On computing functions with uncertainty

We study the problem of computing a function <i>f</i>(<i>x</i><subscrpt>1</subscrpt>,…, <i>x<subscrpt>n</i></subscrpt>) given that the actual values of the variables <i>x<subscrpt>i</subscrpt></i>'s are known only with some uncertainty. For each variable <i>x<subscrpt>i</subscrpt></i>, an interval <i>I<subscrpt>i</subscrpt></i> is known such that the value of <i>x<subscrpt>i</subscrpt></i> is guaranteed to fall within this interval. Any such interval can be probed to obtain the actual value of the underlying variable; however, there is a cost associated with each such probe. The goal is to adaptively identify a minimum cost sequence of probes such that regardless of the actual values taken by the unprobed <i>x<subscrpt>i</subscrpt></i>'s, the value of the function <i>f</i> can be computed to within a specified precision. We design online algorithms for this problem when <i>f</i> is either the selection function or an aggregation function such as sum or average. We consider three natural models of precision and give algorithms for each model. We analyze our algorithms in the framework of competitive analysis and show that our algorithms are asymptotically optimal. Finally, we also study online algorithms for functions that are obtained by composing together selection and aggregation functions.