Did you know that it could take so long when you first started your PhD in Korea? Could this be one of your reasons to move to Japan?
My decision to start my PhD in Korea was under very different circumstances. I was studying a totally different field (automata theory, so closer to mathematics than ML) where longer PhDs were normal, I was used to the US system, I wasn't working or married, etc. I expected it to take ~6 years when I started.
I basically had to choose between living in the Canadian countryside, Helsinki, or Seoul in order to be in a lab studying what I wanted. I was already pretty into Korean culture (I minored in Korean in college, studied abroad there, and was a serious StarCraft player) so that choice was easy!
When I started looking again, I knew I wanted a shorter PhD (especially since I had a masters and industry + industry research experience), so Japan was pretty attractive.
I will admit that I think 3 years is a little short to make substantive progress outside of machine learning, so unless you come in with a lot of research experience and a good idea of what you want to work on, I think a longer PhD would be beneficial, but the opportunity cost is so high that it really doesn't seem like a great option on paper.
One aspect worth calling out as a huge plus of Japanese national institutions which the author didn't disclose is funding for compute is much better than what one would get from a Korean institution (e.g. the author's school runs one of the biggest GPU clusters in Japan).
> at Google I work primarily on on-device language models for keyboards.
It's difficult to reconcile the amount of horsepower your typical keyboard, and a language mode, let alone how a language model might be useful in a keyboard (the mind boggles at the thought it starts correcting your typing).
What are the potential uses of a language model in a keyboard?
"In essentially every other category, my quality of life has degraded significantly since returning home. "
This is the challenge. The market in the US, the salaries, etc. are significantly better/higher, but the quality of life is often worse.
Now that I think about it, one thing I am not sure about is if self-funded students' travel/conference fees were covered. I would guess they are covered through a grant and it is just the tuition that is self funded, but I do not know for certain. Going to >1 international conference a year would be far more than the tuition.
So, I would guess it would cost ~10k/year to be fully self funded. You can recoup some of that cost working (on a student visa this is limited though), and your lab may also have an RA stipend (but this is both very low [<$20/hr] and counts towards your 28 hour/week limit).
My advice though is to never do a self-funded PhD. Being a PhD "student" is a misnomer. You are a research employee and should be compensated as such.
And while I completely agree that a PhD "Student" is a working job that should be compensated. We both know that certain doors are much easier opened when you already have a PhD. I have the financial means to help me accomplish this without taking realy much of a financial toll on it. Hence why it even remains an option.
Your experience mostly tracks with mine, but in my case there was a lot of interaction between Japanese and foreign students, and international journal and conference submissions were preferred. Many courses were offered in English and those that weren't, I could do homework and tests in English. Thesis had to be in English, even for Japanese students. It was a goal of the program to also help Japanese students get comfortable with international collaboration and working in English.
In my case I stayed in Japan for a decade and it was overall a very good experience.
One note about working as a student in Japan. You can work up to 28 hours a week without limitations as you noted. You can even work remotely for a foreign company, which is unique because if you have a work visa and freelance in Japan either as your main job, or a side job, your main client is supposed to be Japanese. This is a slightly grey area, your taxes are to be paid in Japan, and the company will be "operating in Japan" so be very careful about doing this by the books.
One thing to note is that, after I submitted the final draft of this article, the Japanese government ended the MEXT university track or is at least planning to. There was a memo, but the status is not clear.
Anyway, I had a great time doing my PhD this way, and generally I recommend that people work before a PhD and try to work during it (if you are in CS). It gives you a lot of freedom and, at least for me, it is very helpful to have more than one pipeline of work at a time so that when one slows down (as it inevitably does in research), you can draw on the other.