Can adaption help on the average?

Let f belong to a given class F of functions and for the points \(t_ 1,t_ 2,...,t_ n,\) let \(N(f)=[f(t_ 1),f(t_ 2),...,f(t_ n)],\) be an information, that we have about f. If the points \(t_ 1,t_ 2,...,t_ n\) are given simultaneously, then we call N a nonadaptive information. On the contrary, if \(t_ 1,t_ 2,...,t_ n\) are not given simultaneously, that is, if \(t_ 2\) depends on the previously computed value \(f(t_ 1)\), \(t_ 3\) depends on \(f(t_ 2)\) and so on, then N is called adaptive information. Generally, adaptive information is much richer than nonadaptive. From this one might hope that if someone uses adaptive information then he has much smaller error than if he uses nonadaptive information. But this is not generally true. There are many cases when the adaption does not help. More precisely, if, for example, the class F is convex and balanced (f\(\in F\) implies -f\(\in F)\), then nonadaptive information is as powerful as adaptive. In this work the authors study adaptive information for approximation of linear problems in a separable Hilbert space equipped with a probability measure \(\mu\). This measure \(\mu\) is used for the evaluation of error. They prove that adaption does not help, on the average, in the worst case for linear problems, too.

Recommendations

Cites work

Cited in

(34)

This page was built for publication: Can adaption help on the average?

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q760763)