A Marketplace for Data: An Algorithmic Solution
summary
Proposes a real-time data marketplace where M fixed sellers supply features, buyers arrive online, allocation gives a buyer feature when bid exceeds a public price, and payments use Myerson payments equal to incremental accuracy gains. Revenue shares use a sampled Shapley allocation but add a distance-based penalty to stop sellers from profiting by duplicating datasets. Proves a trade-off: robustness to replication and exact budget balance cannot both hold, shows submodular accuracy yields convex price updates, and reports prototypes for retail and energy use.
AI-written summary · about