[HN Gopher] Surprising limits discovered in quest for optimal so...
___________________________________________________________________
Surprising limits discovered in quest for optimal solutions
Author : theafh
Score : 29 points
Date : 2021-11-01 14:22 UTC (8 hours ago)
(HTM) web link (www.quantamagazine.org)
(TXT) w3m dump (www.quantamagazine.org)
| solarmist wrote:
| This result seems bizarre.
|
| You can always increase the dimension of an equation in trivial
| ways.
|
| Why would injection of a dummy variable into a quadratic, to make
| it cubic, make it more tractable?
|
| *edit: a couple of notes from the article that point towards
| hints, maybe?
|
| "Could they do it for cubic optimization problems that carry the
| simplifying condition that they incorporate no constraints?"
|
| "Knowing that both "apple" and "orange" appear would make you
| more confident, but you could still be wrong (Orange is a company
| too). A third term like "melon" introduces a cubic relationship
| and might allow you to confidently conclude the text is talking
| about produce."
|
| Seems like it's related to context, there maybe too little
| context in a quadratic equation to optimize it well.
| adrian_b wrote:
| The results are not about the number of variables, but about
| the degree of the polynomial that is the optimization target.
|
| To increase the degree of a polynomial, adding a dummy variable
| does not help.
|
| You need to add at least one term to the polynomial that is the
| product of 3 variables.
|
| I do not see how you can add such a term without changing the
| value of the function that must be optimized. If you add a
| dummy variable and you add a term that is the product of 2
| variables with the dummy variable, then the optimizer will find
| an optimal solution for a non-zero value of the dummy variable,
| which will offer no hint about the optimum for the original
| polynomial.
| TuringTest wrote:
| My company makes a living by algorithmically finding solutions to
| complex applied mathematical problems.
|
| Theoretically, the best thing to do is to find an optimal
| solution to the problem -even if it is a local optimum- as this
| means that you can stop looking for more.
|
| However, in practice, the user in many cases will not care
| whether the algorithm will approximate the optimal solution a few
| more decimal places; what matters is that the solution does not
| have any major drawbacks, so that it can be safely used for its
| intended purpose; and that it is better of what one expert could
| find in the same time by hand. That's the reason why approximate
| algorithms are used so often.
|
| Under those constraints, you're better off with an anytime
| algorithm with a good heuristic, which gives a viable good-enough
| suboptimal solution fast, and which can be left running while the
| expert analyzes the best solution found so far, to decide if it
| is acceptable.
___________________________________________________________________
(page generated 2021-11-01 23:02 UTC)