Hacker News Reader

Comment by blt | Hacker News Reader

Someone help me... I am applying for CS Ph.D. programs right now. I've been in industry for a few years and I have become a skilled programmer but I never get to work on really interesting problems. I am sick of wiring up buttons and sitting in UI design meetings. The kind of problems I want to work on require a lot of heavy math/algorithms that I don't know like Control Theory and Machine Learning. I have learned a lot from listening to online courses but I never actually do the projects/homework because my boring day job programming makes me too burnt out on programming to dedicate a lot of time to side projects. I want to get the Ph.D. to become a highly skilled expert R&D engineer so I can go back to industry and do the most interesting jobs instead of the menial ones. Should I do it? (of course I'm going to finish my applications, I can always say no, but seems like people in this thread would have useful input...)

View on HN

I'm actually going to suggest yes to you. OR at least you should start one. I, like you, cannot self-learn outside of work because my brain is done and needs to recharge. I, like you, read plenty and do the courses, but don't actually do the problems.

The real solution to your problem is to get another job. And I will disagree with some of the other posters that a PhD is required for ML because that is not necessarily true. I have a BS/MS in applied math and do research into how to apply ML into engineering automation problems for a fortune 50 manufacturing/eng company. Granted, we are not pushing the bounds of ML research. But we are looking at unsolved problems and using ML, control theory, robotics, etc to solve them.

This is what you want. And yes it is very math/algorithms heavy. Whenever we try to hire more people, HR sends us tons of BS cs guys that are sorely lacking the math background. Most of our hires are engineers (mech/ee/comp/aero) or oddballs with math/physics/ computational science like me.

So you can certainly look for other jobs (and if you want I will look at a resume to give more specific feedback). But if you already know that you don't have the background for controls and ML.. then go get it!

Here's the thing with CS or compEng PhDs. You don't have to finish them to get tons of valuable knowledge and move into the 'cool jobs' out there. But (per the rest of this HN thread) if you happen to get an awesome advisor and team relationship then you might as well finish.

This is a great time to get a PhD to work on problems in machine learning and machine perception, especially if you want to go into industry. There is a lot of demand. I finished my PhD about two years ago, and I do R&D at NASA in machine learning and computer vision (although I'm leaving to become a professor).

Once you are in a PhD program, be upfront with your advisor about this goal and you will likely have an easier time graduating than those students that claim they want to become professors. Typically, to get a PhD you just need to finish your coursework (fairly easy) and produce about three published papers (can be brutally hard or fairly straightforward depending on a ton of factors). The latter's ease also depends on your tolerance for disappointment, because the rewards and successes in research are somewhat sparse.

A lot of the problems people have with the PhD is that there aren't enough academic jobs. You aren't seeking one, so you will be fine as long as you focus on your goal and maintain a good work-life balance most of the time throughout school (this can be very challenging to do, in practice).

I would suggest that you follow some of work done in conferences related to the field that you are interested in. See what topics are currently being tackled and see if you can come up with initial hypothesis on the road to a solution.

A PhD usually requires you to come out of "code-it-up" mode. You clearly have to state a sufficiently scoped problem, why it hasn't been solved before and why your solution can claim to solve it.

If you are looking for more interesting jobs, it is easier just to switch companies. Many RnD companies don't exclusively hire PhD candidates. All the best :)

I think it was Yann LeCun who said on a Reddit AMA that if you want to go into Machine Learning, a PhD is very much mandatory.

I'm doing a PhD myself right now in Machine Learning / Deep Learning. I started half a year ago. I don't have any bad experience yet. But I haven't heard too much bad things at all in Germany. My chair: http://www-i6.informatik.rwth-aachen.de/

Yann's answer was in regards to machine learning specific research. If someone merely wants to work on hard computer science and engineering problems that require ML knowledge, a PhD is not required. Though it will definitely not hurt in any way.

Reply on news.ycombinator.com