A Test Score-Based Approach to Stochastic Submodular Optimization

We study the canonical problem of maximizing a stochastic submodular function subject to cardinality constraint, where goal is select subset from ground set items with uncertain individual performances maximize their expected group value. Although near-optimal algorithms have been proposed for this problem, practical concerns regarding scalability, compatibility distributed implementation, and expensive oracle queries persist in large-scale applications. Motivated by online platforms that rely on item scores content recommendation team selection, we special class based solely performance measures known as test scores. The central contribution work novel systematic framework designing score–based broad naturally occurring utility functions. introduce new scoring mechanism refer replication prove long objective satisfies diminishing-returns condition, one can leverage these compute solutions are within constant factor optimum. then extend mechanisms more general welfare-maximization partition into groups sum values. For difficult show be used develop an algorithm approximates optimal solution up logarithmic factor. techniques presented bridge gap between rigorous theoretical optimization simple, scalable heuristics useful certain domains. In particular, our results establish many applications involving selection assignment items, design intuitive practically relevant only small loss compared state-of-the-art approaches. This paper was accepted Chung Piaw Teo, optimization.