3 ms·
I assume you mean norm rather than absolute value because absolute value would give a vector result? But yes, but this is generally unhelpful since in general
by mehrdadn 8y ago
I assume you mean norm rather than absolute value because absolute value would give a vector result?
But yes, but this is generally unhelpful since in general the behavior at a given point would give you almost no information regarding behavior at a nearby point. It's a bit like doing the reverse by turning the objective into a feasibility test. e.g. min f(x) is like min 0 : {f(x) <= f(x') for all x'}... you can do it, but it's not really helpful.
- mathnmusic 8y agoI can see that it will be unhelpful in most cases (because we're losing information by implementing constraints as penalties). But are there any cases at all where this would make the solution easier?