ta46290193

I suppose it's like software development generally. To be good at it you need a sense of what good programming feels like, but advanced books like Knuth's "The Art of Computer Programming" aren't really relevant to the day-to-day work of gluing together libraries and writing code that doesn't go deeper than for loops.

Similarly I suppose most ML work is done with some solid basics but advanced math textbooks aren't really needed, and people actually working at the theoretical advanced math level are rare and in a handful of academic positions or corporate research labs requiring high credentials or other specific qualifications.

Thanks for your input.

Do you know if the typical ML hiring process is able to distinguish between someone at the MOOC/"Machine Learning with [current popular language/library]" level from someone who's done hundreds of problems in a graduate-level math textbook? I'd love to be convinced because as I said I'd like the excuse to work through the book, but at the same time, as has been suggested elsewhere, there's more practical stuff to focus on too.

The number of technical jobs that actually require a math PhD must be relatively small. I'm on the other end of it, trying to avoid the situation where the degree type is meaningless, because this means that, as far as credentials go, you're competing with anyone with a college degree. I suspect in your case you have more room to maneuver than you think, but what do I know? Good luck to you.

My impression of ML has been that there are two types of jobs: one, the most common, where you're basically plugging values into tools like scikit-learn or TensorFlow and spending most of your day-to-day fussing around with data wrangling and general software development; and two, the heavy math jobs involving integral signs and Greek letters, which are much rarer and done mainly by people with top-tier PhDs at research labs.

Since at my age going for a Stanford PhD isn't an option, you probably mean by a better return on investment that I should dig into scikit-learn, Hadoop, the current AWS/Google Cloud/Azure options, that sort of thing. Is that right? Which, that's definitely sensible but not exactly what I was hoping for, since it's basically the same as the regular programming I've done, just with a different set of libraries. On the other hand it makes some use of my math background, and perhaps the dream of reliving grad school at a paid job isn't totally realistic anyway.