Edge Learning as a Hedonic Game in LoRaWAN
Résumé
Federated learning provides access to more data which is paramount for constrained LoRaWAN devices with limited memory storage. Learning on a larger data set will reduce the variance of the learned model, hence reducing its error. However, federating the learning process incurs a communication cost among learning devices that must be taken into account. In this paper, we formulate a Cooperative Hedonic game and introduce a new cost function that captures both the learning error and communication cost. LoRaWAN devices engage in the devised game by identifying if they should keep their learning local or federate with other devices in order to reduce both their learning error and communication cost. We compute the optimal size of formed coalitions and assess their stability. Then, we show through extensive simulations that devices have incentive to form learning coalitions depending on the data characteristics at hand and the communication cost in LoRaWAN.