Efficient Free Rider Discovery and Punishment: Simple InFreD

There are many distributed collaborative systems like Peer-to-Peer (P2P) networks developed so far which do not rely on external servers to bring required resources for their users. Peers in these systems not only download files from the server, but also upload them to other peers, mitigating the server’s burden. Discovery and prevention of free-riders is a key means to ensure performance metrics like fairness. We propose a robust, asynchronous, gossip based protocol that can withstand high churn and failure rates, making it ideal for peer-to-peer networks that prevent free-riders successfully as long as better neighborhood selections. In this setting, each peer trains on their local training examples which could be very few and pass along the trained models to their neighbors. The shared models learn about the environment of peers and dynamically change the overlay network links to choose better neighbors and provide long-term contribution incentives. We describe the details of implementing our algorithm, and discuss our experimental evaluations. It is still an unresolved problem how to generate complex learning models to cover other fairness metrics rather than free-riding in these systems

Keywords:

Distributed intelligence , Gossip Learning , Free-Riding , Incentive Mechanism , Fairness