Federated learning privacy incentives: Reverse auctions and negotiations

Abstract The incentive mechanism of federated learning has been a hot topic, but little research done on the compensation privacy loss. To this end, study uses Local SGD federal framework and gives theoretical analysis under use differential protection. Based analysis, multi‐attribute reverse auction model is proposed to be used for user selection as well payment calculation participation in learning. mixture economic non‐economic attributes making choices users transformed into an optimisation equation solve choice problem. In addition, post‐auction negotiation that Rubinstein bargaining equations describe process theoretically demonstrate improvement social welfare proposed. experimental part, authors find their algorithm improves both accuracy F1‐score values relative comparison algorithms varying degrees.