I am into alot of things rl, dl, now a bit into gpu's as well, currently a research intern, web dev as well. So will keep it short & ofc will start from the start & will mention some of the books that are pure gold.
maths (few things over & over again)
Maths is everywhere in this, from the transformers to rl to regression models. But there are few concepts that will show up like 90% of the times. Here are some of the topics : dot product, covariance matrix, eigen vectors, probabilistic distribution functions, KL divergence, entropy, cross entropy, differentiation basics like what it means & basic chain rule, normal distribution, famous plots & basic statistics.
Now this can expand but instead of like covering things at once, i will choose again that method of learning things when they will show up. I appreciate this method cause i can see the effect up close of like maths is just an tool & when i can see how that tool will help us solve an problem so that makes the process more interesting in my opinion.
books to follow
DL by goodfellow then Hands on ML by aurelion green. So i haven't completed the goodfellow book but i know cause i read some sections, what i liked about that books is that it starts with the linear algebra & probability and information theory that too is in good level.
But after that the flow shifts to deep networks instantly which is where the "hands on ml" comes into play. It is solid structured like how the things should be so firstly it covers the ml algos then dl then diffusion models then a bit of rl & deployment as well.
That book is not that math heavy but it gives you a scope of like how the domain pretty much looks like. After that its either a researcher or a developer with a subdomain like you want to go into cuda shit or rl or ++. But the goal here is to get an idea, also the language doesn't matter concept matters, like i can switch from python to lua as easily as tensorflow to pytorch.
Now ofc don't give me that shitty argument that there is a depth in languages but mf you have google to understand things on the go.
build end to end projects
Yea this is very important to get internships as a beginner or like have an idea of the whole ecosystem where deployment plays a major role. After you go through the book you will be comfortable in the models & languages like pytorch, now its the right time to get a bit into the RAG side as well.
Now that's a slipperly slope but basic idea of langchain & langgraph to test things if you enjoy that then you can ofc go into that subdomain. There is a youtube tutorial on that maybe i'll attach the link later so cover that too.
Now the main thing is tools like Grafana, promethus, AWS instances & s3 buckets, docker, kubernetes, terraform & basically to get an idea of how real world deployment happens. Here we will keep the model simple cause we want to learn how the things integrate.
An end to end model doesn't need the model to be a huge neural network it just needs to get the job done whether its prediction or categorization. I am covering these things too as of now, ik im late but still.
Best thing try to learn from open source repos which contains good documentation which leads to the next point.
repo for everything
Keep your github clean, one mistake i did was that i coded or learned things daily but i didn't focused on keeping it structured in a repo. This gives a clean & impressive looks, so if you start learning a book then create a repo & document things nicely and yup that will help, its subtle but there is no harm in this.
keep in mind this is harder than web dev
I have to say this, web dev is repetitive but there is money there in the early stages. Again this is what i have observed like after a certain time web dev is like doing same thing again & again which i agree might happen with you in ml but this domain is literally growing & will grow in the upcoming years like every months we get dozens of paper with new advancements.
So you can try to get into freelance as well where you can build ML + classic web dev shit and i have heard that this pays good. I haven't done so no experience but i've heard.
That's pretty much it, now don't dm me asking for questions bro like legit. You should firstly go and checkout some of my blogs here : https://tmwork.vercel.app/blog.html
two of em - ml one & homl will answer some of your questions. Rest try to join spaces & ask there, folks will easily guide you. If you're pretty confused then you can dm but show fckin respect, i don't owe you fckin answers. So be nice & i'll be nice as well.
Till then keep learning :)