Hacker News new | ask | show | jobs
by Skaven 2069 days ago
Most methods easily find a local optimum, but it is very hard to tell if it is a global optimum.
1 comments

I disagree -- in industry, it doesn't directly matter if you can globally optimise some theoretical objective function or not, the objective function is always a crude approximation of what's really going on in reality, incorporating a bunch of assumptions, many of which won't be exactly true.

What matters is if you can produce performance at doing a thing that is sufficiently better than the status-quo performance of doing that thing. Things that be completely uninteresting and sub-optimal from a theoretical optimisation perspective and yet provide a lot of real world value and result in successful projects.