Would be great if experienced folks can add in-depth comments.
What about machine learning distribution?
I worked in ML and my output for the half was a 0.005% absolute improvement in accuracy. It was considered good.
The ideal is some combination of distributed systems and deep learning in a user facing product. There’s probably a handful of teams in the whole of tech that do this though.
Couldnt agree more. But such teams will most probably stay closer to headquarters. Folks in other locations might rarely get a chance to work on such stuff. Might be possible 5 years down the line.
What about distributed machine learning?
I wanted to keep a line of demarcation as clear as possible. So didn't add that option. At the end of the day, ML for large scale has to be distributed and distributed systems will take advantage of ML to perform better.