No, it's a reality of the labor market for doctorate holders. Startups are experiencing a shortage of coders, but the supply of CS doctoral students is robust. Just look at how quickly internships/positions at industry labs (AT&T, Yahoo, Microsoft) fill up.

Not all computer vision is "low-hanging fruit," especially if you're pushing boundaries. Similarly, applying ML techniques to "other data that is relevant" is a far cry from using ML to save lives at a hospital due to misdiagnosis. I'm not talking about regressing A/B testing results.

Computer vision is low hanging in the sense that you already have everything you need to make positive contributions to the study. I think the same is true with ML in general, but it was specific about what type of ML, which hangs higher due to the data availability.

> Similarly, applying ML techniques to "other data that is relevant" is a far cry from using ML to save lives at a hospital due to misdiagnosis.

Are you saying that programmers that are not working directly on saving lives are essentially wasting their time? There are a lot of interesting ML problems that do not save lives, but they are still worth working on.

Sure, but to be fair to achompas, that isn't what he said. What he said was

"(3) push the frontier of computer vision in concert with former telecom engineers."

Pushing the frontier of CV (as distinct from implementing/applying some CV algorithms) is hard to do outside a university or industrial research lab. Without formal CS training, it is very hard (Not impossible, but very hard) to access those problems.

My point is that you simply cannot generalize. You have absolutely no idea what talents someone has just by looking at their history. It is simply irrelevant information if you want to hire the best of the best.

Look, I appreciate your attitude. You have a positive outlook on what you can accomplish, and that's undoubtedly a good thing.

But, to be frank, the skills required to access my example problems above are not trivial. Let's consider the autodidactic route for computer vision:

1. You need to be cozy with linear algebra, convex optimization, calculus, and algorithmic complexity if you even want to understand prior research. This, alone, is 1-2 semesters of course load for a full-time student.

2. Then, you need to survey prior research to gain awareness about what already exists. You'll hit Google Scholar, search for papers, and have to circumvent article paywalls.

3. After that, you'll need to code your own framework (non-trivial) or convince other researchers to share their source code (very non-trivial--almost impossible, given that they might monetize or license their work, or their university owns said license).

4. Then you need to collect data to test your CV algorithm, iterate on it, etc.

Universities overcome all of these barriers. Hence, it is unrealistic to suggest one can produce cutting-edge CV work by themselves (or without university help).

2. You have to spend the time doing the research no matter who you are. Alternatively, you can ask someone else. Either way, anyone can do it.

3. This is a fair point, but you are allowed to spend money. If it costs money to access that code, so be it. Students are paying for that access too.

4. Again, true of anyone.

But more to the point, who cares how someone achieved their accomplishments? If it was through college, great, if it wasn't, still great. Why are you immediately discounting the person who did something amazing, just because he did it by himself?

Edit: I confused you with another poster. You may not have been judging people on their past. I do agree that people are more likely to do that work in a school setting, but that remains irrelevant when it comes to hiring.

Thanks for retracting that. :)