Is a masters in ML worth it?

As a hiring manager in the data science/ml world in healthcare, I generally think of degree programs in "data science", "machine learning", "artificial intelligence", "deep learning", etc as being less valuable than degree programs in the corresponding fields that aren't as buzz wordy. I tend to prefer candidates from backgrounds like computer science, math, applied math, statistics, or something domain specific (coming from healthcare) like epidemiology (or variation like computational epi) or bio- or biomedical informatics.

Those programs tend to show me that you're interested in and have done the more boring but foundational coursework that is often cut to make the sexy degree programs. That means that hopefully you won't be upset that 100% of your job isn't deep learning, and that you'll be better suited to pick the right tool for the job.

At one of my last jobs, there was a machine learning engineering team (all boys) and a data science team (all girls and gays) who had the same ML chops. The DS team ended up getting more models into production and more research published than the ML team because they had more "soft" skills to navigate the problems the org was facing. When someone in leadership would say "we're having issues booking appointments", the ML team would set off building some fancy deep learning model while the DS team would generate hypotheses with stakeholders, do some exploratory analysis, run a few prospective studies, and then use those results to inform some regression models that would end up in production. It wasn't as sexy as some deep learning model, but the leadership team wanted full interpretability of their model so deep learning was never going to be acceptable. I generally think of these kinds of skills being taught more the stats, applied math, or epi programs than in the designer ML programs. ymmv

I was in a combo program, got my Master's in ML on the way to a PhD. I really didn't quite understand academia before I entered. It got to the point where I couldn't see the light at the end of the tunnel, I had spent 4 years on a very long paper that didn't represent my own desired direction of work in the field. And the whole time I had simmering thoughts that I'd rather being building software in an engineering context, that academia was not my place.

I eventually quit and did find a job with a smaller company building math-heavy analysis software for an engineering field. I'm quite happy with my choice; I pick my language and tools, I have complete agency over my part of the product stack, and I solve interesting problems. Sometimes I entertain the "what-if," but I see the ML industry as stuck in a massive bull-trap bubble, with a lot of people working on "products" that add zero value to their company's portfolio.

I don't chase easy money, I chase interesting work.

As a data scientist, I would advise against moving from software engineering to ML. There is an enormous glut of junior DS/ML people. The boot camps and masters programs capitalized on the hype, and now there are way more than the market can absorb. Anecdotally, my recruiter spam has gone way down over the past couple years.

The only worthwhile related masters degree, IMO, would be in large AI from a top-tier university.

I work close to the engineering school at a big Uni and often meet MEng students, at times in the past I had them help me out with ML projects that but usually I end up talking to them on the bus or in the 6th floor kitchen. So it's a question I ponder.

I had one student who did all the coding for a successful project that we wrote a paper about. I've talked to others who feel like they are "over their head" but are working on teams with people who know how to do all the math. That might be OK because there is a lot more to successful ML projects than being an individual contributor: very few people know how to manage ML projects and if you learned something about that it could be highly valuable.

If you study for a PhD you will definitely learn the math, but you've got an entirely different problem that you could be out of the workforce between 5-10 years and the market could look entirely different then than it looks now.

I completed the ML specialization for Georgia Tech's online CS MS program (https://omscs.gatech.edu/) a few year ago and found it tremendously valuable. I can't say that it changed the trajectory of my career, but I learned a lot, met some great people, and (in hindsight) had a ton of fun.

Keep in mind that most machine learning fundamentals haven't changed for decades. While new architectures/trends are always emerging, the lessons that you will take away from an academic program like OMSCS will be relevant for the rest of your career.

I think it really depends on where the Masters is. I personally have noticed a lot of degree mills at the Masters level that are superficial academically and primarily designed to lure in candidates to the most profitable part of a university.

So, I'm guessing that the people with the most regrets about joining the California gold rush were the ones who decided to move there after 1850. It might be worth thinking about whether this is because you find machine learning interesting on a personal level, or whether you're just trying to be where the excitement is. If the AI hype has already started to cool off by the time you graduate, are you going to feel regret about the real & opportunity costs of going back to school, or would you still see it as personally rewarding and feel motivated to persevere in building a machine learning career in a slower job market?

That said, if you want to get into machine learning, I've been seeing that increasingly many entry-level machine learning jobs now require a master's degree of some sort. Be aware, though, that these jobs may not be glamorous. I've go a master's degree myself, and usually see all the really exciting work get handed to colleagues with PhDs. I still find it to be a rewarding career, but that's possibly because I enjoy the work itself, and am not particularly worried about the glamor.

It may also be worth pointing out, while all of my colleagues do have at least a master's degree, none of them have one in machine learning specifically. Typically it's one in some other hard or social science. Basically anything that builds a strong foundation in statistics, linear algebra, and research & experimental methodology. The actual machine learning part of the job is kind of the easy part, and mostly belongs in the top row of that "what people think I do / what I really do" meme. I did have two former colleagues who started out as software engineers and then got a master's in machine learning / data science / whatever in order to pivot to AI/ML. Both of them really struggled to settle into the work, and ended up reverting back to just being software engineers, only now with a lot more student debt.

There are lots of high quality answers on the recruitment side here, so I'll focus on the personal.

I did a masters in my mid thirties, and it was definitely the best life choice I made that decade.

Education, ideally, is about learning, not career progression. I had a fantastic time being completely focused on studying.

There's nothing nicer than waking up in the morning knowing that your main task for the day is reading a bunch of interesting papers.

The specific course, well, that's down to you, but as a thing to do with a year or so of your life: magnificent.

Lots of folks end up with kids, or without the cash, or stuck in a job for other reasons. It's fantastic to be even able to consider it. If you can, I think it's a great idea from my experience. Worst case scenario, you will grow in skills and confidence.

> Will a masters in ML give me an advantage when applying

A master's will almost certainly do one or more of 1) Getting you past the resume screener and the posting's absurdly high education requirements 2) Showing that you are not dumb 3) Showing that you are willing to put in work 4) Communicating that you might actually know how to do stuff 5) Letting the hiring manager brag about hiring someone with a master's in ML instead of a bachelor's.

Too early to tell.

How long is a masters going to take? Three years? What's the ML landscape going to look like in three years?

I don't know the answer. In three years, ML could have pretty much maxed out, and people could be not hiring for it any more. Or in three years a masters in ML could be the golden ticket. I don't know. I don't think anyone else knows, either.

My impression is that "these cutting edge tech companies" are not easily able to find as many qualified candidates as they want right now. Emphasis on "right now". Is there something you can do to become qualified (by their definition) that will be faster than a masters?

(I'm pretty sure that their definition of "qualified" has to be less than a masters - there hasn't been enough time for there to be very many ML masters degrees yet.)

a key aspect of the cutting edge companies is their ability to execute on very large and complicated software engineering projects, so you should see if your existing software engineering skills might already be useful.

e.g. look at Anthropic's job postings for engineering roles -- they don't super-emphasize AI knowledge, they seem to also face a ton of classic engineering challenges. there's as much about managing cloud infrastructure and distributed systems as there is about transformers.

I doubt it. Most places hiring want hands on experience with tech stacks; PyTorch, opencv, training, fine tuning models… concrete skills

Make reference implementations from a couple recent papers, publish them on GitHub. Fine tune a demo model to show off during an interview.

Thats shows you lean into the work, not memorization of academias circumlocution

If you feel like you have decent math chops, i would use that time and money on online training, hardware, develop a portfolio of projects that demonstrate your new skills, and make contributions to open source ML libs to pad your resume. But also, many of the new AI startups are just using ML as a service where expertise is not necessarily needed.

I specifically pursued a master of computer science in data science to cover all of my bases. If I'm applying to a job where I want to highlight the focus in data science I can but I don't need to. My diploma just says "master of computer science" on it and in some job applications that's what I want it to be.

I'm stacking my chips in the study of quantum computing. It might be fruitless, it might not. I just think its really cool. Plus it feels like I'm casting a wide net. If quantum does become more practical, what is every tech company in the world going to gravitate to, no matter their field?

I had a friend who started a masters in ai at uw. He really struggled with it because of the opportunity cost. He knew he was losing money and career progression by staying. He ended up dropping out.

In addition to the other comments keep in mind that the number of people actually working directly with ML at these companies is much smaller than the number of people claiming to.

Here's what I would look at. (This goes with any certification or degree). Does the learning experience contained within the masters degree look like it would be worth learning. If so then you can go do that learning and accidentally get a degree while you are there. If not, then you'd be participating in pay to win, a game in which the house always wins.

I work at a startup, and recently built the fraud detection algorithm we use to classify and process payments. I have a BS in CS with a minor in Data Science, and also have applied ML experience. My take is that your skills and knowledge are paramount over your degree, especially if you're building ML systems that are adding value to your company.

Its a dream of mine to get a PhD in computer science or even mathematics. I only have my undergrad in CS and 7-8 years of software engineering experience. Due to some bad decisions Im also quite broke.

Im also thinking of doing a masters at georgia tech or something, or just diving in, but not really sure what to do. I just know Ill regret it if I never do it.

I'll be hiring this year as well. I'm in the manufacturing domain.

As said a few times below, I'm looking for real work experience. Specifically in the domain.

I also agree, there is going to be a lot more really boring work, mixed in with some new and interesting. Pretty much the job description for most of the IT world.

Makes you wonder ... how to break in if you have a master's in AI, but from 2007 (before the mild AI winter ~2006-2008) and have been doing systems engineering since.

So how do you get a job in AI without dropping to an intern's wage?

My masters in info science was not. I'd be farther in my career if I had just put in extra hours everyday instead of spending it in classes.

The sad reality is that most paid Masters degrees (and even non-paid via opportunity cost) are scams that are unlikely to lead to a job.

I did a Masters in Data Science and worthiness depends on a lot of factors.

- The experience is what you make of it, how challenging a course-load you choose, etc. It is often possible to coast thru these programs by taking the easiest classes, if you just want the degree. It is also possible to really go off the deep end and learn a ton, if you want to put in major hours.

- Office hours with professors made it worth-while. I utilized almost every one and grilled the profs hard with questions. I imagine you can get this for free if you work at Google/Meta/etc but I didnt have access to that type of intellectual sounding board, so the MS made a lot of sense.

- Degree diversification makes sense. I had an east coast ivy league undergrad, so I wanted to balance that out with a west coast technical institution with a very different alumni base, so I chose my MS program accordingly.

- The network matters. Find a Uni with a great network.

- Ecosystem matters. Find a Uni with a great ecosystem of startups, accelerators, VCs, and FAANG workers who are adjunct. That limits you to Bay Area and possibly NY/Boston in the US, but I think people underestimate how valuable this is. I compared my undergrad (rural Ivy league) to grad (Bay Area) and the difference was night and day. There is just very limited ecosystem possible when the University is stuck in the mountains.

- Job placement can be good or bad. Uni recruiters have their "target student profiles" whom they work with to place them at top companies. If you arent on their "target student profiles" list, they wont help you. I spoke to great students in 2021 graduating at the absolute peak of the tech hiring craze, who had been sent away by the university recruiters noting "there are no jobs now."

- To learn deep ML you need to start a hard-startup and dive deep into it w/o a dayjob. I did. I did an ML/CV startup diagnostics firm focused on medical imaging. I learned more in 3yrs than I did in 10yrs in industry. I did this in parallel with the MS in DS and that would a wonderful combination because the campus had numerous benefits for students with startups. I could also run hard problems I encountered by my professors and TAs. The MS gave me theoretical knowledge and the startup forced me to truly learn things because otherwise I couldnt ship.

- A small number of students (15%) were outstanding and I keep in touch with them, they made the experience worth it.

- A large number of students (30% ?) were not paying anything for the degree, often it was being paid for by the US Government as part of US Gov benefits programs. I'm sure it was worth it for them. In some cases they didnt value it as much those paying out of pocket. I saw some completely free-riding and doing no work and leaving most of the effort to team members.

- Ambitious projects get seen. I had many, many employers reach out and try to hire me after seeing my masters projects. But you have to learn to manage, market, document and be very vigilant with whom you work with. One free-riding student can seriously tank a major project, and you end up suffering. A cohort of equally ambitious students is unstoppable and a pleasure to work with.

If you want, you are welcome to email me and I'm happy to speak to you on the phone.

We have masters in ML now, wow! What do they teach apart from stats, optimization and linear algebra?

It can be totally worth it. If you put your all into it

Why not start one?