Poster Abstract: Selective Gossiping for SVM Training in Wireless Sensor Networks

The design of an integrated distributed wireless network comprised of thousands of microsensors that can perform tasks with low energy consumption poses some interesting signal processing challenges. In this paper, we propose two “selective gossip algorithms” taking advantage of the sparse representation that Support Vector Machines (SVMs) provide for the decision boundaries, to perform classification in wireless sensor networks (WSNs). Through analytical studies and simulation experiments, we show that the proposed energy efficient, distributed selective gossip algorithms train SVMs equally well with energyconsuming centralized approaches that need to transmit the whole data to a processing node.