Sampling schemes and recovery algorithms for functions of few coordinate variables

Abstract When a multivariate function does not depend on all of its variables, it can be approximated from fewer point evaluations than otherwise required. This has been previously quantified e.g. in the case where the target function is Lipschitz. This note examines the same problem under other assumptions on the target function. If it is linear or quadratic, then connections to compressive sensing are exploited in order to determine the number of point evaluations needed for recovering it exactly. If it is coordinatewise increasing, then connections to group testing are exploited in order to determine the number of point evaluations needed for recovering the set of active variables. A particular emphasis is put on explicit sets of evaluation points and on practical recovery methods. The results presented here also add a new contribution to the field of group testing.