnnnnn non nnnnnnnnnnn
I usually find neural networks challenging to grasp until I watched this lecture. I truly appreciate how you simplified the concept for me.
I want to take this moment to thank UA-cam, MIT and Alexander Amini for suppling this content 4 a person like me who is studding deep learning but was not fortunate enough to study in MIT🙏🙏
Attended Deep Learning lectures at a topmost college of a country, here he clearly explained all that in a single lecture for which the former took 10s of lectures to explain.
This is not for beginners. Having 3+ years of experience in deep learning i found it interesting on how much information is shoved into 1 single video . Note that each concept is very vast if we dig deeper
could you link some real beginner information so i can understand this course?
There is a playlist in UA-cam names 100 days of deep learning by campusx. You can find everything in deep
U know where we can find some real number training example of using a basic liquid neural network ?
@adityaverma1298 you mean this video series right?
@@noelvase4867 Andrew Ng's deep learning courses
What a privilege and great time we live in that most precious courses like these from MIT are accessible for freee.
I'm in my mid 50s now, and I keep telling my kids this same thing. When I was their age, the Internet wasn't available yet. Information was very hard to come by, let alone quality courses like these. Now, it's all at their/our fingertips!
A great time indeed.
This is prolly the best Deep Learning lesson out there. With some maths or stats background, it's easy to follow. This is gold!
Yesterday we started system identification using neural network, I watched your lecture and now I feel quite comfortable using the concept of deep learning. Thank you Sir and love from Pakistan....
I look at these videos every year after the new annual release and it just never gets old. Too bad in my work, I don't get a chance to apply this knowledge. It is still super fun to watch, like a fun show to me
Thanks!
WOW!!!!😍My professor is chinese and I know he knows a alot of things but after watching this teacher teachin, I understood the importance of a good presentation and most importantly, what a good presentation look like.
Thank you, Alexander and MIT for make this information available for everyone.
The clarity you are providing for such a complix scientific subject is remarkable 👏
Hands down, this is the best low level explanation of deep neural networks I have seen so far.
It's not low level... It's High level like programming languages.
@@HeyMr.OO7 What do you mean by low level in your definition? It is as low level as you can get in this field that you can perform calculations on an entire network by hands without having to rely on computers, not to mention programming languages or libraries. Some data scientists or self-taught professionals I have talked to who are fluent in machine learning tools which are considered high levels do not quite completely understand this low level fundamental and I doubt if they could hand calculate an entire network from scratch.
@@paultvshow alright man ! Now, Go get some air !
@@HeyMr.OO7Stop it and get some help if you can’t even reason. You don’t even know what level means lol.
@@paultvshow God bless your brain man ! Now leave 😅😅
I've been attending these classes from IIM-C. You've summarized 6 hours of my professors' class in just an hour.
I'm coming here right on the day of exam to revise everything
Thanks MIT❤
IIM Calcutta teaches Deep Learning?!
@@abhirupmajumder8620 Yes, DL by Prof Soumyakanti Chakraborty
Sir you are doing a great job, I am student of BSCS, last year from Pakistan. But being a student to learn Deep Learning from last 2 year, I am still a beginner, as the system is not very modern. This lecture seems like a new start for me, which feels very promising. Can you please share the other lectures, so I (students like me) can really advance in this field, and maybe start working at MIT someday. Thanks for teaching in such a beutifull way.
Finally I can follow live lectures
since you strongly pointed that out, what are these big advantages over offline lectures that you're so in favor of?
@@webgpu ofline lectures? I guess you meant to say, "What are the advantages of following live online lectures over recorded online lectures? Did I get the question correctly?
@@page002 sorry I most probably was not able to express myself properly. I meant "what are the advantages of the [opposite of live] lectures - so I think that's what you also meant in your past comment 👍
@@webgpu don't worry. So here's my point -
I prefer Live Recorded lectures over only recorded lectures because when we follow live I think we can connect more with the instructors. Also it gives us the impression that we are also a part of it which a recorded and already published can never give(at least that's what I think).
And last but not the least if we follow the live (recorded) lectures here we will have a clear goal and a Dateline to follow. And I think that's a great thing.
So, any day I prefer Live Recorded lectures or Live lectures if possible over recorded lectures specially for technical things and programming.
I am a pretty bad communicator so, I hope you got your answer even a little. BTW, if you don't mind, try to follow Live lectures once I think you will be able to see the difference personally.
Happy Learning
It's nice having up to date lessons on this stuff considering how fast it moves, even if a good amount of the core content presumably largely stays the same.
FREE EDUCATION IS MUST BE THE RIGHTS OF HUMANITY - GREAT VIDEO
I did this stuff 8 years ago in uni - felt like the deep learning stuff was too "stochastic" 😉 and avoided it. Looking to try this out again now without going through all the new research material, and this video has been great - just the right amount of detail!
Really thank you Dr.Alex for making this material accessible to everyone
I loved this, It's my major course......It's extremely helpful...love from Bangladesh
Both theory and actual implementation in industry code! Perfect! Also, great pacing and depth!
After 5 minutes in one episode, and i can already tell this is the best beginner ai lecture series I have seen!
Can you compare this with Coursera's Deep Learning Specialization by Andrew Ng
Thanks in advance
@@mehulnakra2457 this is much harder than DLS for sure, I have studied both. But MLS and DLS from Andrew Ng give u a broad view of ML and DL, so if u are studying these courses, keep studying
YahoooOoo!! Another great season ahead!
How fascinating is it i wanted to learn about neural networks and just searched neural networks mit and found a course thankyou so much youtube and MIT.
0:00 - Introduction
7:25 - Course information
13:37 - Why deep learning?
17:20 - The perceptron
24:30 - Perceptron example
31;16 - From perceptrons to neural networks
37:51 - Applying neural networks
41:12 - Loss functions
44:22 - Training and gradient descent
49:52 - Backpropagation
54:57 - Setting the learning rate
58:54 - Batched gradient descent
1:02:28 - Regularization: dropout and early stopping
1:08:47 - Summary
I'm a beginner in ml and ai fields and it's amazing to have these lectures online and free. I've a doubt: the neural network showed in 33:44 shouldn't be named 'multi' layer rather than 'single' layer neural network since it has an output layer separated of the hidden layer? Thanks!
Thanks for the lecture, please please make a video or provide a pdf of MATH too, I wanna know the math behind deep learning, svms, pca, ML in general aka grad descent etc, how then that changes when many layers are involved (as in deep learning) so basically
normal ML -> i/p -> mat mul -> o/p
deep learning -> i/p -> mat mul = linear x matrix . non linear x matrix . linear or non linear x matrix ..... -> o/p
etc etc etc I mean try and simplify what goes on mathematically then also give enough formalization that some of us can begin to understand a few of the key ML papers on Arxiv. This has been our biggest challenge truly.
Your way of explaining is like movie screenplay or storytelling we are totally into the world you created.
At 22:45 you mention the ReLU function has a discontinuity at '0', IIUC this is not true, ReLU is a continuous function, even at '0'. It is however not differentiable at '0'.
Sir's explanation is better than any Udemy and Coursera course out there fr😮
Make "MORE" of these videos Alexander. I appreciate your effort. Lots of love from Nepal.💝💝😘😘
There was a young lady named bright, who could travel much faster than light, she set off one day, in a relative way, and returned the preceding night.
Thanks MIT! for making this learning available for all!
Thank you for making these content accessible for everyone
No words to salute for exceptional lecture kn Deep learning, its one of the best lecture in my career, hat's off your awesome skills ❤
It's wonderful to listen to the lectures of the MIT professors
I hope one day i will attend it on person if god allows!
I also want to thank UA-cam, MIT and Alexander Amini for this wonderful work, due to their efforts, we students from all over the world can reach the lecutres of worlds most renowned institutions! Thanks for ur efforts again
Thankyou Alex, this was really a great foundational course on Neural Networks. Will continue with other uploads in this series.
This video is interesting because,this video helps me understand the current price and prediction of Palantir stock. The analyst explains incredibly. Thank you for sharing this valuable information.
AMAZING
lOVED THIS WAY OF EXPLAINING THE NEURAL NETWORKS
This course videos are very exciting!!
Really like the fact that everything is explained so simply and in a way that is digestible for most people. Personally I found it a great video for revision and brushing up concepts that build ML.
This is great. The theoretical framework was well explained. The concept is a lot clearer to me. Thanks for sharing this. Thanks, MIT.
studying AI&Robotics at the new UTN and must say your videos are life-saving! Thanks for sharing your lectures. I really like how well and easy everything is explained. Really learned a lot!
UTN Germnay? Hows the uni bro? Can you we connect?
If I were just starting to learn deep learning, I would start with this video
I did this in 2005 in assembler on the development KIT board ADSP-2189. The learning parameter = gradient could be computed after certain epochs. The algorithm was able to compute each weight very easily, so that weach W(i) was tested and the smaller error the better weight vector was taken into the next epoch. For instance the algorithm tested the W(i) within the range -1 to +1, so it started with -0.5, then 0, then 0.5, and divided each range by two on 16 bit numbers.
00:04 Introduction to MIT 6.S191 - Deep Learning Course
03:02 AI's realism in generating hyperrealistic content
07:36 Teaching machines to process data and inform decision-making abilities
10:02 Introduction to foundations of neural networks and upcoming guest lectures
14:22 Introduction to Deep Learning Paradigm Shift
16:38 GPUs and open source tools drive deep learning advancements
20:42 Different types of nonlinear activation functions in neural networks
22:32 ReLU activation function introduces nonlinearity in neural networks
26:30 Sigmoid function divides space based on input value
28:25 Neural networks have millions or billions of parameters, making visualization challenging
32:26 Building a single layer neural network is simple and modern deep learning libraries provide tools to easily implement it.
34:32 Introduction to two-layered neural network with weight matrices
38:22 Building a neural network to predict class performance based on lecture attendance and project hours
40:19 Neural networks need to be trained with data and feedback to make accurate predictions
44:08 Training neural networks involves finding the weights that minimize loss.
46:14 Gradient descent helps find local minimum by updating weights.
50:05 Computing gradients for weights in neural network
52:02 Overview of forward and back propagation in neural networks
55:43 Setting adaptive learning rates to navigate minima and maxima
57:34 Training neural networks involves optimizing weights with billions of dimensions efficiently.
1:01:22 Mini batches offer faster convergence and parallel computation
1:03:24 Overfitting and underfitting in machine learning
1:07:08 Monitor loss during training to prevent overfitting
1:08:57 Fundamental building blocks of neural networks
The best Introduction to Deep Learning ever!
Great lecture. I’ve been studying NNs for a while now and this helped reinforce a lot of it in a holistic context. Thanks for sharing!
This is a gem of a video ! , being a MS student of AI I can see the comprehendible concepts defined here !
my favorite youtuber just dropped a new episode!
ah that moment when someone who produces good content, produces good content!
Such a great content about computer vision , really helpful and thanks 👍❤❤
I’m in grade 6 this was interesting, I learned a lot.
As a society we should be open sourcing education it’s a net + no matter what
I really Appreciate you guys taking this possible... Much love and thanks to you... I hope someday I will be able to continue my studies in such a great university such as MIT. ;)
@43:33 - Depending on the loss function we use, it defers what output we get.
For example: If we use the Binary-cross-entropy loss function, we get a probability distribution as the output
If we use the Mean-squared-error-loss function, we get a real-valued output
Have I got it right?
What is the prerequisites one must know before diving into this lecture?
Excellent video! just a minor comment: about 27:00 i think you should state clear that (1+3x1-2x2) = z and include the "hat" to y (in the graph)...🖖
Same here: a bit of stumbling occured at 27:00 over a few minutes (for me)...
Always be your big fan, really excellent teachings. These are the ones I'd love to go through again and again!
Every year I'm here, you remain the best
Thanks for the videos and the slides, they are great assets for students and teachers.
I wish that you have explained more about back propagation with a numerical example, and the different activation functions we can use in the last layer for the different classification problem, like binary classifications multi-class classification and regression problems
Phenomenal Alexander !! Thanks a ton for your efforts in optimizing & simplifying the concepts. I hv focused on compute gradient went through many but none said the actual pitfall of compute gradient though it plays a crucial role it has some noise & to ovecome that I hope we can use adaptive learning rate algorithms like RMSProp, or Adagrad
You are a great teacher. I wish my professor explained this way. 🎉
Awesome course !! Can't wait to complete it 😁
Great presentation, thanks for always simplifying these concepts to the understanding of all.
Is there any group to follow with other peers? Has anyone made a link?
None yet, but you could start one 😊
If yiou made one, I'll join, if not, I have a Telegram one.
@@mihaidanielbeuca1083 what's the Telegram link?
Please if you send here the Link please
Hey lets create a group together. That would be nice
This video is just perfect to understand working of neural network. Loved it🎉🎉🎉
good Presentation agood overview about deep learning thanks sir Alexander Amini
INCREDIBLE CONTENT, THANK MIT AND ITS INSTRUCTORS
28:24 This is a very basic idea of deeplearning. I should have watch these lectures before I started my computer vision courses.
I learnt: dropout and early stopping. So you should finish all your lectures since the most important is at the end...
So basically what Meta with Llama 3 has done is give to the community the weights for each perceptron?
Hi dear, Thanks for the course. Like always informative and to the fundamentals of DNN.
Fabulous efficiency
sir you don't know how much i needed this! i am begining to start my research very soon, is there anythingyou recommend to get started with dl ?
was waiting from last December. Thnak you
Well done, lectured, Professor.
Extremely efficient & effective.
Thankyou.
Absolutely amazing. Great to be here.
i am also very happy that i am really right here where i am now.
This man is so smart person thank you brother.
Very useful. Cleared much of the jargon of NN beautifully .
This is one the best lecture series for deep learning out there... keep up the good work!!!! Will there be any lecture on the lab assignment - on how do you configure your tensorflow on Google Colab for the assignement/project? I believe that it would be idea/good if there is some lecture video to show how do you configure the Tensorflow on Google Colab. Thank you.
This I got if it may be helpful:
Setting up a TensorFlow lab assignment on Google Colab involves a few steps:
1. *Create a new Colab notebook*: Go to Google Colab and create a new notebook by clicking on "New Notebook" or "File" > "New Notebook".
2. *Install TensorFlow*: Run the following command to install TensorFlow:
```
!pip install tensorflow
```
1. *Import TensorFlow*: Run the following command to import TensorFlow:
```
import tensorflow as tf
```
1. *Verify TensorFlow version*: Run the following command to verify the TensorFlow version:
```
print(tf.__version__)
```
1. *Enable GPU acceleration*: If you have a GPU available, run the following command to
Check this out..
1. _Create a new Colab notebook_: Go to Google Colab and create a new notebook by clicking on "New Notebook" or "File" > "New Notebook".
2. _Install TensorFlow_: Run the following command to install TensorFlow:
```
!pip install tensorflow
```
1. _Import TensorFlow_: Run the following command to import TensorFlow:
```
import tensorflow as tf
```
1. _Verify TensorFlow version_: Run the following command to verify the TensorFlow version:
```
print(tf.__version__)
```
1. _Enable GPU acceleration_: If you have a GPU available, run the following command to enable GPU acceleration:
```
!pip install tensorflow-gpu
```
Then, restart the runtime by clicking "Runtime" > "Factory Reset Runtime" or "Runtime" > "Restart Runtime".
1. _Verify GPU acceleration_: Run the following command to verify GPU acceleration:
```
print(tf.config.experimental.list_devices())
```
This should list the available devices, including the GPU.
1. _Set up the assignment_: Follow the instructions provided in the assignment or project to set up the environment, load the data, and implement the required tasks.
2. _Load the data_: Use the appropriate library (e.g., Pandas, NumPy) to load the data into Colab.
3. _Implement the tasks_: Write the code to implement the required tasks, such as data preprocessing, model training, and evaluation.
4. _Run the code_: Execute the code cells to run the tasks.
5. _Visualize the results_: Use visualization libraries (e.g., Matplotlib, Seaborn) to visualize the results.
6. _Save the notebook_: Save the notebook regularly to avoid losing your work.
Some additional tips:
- Make sure to save your notebook regularly to avoid losing your work.
- Use the "Cells" menu to insert new cells or delete existing ones.
- Use the "Markdown" option to format text and headings.
- Use the "Code" option to write and run code.
- Use the "Output" option to view the output of your code.
- Use the "Restart" option to restart the runtime if needed.
By following these steps, you should be able to set up your TensorFlow lab assignment on Google Colab and start working on your project.
bessssst!
Muchas gracias!!! estas lecturas me han sido de mucha ayuda :)
Excellent presentation.
Greatly appreciated all information. Thank you.
NEW SEASON BOYS
Amazing, top content! Out of curiosity: Why TensorFlow instead of Pytorch?
It's finally out!! 🤗🤗
Hell of an introduction!!
With the mini batch approach, is it necessary to make an effort to assure that the batch used is somewhat representative of the data set as a whole?
Do we have more elaborative video regarding how loss function values are calculated and how the gradient is calculated? And how do we figure out this is how our loss function is going to look over particular weights?
Looking forward to this! 🙏🏻
what not to love in things that seem good and are about to happen ?
some one stated abt the ReLU function has a discontinuity at '0', this is not true, ReLU is a continuous function, even at '0'. It is however not differentiable at '0'. Yes the reason is ,actually, the function has a corner tangent lines which will have a sharp bend , causing a non-differentiable point.
Great course. I apply this to concrete strength prediction in my research
Amini genius is back!!
Every year I look forward to this!
i too love to expect things that occur periodically to happen the next time!
What a time to be alive!
Great lecture! Was wondering if you could elaborate on the thought process behind choosing Tensorflow instead of Pytorch.