A democratically collaborative learning scheme for fog-enabled pervasive environments

The emergence of fog computing has brought unprecedented opportunities to many fields, and it is now feasible to incorporate deep learning at the edge to facilitate the development of pervasive systems (e.g., autonomous driving and smart grids). In this paper, we present our preliminary research on a democratic learning scheme so that fog nodes could collaborate on the model training process even without the support of the cloud, which is urgently needed in the pervasive computing context. The main objective of this work is to utilize the deployed fog nodes to train a well-performed deep learning model together, even with the limited local data from each participant. Instead of relying on the cloud by default, we design a voting strategy so that a fog node could be elected as the coordinator based on both distance and computational power metrics to help expedite the training process. We then experiment the effectiveness of the scheme through a real-world, in-door fog deployment and verify the performance of the trained model through a human moving trajectory tracking use case.