"Learning all the advanced math within a few years is a hopeless endeavour."
Not in my experience. It's possible to get the equivalent of a bachelors and masters in math within two years (which is enough to overcome the issues listed in the post), but it's all you'll be doing for that period of time. Well worth it imo.
> Not in my experience. It's possible to get the equivalent of a bachelors and masters in math within two year
Seriously.
In fact, most people who learn this math learn it in a period of 2-3 years. Far from being impossible, learning this math in a few years is normal. It's not even a full time job. Most people learn all this math while also doing other classes and school stuff. Even a very dedicated math major probably only spends 20-25 hours a week actually studying math. I'm not sure much more than that is sustainable for most people anyway.
Now, I'll grant, this is going to be a lot harder to do without the structure of well thought out syllabi and lectures, but it's certainly manageable.
If you're already a grad student you can usually take any undergrad course at your institution for free. It will slow down your progress on your graduate degree, but you might as well do it right if it's what you really want.
If you want to learn decent undergrad math you really dont even need that much. 1yr analysis, 1yr algebra, topology. Then maybe a combination of a more indepth linear algebra course, an optimization course, a probability course, a stats course. You could probably do all that and a few more electives in one year full time.
From what I've heard from some foreign students, the 400 level American undergrad courses are what they are expected to learn as freshmen.
> It's possible to get the equivalent of a bachelors and masters in math within two years (which is enough to overcome the issues listed in the post), but it's all you'll be doing for that period of time.
About that - I don't think OP (or their research group) can afford to hold off publishing and attending ML conferences for 2 whole years while brushing up on math.
The ratio of explore /exploit depends on what your goals are. If it's just to bang out a PhD in applied ML then perhaps, but if you're in it for the long haul it's well worth it. It would unlock a whole bunch of research directions and catching back up to the field with solid math skills in hand would be quick. If you pay attention to the leaders in AI, they're mostly applied mathematicians in disguise.
Not in my experience. It's possible to get the equivalent of a bachelors and masters in math within two years (which is enough to overcome the issues listed in the post), but it's all you'll be doing for that period of time. Well worth it imo.