I watched a lots of deep learning tuturials before this, some of them were even twice as long as this and yet this explained the best and the most. Thank you for the awesome tutorial without ads for free. You are a hero.
@@claudioa.dmedina2020yes. Agree. This video best fit for one with certain background Else it overwhelms Don’t get me wrong, this video is very good. I will watch again
For absolute beginners, go through this course 2-3 times, try to note important terms, topics, and processes, and then take every topic and go deep sequentially. I am happy that you have such a great mentor like him now. Best wishes all.
After watching a lot of tutorials or courses about deep learning, i can truly say this is probably the best! Everything organized and clear. Congratulations, it will help us a lot! Thank you
⌨️ (0:00) Introduction ⌨️ (1:18) What is Deep Learning ⌨️ (5:25) Introduction to Neural Networks ⌨️ (6:12) How do Neural Networks LEARN? ⌨️ (12:06) Core terminologies used in Deep Learning ⌨️ (12:11) Activation Functions ⌨️ (22:36) Loss Functions ⌨️ (23:42) Optimizers ⌨️ (30:10) Parameters vs Hyperparameters ⌨️ (32:03) Epochs, Batches & Iterations ⌨️ (34:24) Conclusion to Terminologies ⌨️ (35:18) Introduction to Learning ⌨️ (35:34) Supervised Learning ⌨️ (40:21) Unsupervised Learning ⌨️ (43:38) Reinforcement Learning ⌨️ (46:25) Regularization ⌨️ (51:25) Introduction to Neural Network Architectures ⌨️ (51:37) Fully-Connected Feedforward Neural Nets ⌨️ (54:05) Recurrent Neural Nets ⌨️ (1:04:40) Convolutional Neural Nets ⌨️ (1:08:07) Introduction to the 5 Steps to EVERY Deep Learning Model ⌨️ (1:08:23) 1. Gathering Data ⌨️ (1:11:27) 2. Preprocessing the Data ⌨️ (1:19:05) 3. Training your Model ⌨️ (1:19:33) 4. Evaluating your Model ⌨️ (1:19:55) 5. Optimizing your Model's Accuracy ⌨️ (1:25:15) Conclusion to the Course
This is honest to god a great lecture and perfect for introducing deep learning! I hope there’s another video in the future that shows some light programming.
Short and crisp overview of deep learning with cool intuitions. This is the only video that I will recommend to anyone who wants to start with deep learning and even machine learning in general.
Some (hopefully constructive) annotations to improve the video for better clarity: * 6:33 - "Channels have weight". And the slide matches it. But 7:08 says the weight is something of a neuron (how important is the neuron, rather than how important is the relationship). I think it is confusing; aren't weights a property of the relationship/channel, in graph theory? * inside slides that list advantages/disadvantages you might use the color red for disadvantages, and not just always green to highlight terms. Ex. "small" at 29:08. * I like the examples of descending the Everest, and the one about memorising songs. * 40:21 - the slides disappears till 40:57 * same at 52:01; till 53:11 * 58:15 - "sometimes you may find the ???? depicted over time" (the automatic subtitles don't get it either) * 1:04:17 - audio says "input gate, output gate, and a forget gate"; but the slides shows "Update, Reset & Forget gates". * 1:10:30 - "although if you are interested I'll leave them in the notes below"; yet, it would be useful if a list (or a link to more info) could be added in the description. These are the major notes that I think should be fixed, for better clarity. Anyway it is well made, a good explanatory overview of the neural networks world, that I had no idea about it. It was ok to understand for me, if I skip on the name of the specific algorithms (that in the end are implementation details). But I already got some basic knowledge of statistics & data-analysis, and about graph theory. My only doubt is if others that never dig in those topics can follow this video as well. Keep up with this interesting contents! Thanks for your time and effort!
This is a way better introduction than a University's Introduction to Deep Learning Course for beginners. While the university assumes some ML and computer science background before starting DL courses, this video is for complete beginners. I appreciate it!
Great job! As a FYI, when I used to teach, by the time I got to ANN's, I'd already covered statistical modeling. So it was always an "a-ha!" moment when I'd ask the class if they remembered the sigmoid function from before ... and that all the sigmoid functions acted like mini logistic regression models.
Thank you for this amazing video. Honestly, out of all of the tutorials I've watched, it is the first time that someone explains it in such clear and understandable way. Again, thank you for sharing your knowledge!!!
Hi Jason, Nice video,. The range for the si gmoid function (16:24 on your video) is not [0,1] but rather ]0,1[ as a 0 and a 1 can only be obtained towards +-infinity. Same applies to the hyperbolic tangent ranging from ]-1,+1[ and not from [-1,+1]
Thank You for this, this is my 2nd day into this field, and I think I know the big picture of how it all works and learnt around a month of stuff in a day or two.
for those examples we can assume all possible directions until the ball moves as for "dog" we can assume all statements/questions are accurate until more content is provided at least that would be a temporary solution.
It was a good introductory course! Although there's lack of examples in the explanation of neural network part. But overall you can get the idea of how deep learning works.
Nice introduction. Some of the ideas probably should be reigned in a bit. For example, people who study learning do not think machine learning models how the human brain learns at all. Calling neural network nodes "neurons" misleads people into thinking you are actually trying to imitate a neuron instead of just using a software node. Stating that activation functions are non-linear is not always correct - in fact the equation you showed when you said that looks like a linear sum of terms. All this doesn't take away from the work you've done to make these ideas accessible. It's a good intro - but hopefully people understand there are some sweeping statements that might not hold up under close scrutiny.
Neuron and node are interchangeable enough in this context. Had 'neuron' not been used, a consumer of this course will still eventually run into each term. This course certainly wasn't the first to use 'neuron'. NNs do roughly model brain functions. Like Jason said, each works to identify patterns based on data. Jason also goes on to explain the case where linear activation functions are less than optimal. Of course there are exceptions, but beginners probably wouldn't be concerned with them
It a wonderful course that I've ever seen on youtube s about DL. Thanks a lot, Jason for making this one. :). Expecting more vids from you to change lives.
0:10 The computer runs an algorithm on each players possible moves and picks the one that's closest to the king WITHOUT BEING EATEN. There is no ai in that. It's just iterating all the players and iterating all of their moves going from player to player. There are actually many moves to remember and remembering all of them on a move can be sometimes difficult when there are so many possibilities A computers algorithm in that framework is 100% correct all of of the time. In terms of the facial ai it's just objects which are outlined and marked with a set of attributes Trigonometry is really important here. Also business functions play here quite frequently It's also remembering situations and figuring out the best path I took at a path. A path is a set of steps I took. How many times has that played up before. What were the moves that won me the game at that path. That's the hard part. That's deep learning. That's path memory. Very difficult This is where you learn to flip the binary tree on its side and use them as open ports. Drawing it differently and analyzing is all we need to see. DNA Path memory
Thanks for your all videos. But I request you please upload two different videos of 6-7 hour on Calculus and Linear Algebra required for Machine Learning and Data Science
The output shouldn't be right or wrong, it should be a variable no? Because not all input we receive is binary. If anything it should be on a scale of 1-10, 1 being very negative (lots of adjustments to neurons) and 10 being perfect and should increase the weight in the future.
Is there a way to have access to the slides used in the video? It would be really helpful for those who want to revise everything and doesn't have to go through the entire video.
I found the fake news detection statement amusing. This was clearly before we realised that every side of an issue has fake news (and some truths are just not available on any network because the data regarding them is deleted). If created today I would suggest the software would be used to determine the truth according to the clients parameters and specifications. Additionally the software would need to be directed to what are 'reliable sources' or otherwise risk verifying undesirable information. Sorry for overly discussing this issue. This video was very interesting, thank you.
Excellent.....please upload the very basic idea of machine learning and artificial intelligence.... it's too useful for us like intern and upcoming future...please upload freecodecamp...
