How I'd learn ML in 2025 (if I could start over)

all you need to learn machine learning in 2025 is a laptop and a list of steps you need to take I said it last year and I'll say it again but this time I'm an actual research scientist at one of the best AI startups in the world and it took me over 6 years to get to this point and nowadays you have so many new and amazing resources that way too little people know of so today I'll share how I would learn machine learning in 2025 if I could start over by revealing the six key

steps you need to take let's get going well this one feels obvious doesn't it but what might not be so obvious is how much python you actually need to learn in the beginning in general all these steps don't have to be strictly followed in any particular order but I would recommend to not start with the final and arguably most important part but what I do recommend to start with is learning python python is the programming language used by anyone working in ml so it is very important to get the hang of it many of the

following steps build on your fundamental understanding of the basics of python so I would definitely want to know what a list and dictionary are and what the differences are how to implement a for Loop and if elf statements I would even go as far as knowing what list comprehension and class inheritance are and honestly I don't really know what else to say than and just type in beginner Python tutorial into YouTube or Google and get started there are so many amazing extensive free resources that you can use to learn the basics of python but make

sure to always code along now here comes perhaps an important step which is I would want to build some cool little fun projects after learning these Basics straight up jump into the cold water and build something like a calculator or a simple website a snake game or or whatever these beginner python projects are I would honestly just want to really have fun here but not spend too much time here especially if you already know how to code and just want to learn the basics of python don't spend too much time trying to nail the python

experience because although you have to know Python and be good at programming I would not tackle that in this stage yet so let's go to the next one learning machine learning does not require complex math after studying it for multiple years I really stand by the statement so much of machine learning can be covered or be understood with really fundamental concepts that are mostly undergraduate level even professional ml Engineers a large part of them don't require complex Marth in their day-to-day life or even in the interviews if I were to start over I would want

to learn Main M three to four things how to do derivatives and integrals although even integrals are rarely super relevant what vectors and matrices are and how the basic operations work which are also just rules you learn and to have a lot of intuition behind them the basic concepts behind probability Theory which boil down to a fairly small set of Concepts and mainly base Rule and finally some random math tricks that make lives easier for example the log rules and summation rules but you pick those up on the way these Concepts and rules will get

you immensely far in machine learning with many of the new and different models that you learn they're all just a different application of some of these rules understanding them on a intuitive level especially the probability theory part that might be a bit more challenging at least it was for me but it's still not a lot and it certainly isn't rocket science that said The more you learn the better of course and especially if you want to go into research you will want to learn a bit more math get a bit more practice applying the math

and perhaps just get a better intuition so with all that said how would I actually learn math in 2025 well there are of course many resources but one that is pretty new straight to the point and just amazing is this book called why machines learn I know I know books are kind of scary and perhaps a bit outdated for some of you because video courses are so much more convenient but I really don't think they have to exclude each other in fact this is something that I will recommend a lot in this video that there

are different resources that you can use to learn the same concept there's not one book lecture or course that covers everything or that will help you understand a concept from the first go but this book covers a lot of the fundamental math in a very fun and understandable way I can really recommend it it builds up the idea of how linear equations work and how they directly apply to neurons in neural networks what vectors are and the most important operations like the dot product how matrices and the relevant operations work and of course the math

behind training and network gr and descent a ton of amazing explanations and intuition on probability Theory and so much more all the way up to how convolutional your networks work now does this book cover absolutely everything no of course not but it covers a ton of fundamental math that you need what it for example doesn't cover is actually teaching you how to do derivatives so in any case that I would not understand something or he just skips something I would just look that up myself type it into YouTube for some some nice YouTube videos type

it into Google for perhaps some nice blog posts or something that you can do in 2025 is even use an LM those are surprisingly powerful and I genuinely use them in my day-to-day to ask questions that said be aware that LMS can Linate I State factually wrong things so yeah just be aware of that now if you do want a specific recommendation for a math course that I might use myself I would then probably recommend the same thing I recommended last year which is KH Academy nevertheless in general I would try to stick to this

book and expand from there the amazing thing about this book is it teaches you math in the context of machine learning and it already touches on many of the concepts in ml but that said let's have a look at what I would do after reading this book I would now continue to learn everything relevant to machine learning and deep learning finally right now I Del L split up machine learning and deep learning because they are a bit different and they are taught separately but the machine learning part or the classical machine learning is perhaps not

the most flashy things you see nowadays with all these deep newal networks but it definitely is core knowledge so I would not want to skip this part there was already a lot covered in why machines learn but of course not everything so next I would simply take andr nun's machine learning specialization course if there learn many new machine learn learning models like logistic regression decision trees recommended systems and more practical advice on how to develop machine learning models another great thing that comes with this course is practical exercises which is amazing you learn things like

tensorflow and you will actually Implement your first machine learning pipelines and train your first models which is super exciting so up until now I've learned basic python the fundamental math necessary for machine learning and steep learning have learned some of the basic machine learning models and have actually coded up my first machine learning pipelines and trained my first models but now comes the really exciting part this is the exciting part but also the part where you have to make an important decision do you want to learn deep learning to a level where you can understand

the current machine learning techniques and models and apply them to a certain problem or do you really want to learn deep learning looking at non-conventional models and applying the fundamental math in more complex ways and with a focus on a bit more Theory you see machine learning is still a very new and empirical field which often requires much more other skills than just an understanding of machine learning theory building cool ml projects or even ml products in reality simply often does not require a fundamental understanding of the theory so if I wanted to learn deep

learning to a point where I could get a job as quickly as possible I would pick the first option which I would call the applied path in that case I would not spend too much time learning a ton of deep learning theory so I would just take Andre's deep learning specialization where you learn again a lot of fundamentals and actually get some practical coding exposure yet again no resource pass at all unfortunately this course does not really cover the Transformer architecture which at this point is just mandatory to know so for that I would simply

watch something like Stanford's cs25 series which is available on YouTube and I'd also watch basically all of Andre kath's YouTube videos and try to code along now I should be decently equipped to jump to the next chapter of this machine Learning Journey which is probably the most important one but before we get to that let's have a look at what I would do if I really want to learn deep learning and then get a job at one of the top companies and even set me up for my path as a researcher okay so what I

would do is work through another book called understanding deep learning this is an absolutely amazing resource with a ton of content about deep learning it basically touches on every relevant model in deep learning that you will needs to know I will not list all the different topics that this book covers because it's a ton but you will see them here on screen the amazing thing is that this is actually a free resource it's available as a free pdf online and you don't have to buy the book I bought the book because I want to support

this actual project what's even greater is that it actually has a lot of theoretical and practical exercises for the different topics it covers I mean even this book does not cover everything it doesn't really cover rnn's and lstms which is a bit unfortunate but I guess it is because it puts more emphasis on the Transformer architecture which is the very dominant architecture these days everything exists out there on the internet what you really need is someone to take you by the hand and tell you what to learn or in other words a curriculum and this

is what this book offers it is a very dense book and it covers a ton and it will take a lot of time to work through but you don't need to rush through this book there's at least one more chapter along this machine Learning Journey that is at least as important as understanding the theory in fact I wouldn't rush through this book in one go I would mix it with the next step that we'll gets to projects I cannot emphasize enough how important projects are as mentioned before machine learning is still a very empirical and

practical field so basically no matter what job you are striving for you will have to code a lot first there of course are some fundamental ml developer stack libraries like numpy pandas and M plot lip for manipulating and visualizing data they again are really nice tutorials on YouTube and even a 20 minute tutorial will be enough after that you need to use the tool and when it comes to the actual machine learning Frameworks such as py Doge tensorflow or perhaps even Jacks there again are amazing tutorials online but there's nothing that teaches you a skill

or a tool or anything better than actually applying it so in the beginning I would simply start with kegle and do not underestimate the complexity really start with earlier or beginner level projects so you don't get frustrated and demotivated and if you do try to attempt more challenging challenges where you can actually win a prize money don't expect to win one because it really is difficult and you often need a lot of compute available with kegle alone you can already go quite far especially if you continue on progressing to more and more advanced challenges but

if you want to go to more complex projects that are outside of just a jupyter notebook that would be relevant for machine learning engineers and researchers you would want to go to The Next Step Building actual bigger and challenging projects and my favorite project to work on which I already mentioned in my last year's video is reimplementing a paper and how to do that is for a whole another video or way too much for this video but it is a challenging problem and you learn a ton so make sure to pick a paper that is

more suited to your level so you don't get frustrated and give up because that is the worst thing you could do you really learn a ton by reading papers alone that's a whole art in itself and you learn even more by just looking at code that other people have written and if you're getting really good at machine learning you can perhaps even find ways of improving some of the recent works and at that point you are basically already a researcher or at least an engineer okay but still what I would make clear to myself is

that my first project will not be a good one my second one will probably also not be the best one but with every project that you do you will work on more increasingly complex projects that are more and more impressive and this is really important to know because this makes my next bonus tip much less scary no matter at which point I would be on my machine Learning Journey if I were to relearn machine learning I would want to somehow show my work and present myself while learning machine learning fundamentals I would perhaps write a

little X or LinkedIn post when learning cool new deep learning techniques I would perhaps write a whole blog post explaining it or a LinkedIn post but most importantly whenever I work on a project I would somehow want to again write a blog post or perhaps even make a working demo website I mean this is what once got me an internship interview with a Google deepmind researcher I worked on a project and then wrote a blog post for this project and the final level would be actually writing up a paper about your project you can always

do this and upload it to something like archive but the final Bo of the final level is actually getting your paper published at a pretty nice conference but at that point you are pretty pretty far in your machine Learning Journey it really really takes a lot of time look I will repeat this again because it's important it takes a lot of time at least depending on where you start your machine Learning Journey and how much time you can dedicate to active study it will very likely be really really challenging at some point and you will

very likely be struggling when learning the math or coding at least I was when I was learning it that's why I would recommend watching this video next I there share the secret tips I learned over many years by myself that help me learn machine learning in a more easy way so yeah in the end don't give up and try to have as much fun as possible bye-bye