How to get started with Machine Learning

Key Takeaways

in this video i'm going to give you a five step curriculum for mastering machine learning and six bonus tips so i'm currently working as a machine learning engineer microsoft where i've landed thanks to this curriculum i've never taken a single official university course on machine learning why because we don't have any here in serbia and i guess most of you won't have one on your university also so my background is in electrical engineering uh i did have some computer science subjects but i had to take additional software engineering self-education path and i've landed my first job ever as a software engineer at microsoft where i'm currently working as a machine learning engineer thanks to my curriculum so machine learning is this super useful field and you must have seen these videos from open ai where this robotic hand learns how to solve rubik's cube in a hundred percent simulated environment why the [ __ ] do i have this i don't even know how to solve this thing what we've done is trained an algorithm to solve the rubik's cube one-handed with a robotic hat or these videos from waymo where a car is learning how to drive by itself by using computer vision and deep learning it knows exactly where it is on the road the street is mapped it can also identify everything around it in full 360 degrees objects are identified and analyzed and then predict what those things might do next paths are charted so the biggest problem with machine learning is that there are too many resources out there and in the process you just get a decision making fatigue and you give up the truth be told they're all just reiterating on the pretty much very same things and it's actually really easy to learn machine learning and no matter your background i promise you you can learn it i'm gonna give you five simple steps and six bonus points to get you started step number one learn to code in python so the better your software engineering skills in python are the easier it will be for you to just follow along the next steps so if you're a complete beginner uh you should go through this four hour long video and uh just follow along and do and encode everything that that guy just teaches you to do don't just binge watch the thing you gotta learn how to code and the best way to learn something is to do it yourself so if you're not a complete beginner i recommend you going through this book called automate the boring stuff with python and if you still feel like your python coding skills are kind of weak just go through the first eight chapters if you feel confident enough just skip the eight chapters and uh just do the chapters nine uh to the end of the book uh you'll learn so much from this you'll learn how to automate stuff you previously uh maybe did manually or something it's such an amazing thing to have in your tool set you'll further improve your python skills by just learning stuff on the fly so you're doing some concrete project or test whatever you just go to stack overflow you query something you figure it out and that's it that's how you learn python don't go about doing bunch of courses bunch of books just do this believe me it's enough so just get used to this workflow in python you don't know how to do something you just go ahead and you google it like reverse list in python and you pretty much open like uh you can open like stack overflow and find some results here or another cool resource i'm i'm often using is gigs for gigs this website is also pretty awesome you'll find your answers there step number two get a high level overview of what machine learning can do and what machine learning is so in this part you're gonna go through two courses the first one is a general machine learning course taught by andrew eng so he's a famous professor and you probably even heard of this course and this course you're going to learn about different techniques such as linear regression logistic regression even a little bit about neural networks uh ie deep learning and some other techniques like svms and you're gonna get a glimpse of what like the field looks like the second course you're gonna take is a deep learning course taught also by andrew eng and his company deep learning ai and in this one you're going to get a understanding of what deep learning is a field is you're going to learn how to structure your projects and you're going to learn something about convolutional neural networks which are really popular in the field of computer vision and also something about rnns or in general sequence models if this all sounds gibberish at the moment just stick with me and you'll be able to understand all of this as soon as you start following this curriculum so it took me approximately two and a half months to go through these two courses and the reason being i was actually working full time in microsoft and i was also doing some other resources which i'll tell you more about like in the bonus points on the end of the video so uh you basically can finish this one in less than a month and you'll learn a lot so during these two courses you're gonna learn a lot of new terminology just don't get frightened uh it's totally okay not to understand everything you're still like on some uh higher level of the knowledge pyramid you'll get deeper uh down the pyramid and you'll learn uh more details as the like time goes by so just stick with the course do everything and when you finish with the course create some project that's like immensely important for you to understand once you finish the course you have to create something on your own and you have to publish it somewhere like on github just open source it and you learn a lot and you'll have something to show off it's not only about certificates certificates that are nice but once you have some project something concrete that you can show to other people that's so much better and you can see my github repo here and some of the projects i've been open sourcing over the last couple of months so you learn a lot by doing that and you just get some kind of public artifact so other people can know that you're at least this good also i encourage you to go ahead and write some blog about the things that you learned during the courses and in general try and and and switch between these two modes which i like to call input and output modes because i'm a nerd and engineer so basically input mode is you're taking ingesting information you're just learning you're reading you're focusing output mode is when you're actually trying and trying and structuring the knowledge and trying and outputting it in the sense like either code or or blog or whatever or video just go ahead and use that knowledge and do something practical so please please do not skip this step i cannot stress enough how important it is for you to create some concrete coding project step three we're getting deeper into the middle level of the knowledge pyramid here you're going to take two fast ai courses which are much more practical than coursera courses so the first one you're going to do is called deep learning for coders and if you notice there's some overlap with things you already learned feel free to skip some content but i strongly suggest you try and do this course also the second fasta course which you should take is called deep learning from the foundations and you're going to dig a lot deeper into how deep learning works how the back prop is implemented how the uh how the optim optimizers are implemented etc so go through this it will help you immensely in the long term and again after you finish these two courses do create your own project and do open sources on github this will help you tremendously and at this point of time you should already start focusing on one specific deep learning framework and i strongly suggest you start with pytorch uh fast ai's deep learning from foundations will also teach you a lot of stuff about pytorch just keep on using a single framework do not try and learn tensorflow keras and stuff just focus on one single framework it will be easy for you to switch if needed so the first three steps i showed you can easily take you up to six months if you're a total beginner if you're not feel free to crop this curriculum so as to fit your level of expertise step 4 we're getting deeper start reading research papers and implement a single paper so i'm obviously trying to linearize this ml learning process but learning is an extremely non-linear process so there is a huge chance that you'll start reading research papers before you get to this step now if there is one site that you should know about when you start reading research papers that's archive and that's just a place where people publish their papers and you can see one paper here on the screen and if there is one single word of advice i can give you here when you start reading the papers that's to just embrace the suck that means you're going to feel so dumb there's going to be so many mathematical symbols you're not aware of algorithms terms terminology you've never heard of just be ready to read the thing from end to end and not understand anything on the first pass reading papers is an extremely important skill for both machine learning engineers and also for machine learning researchers so once you read a couple of papers in a specific field like say i read 20ish papers in neural style transfer literature go ahead and implement the damn thing you're going to learn so much by doing this on the screen you can see my implementation of the original neural style transfer paper and i really preach what i do myself the things that i know that were useful for me and also for others last but not least step 5 mathematics listen up in the end you have to have at least one person on the team who'll know what the heck is going on and that person will know mathematics i'm not saying you should have that knowledge i'm just saying if you have a startup or whatever you'll have to have at least one person who have a thorough understanding of mathematical foundations for machine learning if you were following along this curriculum you'll you'll have some basic at least basic mathematical foundation so far uh the thing i recommend you do here is go over this book called mathematics for machine learning it's free it's an awesome resource it took me around three to four months to finish it but i'd supplement i supplemented it with three blue one browns uh playlists for linear algebra and calculus and also with this python for data science handbook and i was also working full time so you can do it faster but it will take some time you just need to go through the fifth chapter of this python data science handbook uh it's about machine learning for the end of this curriculum uh if you want to build the rock solid deep learning foundations uh there is this book called deep learning by ian goodfellow and his colleagues it's super academic super thorough it will just give you additional uh additional knowledge you need to have in order to do some serious research in the field of deep learning so that's it for the curriculum if you stayed so far with me uh congrats i'll now share the six bonus tips with you tip number one so this curriculum will require you to have uh self-discipline obviously and uh good learning skills so before you even start doing this curriculum if you're new to learning stuff on your own i'd strongly recommend you go through this course by coursera called learning how to learn i finished it exactly one year ago and found it super useful even though i have a long track record of learning on my own tip number two do not learn tooling for the sake of tooling like this for example this python for uh python data science handbook has this uh chapters on numpy matplotlib and pandas uh just don't go and teach yourself these tools without previously needing them so for example i've been i've been working full time for two years already in this field and i've been doing a bunch of stuff aside and i still haven't had to use pandas because i'm not working with structured data i'm working with mostly with imagery because i'm working the field of computer vision tip number three i can't emphasize this enough build your own stuff after you go through the theory this is super important so set a project goal that's the what and a lot of house will follow up so if you'll need pandas to accomplish the project you'll learn pandas on the fly tip number four focus focus focus so in the beginning just focus on a single deep learning framework that's by torque for me focus on a single application area that's computer vision for me uh you'll get to learn some a lot of terminology you'll get some self-confidence nobody can keep up with all of the fields so i have friends in deepmind who are experts at computer vision but they can't keep up with reinforcement learning tip number five follow people on twitter there are a lot of cool people on twitter which post regularly and that will help you keep up with the newest things that are happening in the artificial intelligence field tip number six follow lex friedman's podcast he has a youtube channel and he's been having amazing guests like i've been following him for almost two years now and you can learn so much from the best people from the industry and from research so those were all the steps and tips i had please comment if you think i've missed something or feel free to add your own suggestions in the comment section also subscribe to this channel if you found this video useful and hit the bell icon to be notified when i upload a new video until next time keep learning [Music] you