HN Explorer

Ask HN: How many machine learning or other research-type jobs are there?

2 points by epicureanideal ↗ HN
As patio11 says[1], 90% of all software is line of business software.

Where are the other 10% of jobs? How are they distributed? One or two people per company? (I worked for a company with 2 research-type engineers out of 10ish total.) How large is the team inside Google or Amazon? What are the qualifications needed to get into those positions? Are they significantly better paying? Are there more skilled people for these positions (like in academia), or more positions than skilled people?

Basically, it seems that there is some missing information about the market, and I'm not sure how much effort to put into training for a highly specific skill set without specific information about the potential benefits.

[1] http://www.kalzumeus.com/2011/10/28/dont-call-yourself-a-programmer/

1 comment

[ 3.2 ms ] story [ 14.7 ms ] thread
tgflynn
Several years ago I worked for a very large company in a rather small group (about 40 people) that worked on what were then challenging image recognition problems. I think that most work in these areas is done either in large companies like Google, Microsoft, IBM, etc. or in a few small startups that are founded specifically to tackle these types of problems by people with very strong backgrounds in them (often from academia).

What are the qualifications needed to get into those positions?

In my opinion the ideal background would probably be an undergraduate CS major with a minor or a lot of course work in applied mathematics (at least at the level required for an EE or physics type degree) who has developed very strong practical programming skills. This would ideally be followed by a masters or PhD in machine learning. Many people get into the field without these kinds of qualifications through self-study, however (unless they're looking for a job in academia or a place that is specifically considered a "research lab"). The minimal requirements to learn the field are probably being comfortable with applied math through linear algebra and vector calculus with at least some statistics and probability along with strong practical programming skills.

Aside from all this the sine qua non of getting a job seems to be getting good at handling technical interviews.

Are they significantly better paying?

I doubt it. That's mostly a question of supply vs. demand (see next question).

Are there more skilled people for these positions (like in academia), or more positions than skilled people?

This is a complicated question. On the one hand the field seems to be exploding now and there are potential applications in practically all areas of human activity. There certainly aren't enough qualified people in the world right now to even scratch the surface of what can be done.

But that doesn't necessarily equate to there being a lot of jobs. I don't know what the job market is currently like but in 2011 it certainly wasn't easy for an experienced person to get a job in this field (it might have been easier for a young person with strong academic qualifications, however).

The problem is that developing a machine learning application to the point of generating revenue is typically very difficult and can take years. From the point of view of most investors it makes more sense to put money into developing a mobile app or a SaaS or something technically much simpler. Also even if you do want to found a machine learning company finding enough skilled people is probably not easy.