Robust probabilistic inference in distributed systems

摘要: Probabilistic inference problems arise naturally in distributed systems such as sensor networks and teams of mobile robots. Inference algorithms that use message passing are a natural fit for systems, but they must be robust to the failure situations real-world settings, unreliable communication node failures. Unfortunately, popular sum--product algorithm can yield very poor estimates these settings because nodes' beliefs before convergence arbitrarily different from correct posteriors. In this paper, we present new probabilistic which provides several crucial guarantees standard does not. Not only it converge posteriors, is also guaranteed principled approximation at any point convergence. addition, computational complexity updates depends upon model, independent network topology system. We demonstrate approach with detailed experimental results on calibration task using data an actual deployment.