Multi-factor incentive mechanism for federated learning in IoT: A Stackelberg game approach

In  the  era  of  the  Internet  of  Things  (IoT),  remotesensors and endpoint appliances generate vast amounts of data.Decentralized and collaborative learning builds on these IoT datato enable classification and recognition tasks by inviting multipledata owners. Federated learning (FL), as a popular collaborativelearning  framework,  can  significantly  improve  the  performanceof  models  without  collecting  the  original  data.  To  invite  dataowners  to  participate  in  FL,  various  incentive  mechanisms  aredesigned  to  address  this  issue  by  researchers.  However,  existingsolutions  still  face  high  costs  and  low  utility  due  to  informationasymmetry,  where  the  reputation,  computation  power,  and  dataquantity of the data owners are not known in advance. Therefore,we  propose  a  Stackelberg  Game-Based  Multi-Factor  IncentiveMechanism for Federated Learning (SGMFIFL). First, we designthe  Top-K  cost  selection  algorithm  based  on  reverse  auction,which   can   reduce   the   cost   of   selecting   data   owners.   Next,we  devise  a  multi-factor  reward  function  based  on  reputation,accuracy, and reward rate, the data owners with high reputationand  high  accuracy  will  be  of  more  reward.  In  particular,  toensure  that  SGMFIFL  can  provide  reliable  incentives  in  IoT,we use blockchain to provide a secure and trusted environment.Finally,  we  construct  a  two-stage  Stackelberg  game  model  forthe  task  publisher  and  the  data  owners  and  derive  an  opti-mal  Equilibrium  solution  for  both  stages  of  the  whole  game.Experiments  conducted  on  two  well-known  datasets,  MNISTand  CIFAR10,  demonstrate  the  significant  performance  of  theproposed mechanism.