COLLABORATIVE TRAINING IN SENSOR NETWORKS: A GRAPHICAL MODEL APPROACH

One of the advantages of modeling distributed inference problems in sensor networks as inference problems on graphical models is to find communication-efficient “messages” that are exchanged among the sensors. In many adhoc algorithms for distributed inference problems, the messages transmitted among the sensors are problem-specific. If we can successfully model these problems as graphical models, the messages exchanged among the sensors turn out to be exactly the messages specified by the corresponding graphical model message-passing algorithms such as the sum-product algorithm. Another issue, much more important in the area of wireless sensor networks than in graphical models, is the communication cost. In wireless sensor networks, communication is usually constrained and expensive, unlike in centralized