Hacker News new | past | comments | ask | show | jobs | submit login
Ask HN: Master vs. Udacity nanodegree
49 points by pgcosta on Sept 16, 2016 | hide | past | favorite | 25 comments
What is your opinion, in terms of added value, of an university master in machine learning vs udacity machine learning nanodegree?

I want to further progress my education, but am very divided: masters are expensive and long (2 years in Barcelona).

Udacity seems more accessible(moneywise), and has also the advantage of being taught by world leaders in the field.

What's your opinion on the matter?

Lets start out by being clear here, Udacity "nanodegree" isn't an accredited degree. So please keep that very clearly in mind.

That being said in the SF Bay Area "tech bubble" no one really cares about degrees. They care about what you know and what you can do. That may not be true where you're looking for a job though. You should look at what real job postings are looking for and perhaps talk to a recruiter or two (if possible).

Anything from a real university is likely to provide you a lot more information and access, way more than you can ever get at Udacity. However if you can't afford the time/money, Udacity or similar learning courses may not be a bad option.

Have you considered an online Master's degree in machine learning from Georgia Tech? Its offered through Udacity so its cheaper and more flexible, but you also get a full degree from Georgia Tech.

Full disclosure- I'm currently in that program and loving it.

Is it possible to do a Master's from GA Tech without a BS/BA from a STEM major? I have a BA, but it's in the humanities, although I'm currently employed as a software engineer.

Yes, I think it should be fine considering you have some work experience.

Once you're accepted though, you'll have to complete two foundational courses within a year (with a grade of at least B) in order to continue to be eligible. This applies to all new students.

From my own experience, some of the courses are pretty easy, so getting a B or higher isn't difficult.

I've done both the nanodegree and am nearly done with OMSCS. the nanodegree is pretty value less in general.

It's not that rigid or difficult and carries no real weight.

I keep getting marketing emails from Udacity about their nanodegrees. Some of them sound pretty attractive (like the ones related to self-driving cars) and I've been tempted to take them. Glad to hear an actual review from someone :)

They're very light. Since they don't carry any weight you'd at least hope to learn something.

You'll get as much if not more from the free machine learning Georgia tech course with Tom Mitchell's book than the nanodegree.

As a follow up, I took Thruns robotic driving course and it suffers from being a purely software course. There are optional hardware projects but no imparted hardware instruction.

So I'd be especially leery of an automated driving nanodegree degree online.

How many hours per week is the program and how long does it take? how far are you into it?

I only take 1-2 courses each semester. Which means it'll take me a total of 3 years to graduate. Right now I'm about a year in, with 3 courses under my belt. One of the courses I took was time-consuming and difficult- maybe 15-20 hours a week. The other two however have been relatively easy, where I've spent less than 10 hours a week on each.

Adding to this, there's an informal course review site that gives you pretty good reviews from other students. It should give you some more context: https://omscentral.com/reviews

What was your math background before beginning?

I have an undergrad degree in Computer Engineering, and have been working in the field for 3 years. So I took 1st year and 2nd year Math courses during my undergrad, but I haven't really used them since.

I just finished the Udacity machine learning nanodegree a bit over a month ago. While it is good (and getting better) at teaching techniques and material, you miss a lot in terms of access to research faculty.

There aren't any research projects, and you won't ever operate at the cutting edge (WaveNet!) at Udacity. It's like trade school for software, which is fine if you're highly motivated or just want to acquire some skills.

> There aren't any research projects, and you won't ever operate at the cutting edge (WaveNet!) at Udacity

Besides doing specific research, the CS grad curriculum at most schools is hardly cutting edge either. Sometimes it lags quite a bit behind the real world because of faculty being more focused on their own research vs teaching.

The goal of a Masters degree in Machine Learning is ideally to prepare you for core research and engineering roles where you work on improving, adapting and inventing new algorithms and techniques.

However, such roles are very limited in the industry. Most jobs you will find expect you to be able to apply, combine and optimize existing algorithms to given real-world problems. This requires a different set of skills that are seldom taught at University (you're typically expected to pick up these practical abilities on your own).

Udacity's Machine Learning Engineer Nanodegree, like its other programs, is heavily project-based, and has been developed with feedback from industry partners in order to emphasize the skills and concepts that are most relevant for the vast majority of jobs that are out there. This focused curriculum allows people with limited time or a related background to efficiently get started in machine learning.

Keeping this mind, ask yourself what your ultimate goal is, what time constraints you have, and choose accordingly. There is no shortcut to success, esp. in a competitive and highly technical field like machine learning - whether you opt for a Masters degree or a Nanodegree, you will have to spend considerable effort building a strong public profile (e.g. by participating in Kaggle competitions, and working on additional projects) in order to make yourself stand out from the crowd.

Good luck!

Disclaimer: I work at Udacity, in case you didn't realize by now :)

Which companies are hiring the ML nano-degree graduates?

Several companies have hired MLND grads, including Google, PwC, Cyient Insights, Henry Ford Health System, TransVoyant, AirBnB and Udacity itself. Also, it is common for companies to hire candidates internally once they demonstrate they've acquired the relevant skills.

Unless you're career progress is dependent on a traditional degree (university professor, old school corporation, etc.), I would say go with whichever route allows you to actually both learn and apply the materials to the fullest extent.

The important thing is that you're able to learn and master the material and use it to create real value in the world. If you can do that, I don't think anyone will care how you came by it.

The great aim of education is not knowledge but action.

As a third, shorter (and cheaper) option I'd suggest the new Coursera ML course [0]. If you're short on money, they'll let you take the courses for free.

I don't know how expensive Master's degrees are in Barcelona, but GA Tech has a online Master's in CS for ~$500 per course [1], where you could focus on ML.

[0] https://www.coursera.org/specializations/machine-learning [1] https://www.omscs.gatech.edu/

A Udacity nano degree is more like a Graduate Certificate but much much cheaper. What udacity has going for them is that at the end of the nano degree you have projects you can show off. Not all certificate programs will make you have a final project for you portfolio.

In 50 years people will know what your degree is and your university will probably still be around. Udacity will be long gone and no one will know what a nanodegree is.

I hope to be either retired or dead in 50 years :D

a masters in machine learning for sure, university adds very important value of 'connections' mostly, I doubt if udacity can provide that.

I was in a strikingly similar position as you at the beginning of this year. In December 2015, I setup a plan to quit my job and take some time to self-learn Artificial Intelligence, when I was accepted into a master's program in Barcelona (UPC's* - assuming it's the same as the one you're considering).

This blogpost outlines what my plan and concerns were about the self-taught route:


Since then, I've been asked why I ultimately pursued the formal degree route, and this was my response:

"Without a formal CS background, I was pretty skeptical about my chances of getting accepted, but I applied anyways. I was so skeptical, that I convinced myself it wouldn't happen and set off to teach myself. But I ended up getting accepted into a Masters program in Barcelona, and I couldn't turn down the opportunity. I love Barcelona as a city, the tuition is reasonable and the program was inline with what I was looking for - a larger focus on application with foundation in theory as opposed to full on theoretical research.

I chose to do the conventional degree because of the above, plus the allure of receiving a piece of paper that people respect. Regardless of my thoughts on the real value of conventional degrees, it's hard to argue against their societal credit.

I'm new to this industry and pretty young, so take everything I say with salt, but my main advice would be to just build cool stuff. Whether you do it at a university or through autodidactism (learned that one from the HN thread), just work on cool projects. My naive hope is that people will care more about stuff you can actually build over a piece of paper with your name on it - but it doesn't hurt to have both."

That was in response to a thread about this guys blog, which gives some further perspective on the self-learning route:


I'd like to add, that I've since decided to do both. I'm using the curriculum I developed for myself with online courses to compliment my formal education from the master's program, which has been working well so far.

ps. If UPC is the program you're looking into, it can be completed in 1.5 years (3 semesters) instead of the full 2. The last semester is dependent on how long it takes to finalize your thesis. Also, if you have questions about the program (again, assuming it's UPC's), my email is available from the site in the first link.


What is your goal, as in, what type of job would you want to have post finishing either option?

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact