A Randomized Gossip-based Algorithm for Classification on Peer-to-Peer Networks

Peer-to-Peer (P2P) networks are distributed systems in which nodes of equal roles and capabilities exchange information and services directly with one another. In recent years, they have become a popular way to share large amounts of data. Such architectures, however, complicate the process of knowledge discovery and data mining since algorithms must deal with distributed (and often) dynamic sources of data and computing. In this paper, we present a distributed algorithm for learning linear classifiers in P2P networks. The problem is posed as a linear program such that each peer has its own constraints, but needs to solve a global objective function. A randomized-gossip based approximate algorithm is presented which reduces communication cost in the network significantly while ensuring convergence of the algorithm.