Ask HN: back to math? | Hacker News Reader

Ask HN: back to math?

13 pointscageface14 comments

It seems like a lot of the most interesting jobs in software these days require a pretty heavy math background. Things like analytics and machine learning and computer graphics in particular. It's occurred to me that it might be worthwhile to go back to school and get a graduate degree in math, which is something that appeals to me intrinsically because I've always enjoyed math anyway.

Has anyone else done this? If so, did you find it helpful in getting more interesting work or did you just find yourself overqualified for the typical software job?

View on HN

Analytics is actually very cave-man level math and statistics. You could just work through Strang for linear algebra and a stats book and you'll know more than 80% of the people employed doing this work...and that's probably not even necessary. I will warn you that if you're already an infrastructure programmer, most of what analytics teams really need is someone to babysit the data. If you don't want to deal with that, I would disclaim all knowledge of computer programming beyond Microsoft Excel.

There's loads of blogs where people discuss statistics/ML and echo Hal Varian's quote that it's the next "sexy career" but if you look at the authors of those blogs they are all still in grad school, so take that advice with a grain of salt. Actuaries have been employed doing stats for centuries and they have never been "sexy," but they get paid ok.

Personally, I find statistics heinously boring so the excitement is lost on me. Instead of thinking "wow, we're awash in data, finally a reason to use my Stats 101" I tend to think: "oh fuck now that there's all this data I better not let anyone know I was an applied math major...I'll get stuck doing Stats 101 all over again." Now if someone wanted to pay me $200K a year to do knot theory, I'd be stoked...

Graphics is math heavy but you learn that math in a computer science grad degree, not a math grad degree.

Aside from Wall Street (where the useful math is often from the EE department, anyway) most math heavy programming jobs (i.e. "scientific computation") probably pay worse than what you're making now.

Signed, your friendly neighborhood math teacher turned programmer.

It should be noted that an M.A. in statistics will require aptitude in both the calculus and algebra. Moreover, a generalized applied degree will require preliminary knowledge of ordinary differential equations and real analysis. I’ve also heard from professors that some competency in scientific computation (i.e. familiarity with both a well-used computer algebra system and statistical package) is usually expected.

I haven’t pursued either, but I suspect that I will--I’m still an undergraduate studying general mathematics. But my assumption is that a generalized applied program is more useful than statistics. I’ve found breadth to be more important than depth in curriculum. That is, if you have a rigorous foundation in probability and mathematical statistics, then you could conceivably learn sampling theory, Bayesian statistics, or linear models outside of the classroom. Likewise, a rigorous foundation in differential equations might enable you to pursue chaotic or dynamical systems independently.

I can’t speak to computer graphics specifically, but some of my peers work in quantitative finance, and they’ve echoed my sentiments on breadth. I’ve also heard from professors that one unexpected occurrence about students who enter either program is that many enter with a specific research interest, say quantitative finance, but realize they prefer to work in a different field, say biological statistics. I find the mobility and flexibility attractive.

Thanks for the insights. I really enjoy studying math for its own sake so I'm naturally more inclined to study broadly and build a solid, common foundation than I am to zero-in on one speciality. My experience in FP, for example, has kindled an interest in studying algebra in greater depth.

Yes! I'm in the final stages of finishing my M.S. in Math. I've been doing it part time and it's been difficult and time-consuming, but very enjoyable.

Interestingly, studying math may have a non-obvious benefit of helping you understand functional languages like Scheme, Lisp, Clojure, etc. And I believe that functional languages are going to be quite useful in the years ahead. So even though you may not get a job in applied math, it may be helpful in all other kinds of software development.

It was actually my experience learning various FP languages that inspired this idea in the first place.

I'm just starting my 4th year back studying math formally, 2nd year of PhD. I couldn't be having a better time. Who knows what the other end will look like, but I doubt I will ever work the "typical software job" again.

My hero! Did you have to start from the beginning with an undergraduate degree or were you able to pick up in the middle somewhere?

Mathematically, the beginning, more or less. I had the advantage of having dropped out of my undergrad, so I went back (after 13 years), changed my major to math (from CS), finished in two years and applied to graduate school.

If I hadn't been able to resume my undergrad, I'm not sure what I would have done. Masters? Part-time classes until I'd bulked up my background and impressed a few professors to write me recommendations? I don't think I would have done very well applying straight to PhD programs.

Analytics seems more stats oriented. Actually, I'd say the same about machine learning, too. Not that math and stats are fundamentally distinct, but there are different graduate programs offered for the two.

That's a good point. I've also considered a stats focus. My limited experience with ML suggests that linear algebra is pretty important too but certainly statistics is fundamental to both.

For ML, I think it depends heavily on what you want to do. If you want to develop new algorithms, prove theorems about them, etc., there is a heavy math focus, although in terms of departments, you'd probably be more likely to do it in a CS department with a strong ML group, than in either a math or stats department. If you want to use ML though, the skills are quite different. A good handle on statistics is still necessary, so you know when to use which techniques, and how to interpret the results, but a lot of the math used in ML as a research area is used mainly to develop the algorithms and prove theorems about them, and isn't as relevant if you're looking at them mainly as predictive modeling or data-mining tools to apply to various domains.

did someone tried to learn math ( linear algebra ) from the net ?

Try this (lectures from the man himself):

http://academicearth.org/courses/linear-algebra

I've been working through these myself. Great lectures. The whole MIT OCW thing is a godsend if you're trying to learn any of this stuff outside of a classroom.

Reply on news.ycombinator.com