How to become a better MLE as a new grad?
I'm graduating with a degree from a T10 school's fifth year masters program and will be starting at FB in a month. I want to be an MLE and hopefully grow to high levels like E6/7, but I don't actually have a handle on what skills I need to prioritize and develop to succeed as a MLE. I took a bunch of ML classes in school and did some ML research, but I've honestly forgotten a lot of the more esoteric stuff I've learned. Are there any resources out there that contain a lot of conceptual knowledge MLEs should know? What skills are important to succeed as a MLE? How can I train them? When you're learning something, it's often better to do targeted drills/focused learning than simply practicing it over and over again - is there something like that for MLEs? Is it stupid to aim for high levels as a MLE at FAANG without a PhD? #machinelearningengineer #machinelearning
Where can I find that T10 school list?
There are plenty of lists out there, I just googled top CS schools. My schools tend to be in the T10 in almost all the lists I've seen.
US News is the one most people use: https://www.usnews.com/best-graduate-schools/top-science-schools/computer-science-rankings
You don’t need a phd for MLE at Facebook. On the other hand, applied scientist at Amazon often prefers a phd. But you are already at Facebook. So why bother
Do you know your Facebook team? I don't know how hard it is to transfer teams there, don't want to be stuck with SQL/PHP role.
No, but I interned on an ML team and implemented techniques from papers for the most part, so I'm fairly confident in my ability to get on a ML team during bootcamp. At the very least, I could always just go back to the team I interned on.
In some companies there’s less of a fine line between MLE and ML research (e.g Google and Facebook) but ideally you want to be very comfortable reading papers and implementing the models in tensorflow or pytorch. Get use to that first. The best way to do that is just practice, read some popular papers, find implementations on GitHub and try to understand how it’s done. As an example, I spend most days reading papers and training models, and my role technically is SWE. Also I’d prioritize on getting up to speed on deep learning, as most ML research at FANGG is working on these types of models. So something like fast.ai would be a good source to get up to speed fast.
What this guy said + there is often more E to MLE than ML, so the usual engineering stuff applies. Also haven't had any experience with fast.ai, but Andrew Ng's deep learning specialization on Coursera is excellent (later courses which are not delivered by him personally - less so).
What area of ML are you interested in?
Not quite sure yet. My research was in NLP. I think the types of problems I want to focus on the most are recommendation systems since that seems like a large space w/ very relevant business needs. Probably something w/ ads?
AFAIK MLEs are required in building pipelines, big data sort of stuff, scaling models rather than mode creation itself which works perfectly. It’ll be much harder getting there as a data scientist type, simply because you’ll spend your intitial years catching upto phds