A collaboration-driven mechanism for AI diagnose with multiple requesters under incomplete information

In this paper, we design a collaboration-driven mechanism with multiple requesters cooperating together to collect massive accurate data and minimize costs in AI diagnose. Under the process of this incentive mechanism, two issues, the accuracy level of data and how to encourage requesters to cooperate, has been solved. A novel payment policy, with which the most accurate data gets the highest payment, is presented to assure the accuracy level of data. A Stackelberg game, which describes the interactions between requesters and workers, is put forward to stimulate multiple requesters jointly collect data. Considering requesters’ selfishness, Stackelberg equilibrium with multiple leaders are presented to maximize both requesters’ and workers’ utilities simultaneously. A cost allocation method inspired by separable costs remaining benefits method is proposed to ensure the increase of requesters’ utilities when collecting data as a team. However, since the worker’s (mobile device users’) privacy information is unknown, the game mentioned above is incomplete information dynamic MCS game. Hence, Q-learning is used to learn from historical data to estimate the unknown private data of workers. Several simulations are presented to prove that Q-learning can obtain theoretical solutions independent of parameters in the case of incomplete information.