Hi folks! All of the links are in the description! Here is the rough skeleton of the video. 1. Learn software engineering/coding in Python 2. Get a high-level overview of the field (Coursera) 3. Start to get deeper (fast.ai) 4. Start reading research papers and implement at least 1 paper from scratch 5. Get a sound maths foundations Happy (deep) learning!
These tips are precious and pretty accurate. I really like how you try to explain in the best way and give us the best practices from the tech industry. Keep going!
Thanks buddy! Hope you find it useful I know you're currently working mostly on pure software engineering but in case you start with ML you have a place to start!
Man after hearing your very first advice, I know you know what you are talking about! I see so many bright "researchers" handicapped by their coding skills. Software engineering skills indeed should be ranked THE top skill to have if you ever want to be truly productive in any technical field, e.g., ML.
Dude, you are the best! I share your passion for passing on what I have learned, thanks! Also, I appreciate the time you put into making these videos. Also++, I can't believe this has so few views and likes. Gracias pibe!
thanks man, its really helpful.i have been learning these stuff for about five monthes now, starting with books like introduction to algorithm, probalistic machine learning byKevin Murphy' , reinfocement learning by Sutton & Barto, and online course from Probabilistic Machine Learning by Philipp Hennig , and David silver's lecture i on reinforcement learning is excellent! as well the deep learning course lectured by other DeepMind researchers. and its great pleasure to read Petar Veličković's paper. iits encouraging to see that someone self taught can get into DeepMind. hopefully i can get somewhere one day
Bunch of useful stuff, even someone who is quite deep into ML can find a lot of useful information in this video. But I do have to disagree with your first statement, that in the School of Electrical Engineering in Belgrade there are no ML classes: prof. Predrag Tadic is doing a great job, he's teaching two ML courses and I highly recommend them to all students.
Exactly what i was looking for. I did computer engineering in college, with a focus on ML, but havent used it since. And i really want to take my career in that direction. Thanks for this video, and all the resources
Hi Marc! Super glad to hear that! Keep putting in the effort and you'll get there! It just takes time as pretty much everything in life. #gettingphilosophical
Awesome ! i'm trying to find proper roadmap since few weeks..it was really frustrating as there so much material...and i can surely say this is the one. thank you so much Aleksa✌
I really found ur video inspiring and straightforwards. already subscribed to it. Generally new comers in this field are confused. your video makes the way
Banging video! Thank you very much! I have a question tho: I'm really interested in NLP. This is what interest me the most. You said the framework you focus on is Pytorch and Computervision as your application area. My question is if I wanna do NLP what are tools I really need to focus on ? Maybe you can help me out a bit dude :D
This was super super helpful and detailed just what i needed to for a roadmap and thank you so much for all the links and book suggestions, you made my ML journey a lot clear now.
dude so awesome! thanks so much! what a great boon for self learners like me ! :) Couple of questions: 1. You can do the Math last? I though you had to do it somewhere in the middle 2. What do you think of Kaggle ? I was thinking of doing that Instead of or as addition to projects which are hard to find . Thanks so much!
Slow response but I hope it helps the others at least. 1. Well, TBH, if you get into the bleeding edge ML you'll definitely need math but then again it depends on the specific subfield you end up in. Transformers are less mathy whereas RL uses way more mathematical formalism. The sooner you're good at maths the better (again some people will just be using high level API and it's questionable whether they need to learn maths in great depth). 2. I've personally never tinkered too much with Kaggle, I was more into developing my own projects, playing with them, creating nice visualizations, etc. Check out my GitHub here: github.com/gordicaleksa If you have time definitely check Kaggle out, but don't overfit to their format. Let me clarify what I mean by overfitting. I guess it's similar to the competitive programming - software engineering relationship. If you do SOME competitive programming that's great - if you do it too much it can actually hurt you becoming a good software engineer as you tend to focus on writing dirty code all the time in competitions.
Hi Aleksa, thank you for the video. I was wondering if you could create a list of papers to implement of increasing difficulty. That would really help a lot because it is a bit messy to find papers to get started with that are relatively easier to implement. Just my two cents.
It is great roadmaps, and your story is very inspirational. Currently, I am on step 2, follow course of Andrew Ng on Coursera. If I didn't understand in depth what professor said I am trying to populate my math knowledge with course from Stanford, where professor Ng explain everything with much more mathematic point of view. Pozdrav:)
Until next time "keep learning deep learning". Awesome work, very good material... I was just looking for what to learn next after Deep Learning specialization. What type of projects would you recommend after the DL specialization? Would a neural style transfer be a good project for that phase of the learning?
Hahhahha yep, thanks Vojislave! I am literally about to create a video about that this evening but I'll first propose some more beginner-friendly projects - though I think they may also be helpful to you too. If you're totally new to the field even if you completed the DL specialization I'd recommend you go and play with classification on the MNIST dataset, first just use off-the-shelf models (VGG, mobile nets) , then try and build a feedforward net and get it to work with decent accuracy. Create a confusion matrix understand where your algo is failing. Than build CNN and do the same thing. That's a nice way to start. Next up, yes NST and deep dream are super beginner friendly and a nice way to start getting deeper. Hope that helps!
Yup fellow Serb here. Hahaha winter is coming brace yourself. Never been there, but I've been to St. Petersburg during the white nights, what a beautiful city!
Great video! I'm stuck after finishing step 2. I don't know what kind of cool project I should start. I started a kaggle competition and failed miserably. Can you help? What kind of project should I pick.
What's your opinion on people diving right into deep learning without going through traditional ML algorithms like decision trees, gradient boosting, PCA, SVM, etc?
@@gustavojuantorena Gracias de nuevo! Tengo muchos amigos en America Latina.. Hice una practica en Brasil hace 2 años. Espero irme a la Argentina el año que viene!
Thank you for the awesome video! I really appreciate it. I do have a question though. In the video description, you put the two fastai courses under "mid-level intro" and both coursera courses under "high-level intro". How can I start with high level first? because you mentioned that we better start with Coursera Andrew Ng course? Also you mentioned that fastai courses are more practical. I am kinda confused, which ones I should start first please? Thanks a lot again!
Hey! Thanks Adam! So start with the high level resources first (Coursera), those don't go so much into depth but more into breadth and they give you a (high level) overview of the field. Hope that clarifies it, I guess the confusion came from the "high level" part, that simply means something doesn't go to deeply into the topic (ML in our context).
i feel it wll be much better to learn some (necessray) Data structures & Algorithms otherwise jumping directly to building models feels so overwhelming .
Very interesting video. I have a question. When you talk about reproducing research papers, does it include training? Because it can be very difficult for some if not most of the big papers.thanks
Thank you so much for this! Currently at the stage where I finished DLAI Cousera Courses. May I know, by projects, what kind of project do you mean? Whats the difference between this and implementing research paper's work? thanks!
Implementing a research paper is the last step you on your journey to becoming an ML/research engineer. Find some simpler projects, do a walk-through of them, understand them in all the details. Then try and play with that code, modify it, visualize stuff, debug. PyTorch tutorials are not a bad place to start. Happy learning!
@@TheAIEpiphany Thank you!!! I sort of jumped ahead, and I am currently trying out NST using transformer encodings. Managed to get the content image reconstruction to work (yay!), trying to figure out the style part now. Cheers!
I'd only add that if one doesn't know math fundamentals before starting to learn ML then one will end up better with some math guy on the team rather than trying to learn it on one's own ;)
![ILLSLICK - WINTER IS COMING [Official Video]](http://i.ytimg.com/vi/dwTMeM1zIhQ/mqdefault.jpg)
'embrace the suck' ahahah i swear this was the most important aspect of learning ML and reading research papers.
Hi folks! All of the links are in the description!
Here is the rough skeleton of the video.
1. Learn software engineering/coding in Python
2. Get a high-level overview of the field (Coursera)
3. Start to get deeper (fast.ai)
4. Start reading research papers and implement at least 1 paper from scratch
5. Get a sound maths foundations
Happy (deep) learning!
These tips are precious and pretty accurate. I really like how you try to explain in the best way and give us the best practices from the tech industry. Keep going!
Thanks buddy! Hope you find it useful I know you're currently working mostly on pure software engineering but in case you start with ML you have a place to start!
@@TheAIEpiphany Absolutely! Thank you so much! I'll turn on the notification bell!
One of the best roadmaps i have come across while searching how to start ML. Thanks buddy
Thanks Kaushik! Glad to hear that!
you are the best thing that's ever happened to me. I miss you dreadfully
I was lost in the learning process lately so bad, and it was really helpful; thank you.
Man after hearing your very first advice, I know you know what you are talking about! I see so many bright "researchers" handicapped by their coding skills. Software engineering skills indeed should be ranked THE top skill to have if you ever want to be truly productive in any technical field, e.g., ML.
Dude, you are the best! I share your passion for passing on what I have learned, thanks!
Also, I appreciate the time you put into making these videos.
Also++, I can't believe this has so few views and likes.
Gracias pibe!
Thanks man! 🚀
Great advice. You’ll hit 100k subscribers quickly. Keep up the great content
Thanks Nick!!
thanks man, its really helpful.i have been learning these stuff for about five monthes now, starting with books like introduction to algorithm, probalistic machine learning byKevin Murphy' , reinfocement learning by Sutton & Barto, and online course from Probabilistic Machine Learning by Philipp Hennig , and David silver's lecture i on reinforcement learning is excellent! as well the deep learning course lectured by other DeepMind researchers. and its great pleasure to read Petar Veličković's paper. iits encouraging to see that someone self taught can get into DeepMind. hopefully i can get somewhere one day
I do agree with you on taking Mike's lessons, he is very good at making the language curve less bumpy
Bunch of useful stuff, even someone who is quite deep into ML can find a lot of useful information in this video. But I do have to disagree with your first statement, that in the School of Electrical Engineering in Belgrade there are no ML classes: prof. Predrag Tadic is doing a great job, he's teaching two ML courses and I highly recommend them to all students.
Thanks!! For the second part what I meant is we don't have master program or complete curriculum in general - which is afaik true.
That's right, these ML courses are (unfortunately) just a minor portion of the curriculum.
HI thanks man. I am getting started with Machine Learning. I found this video incredilby useful.
Thanks for this roadmap and tips. Already starting a blog and planning my DL MVP
Exactly what i was looking for. I did computer engineering in college, with a focus on ML, but havent used it since. And i really want to take my career in that direction. Thanks for this video, and all the resources
Hi Marc! Super glad to hear that! Keep putting in the effort and you'll get there! It just takes time as pretty much everything in life. #gettingphilosophical
Comment here to start following your advice from today. Thank you so much for the knowledge.
Great that you can make a video on the side too. I remember editing my karate kata video, which took far more time than I thought that it would.
Loved this, thanks!
Awesome ! i'm trying to find proper roadmap since few weeks..it was really frustrating as there so much material...and i can surely say this is the one.
thank you so much Aleksa✌
This video is a holy grail. Thanks dude
I really found ur video inspiring and straightforwards. already subscribed to it. Generally new comers in this field are confused. your video makes the way
Thank you! That's what fuels me!
The best way you could support this channel is by sharing, stay safe!
Thanks for this practical video including learning path, resources and tips. Its super helpful! 👍
You're welcome! 🙏
Thank you very much for sharing your expertise with us.
Its huge act of generosity.
🙏🙏🙏
Thank you for sharing these gladiator tips
Banging video! Thank you very much!
I have a question tho:
I'm really interested in NLP. This is what interest me the most. You said the framework you focus on is Pytorch and Computervision as your application area. My question is if I wanna do NLP what are tools I really need to focus on ? Maybe you can help me out a bit dude :D
This was super super helpful and detailed just what i needed to for a roadmap and thank you so much for all the links and book suggestions, you made my ML journey a lot clear now.
Thanks Akshat! I'm super happy that you found it useful!
dude so awesome! thanks so much! what a great boon for self learners like me ! :) Couple of questions: 1. You can do the Math last? I though you had to do it somewhere in the middle 2. What do you think of Kaggle ? I was thinking of doing that Instead of or as addition to projects which are hard to find . Thanks so much!
Slow response but I hope it helps the others at least.
1. Well, TBH, if you get into the bleeding edge ML you'll definitely need math but then again it depends on the specific subfield you end up in. Transformers are less mathy whereas RL uses way more mathematical formalism. The sooner you're good at maths the better (again some people will just be using high level API and it's questionable whether they need to learn maths in great depth).
2. I've personally never tinkered too much with Kaggle, I was more into developing my own projects, playing with them, creating nice visualizations, etc. Check out my GitHub here: github.com/gordicaleksa
If you have time definitely check Kaggle out, but don't overfit to their format. Let me clarify what I mean by overfitting. I guess it's similar to the competitive programming - software engineering relationship. If you do SOME competitive programming that's great - if you do it too much it can actually hurt you becoming a good software engineer as you tend to focus on writing dirty code all the time in competitions.
Thanks a lot man
really helpful and organized
Thanks Motaz my friend! Glad you found it useful!
Great advice, really appreciate your structure and approach to learning!
Nice. It would be so great to have some theory or code videos on SOTA transformers and attention for computer vision. Awesome channel!
Thanks! And thanks for the feedback it's super valuable for me. I'll definitely put that on my backlog and consider doing it sooner rather than later.
Hi Aleksa, thank you for the video. I was wondering if you could create a list of papers to implement of increasing difficulty. That would really help a lot because it is a bit messy to find papers to get started with that are relatively easier to implement. Just my two cents.
GOLD!
It is great roadmaps, and your story is very inspirational. Currently, I am on step 2, follow course of Andrew Ng on Coursera. If I didn't understand in depth what professor said I am trying to populate my math knowledge with course from Stanford, where professor Ng explain everything with much more mathematic point of view. Pozdrav:)
Thank you! Awesome, just keep on working on it, you'll get there! Poz! 😄
Until next time "keep learning deep learning".
Awesome work, very good material...
I was just looking for what to learn next after Deep Learning specialization.
What type of projects would you recommend after the DL specialization? Would a neural style transfer be a good project for that phase of the learning?
Hahhahha yep, thanks Vojislave!
I am literally about to create a video about that this evening but I'll first propose some more beginner-friendly projects - though I think they may also be helpful to you too.
If you're totally new to the field even if you completed the DL specialization I'd recommend you go and play with classification on the MNIST dataset, first just use off-the-shelf models (VGG, mobile nets) , then try and build a feedforward net and get it to work with decent accuracy. Create a confusion matrix understand where your algo is failing.
Than build CNN and do the same thing. That's a nice way to start.
Next up, yes NST and deep dream are super beginner friendly and a nice way to start getting deeper.
Hope that helps!
@@TheAIEpiphany That helps a lot, I like those ideas, thank you! Im looking forward to see that video once you upload it. Keep up the good work!
@@vojislavmladenovic2968 thanks buddy!
on point. Awesome channel, thx!
Superb. Thank you.
Glad you found it useful ^^
hey, i didn't know you are from Serbia! greetings from Siberia :D
Yup fellow Serb here. Hahaha winter is coming brace yourself. Never been there, but I've been to St. Petersburg during the white nights, what a beautiful city!
Great video! I'm stuck after finishing step 2. I don't know what kind of cool project I should start. I started a kaggle competition and failed miserably. Can you help? What kind of project should I pick.
What's your opinion on people diving right into deep learning without going through traditional ML algorithms like decision trees, gradient boosting, PCA, SVM, etc?
Absolutely fine, they re very different beasts
Awesome video! Greetings from Argentina
Muchas gracias Gustavo! Jajaj
I hope I'm not annoying people by talking to them in their language even though they talk to me in English. 😅
@@TheAIEpiphany Jaja para nada! Excelente contenido. Saludos!
@@gustavojuantorena Gracias de nuevo! Tengo muchos amigos en America Latina.. Hice una practica en Brasil hace 2 años. Espero irme a la Argentina el año que viene!
@@TheAIEpiphany Qué bueno! Yo vivo en Buenos Aires. Ahí empecé a seguirte en Twitter, so we can keep in touch! 👍
Thank you for the awesome video! I really appreciate it.
I do have a question though. In the video description, you put the two fastai courses under "mid-level intro" and both coursera courses under "high-level intro". How can I start with high level first? because you mentioned that we better start with Coursera Andrew Ng course? Also you mentioned that fastai courses are more practical. I am kinda confused, which ones I should start first please?
Thanks a lot again!
Hey! Thanks Adam!
So start with the high level resources first (Coursera), those don't go so much into depth but more into breadth and they give you a (high level) overview of the field.
Hope that clarifies it, I guess the confusion came from the "high level" part, that simply means something doesn't go to deeply into the topic (ML in our context).
@@TheAIEpiphany I got you now. Thank you very much for the quick reply. Appreciate the great info! All the best!
i feel it wll be much better to learn some (necessray) Data structures & Algorithms otherwise jumping directly to building models feels so overwhelming .
Very interesting video. I have a question. When you talk about reproducing research papers, does it include training? Because it can be very difficult for some if not most of the big papers.thanks
Thank you so much for this! Currently at the stage where I finished DLAI Cousera Courses. May I know, by projects, what kind of project do you mean? Whats the difference between this and implementing research paper's work? thanks!
Implementing a research paper is the last step you on your journey to becoming an ML/research engineer. Find some simpler projects, do a walk-through of them, understand them in all the details. Then try and play with that code, modify it, visualize stuff, debug. PyTorch tutorials are not a bad place to start. Happy learning!
@@TheAIEpiphany Thank you!!! I sort of jumped ahead, and I am currently trying out NST using transformer encodings. Managed to get the content image reconstruction to work (yay!), trying to figure out the style part now. Cheers!
I'd only add that if one doesn't know math fundamentals before starting to learn ML then one will end up better with some math guy on the team rather than trying to learn it on one's own ;)
Hahah well I dunno I learned ML math by myself also 😅 It's just the question of priorities and whether it's necessery given your situation
Is possible to use PyTorch for The Deep Learning Specialization instead of TenserFlow?
Hello sir, is it okay to skip data structures and algorithms?
Data Structures and Algorithms are mainly useful for software development and software engineering not much for Artificial intelligence.
Please upgrade microphone, you have 2000e salary for god sake
Hahahah these were my older videos I do have a Rode mic now, and my salary in Microsoft was bigger than 2ke 😂
врхунски
Хвала!
Good Stuff! Thanks :)
Thanks Syed!