Maximum-Likelihood Registration in Networks
A key aspect of integrating and ultimately exploiting information collected across a distributed network of assets is establishing and maintaining synchronization across the nodes of the network. In this presentation, a statistical framework to allow foundational issues in this type of problem to be addressed a rigorous fashion. While the work is applicable to a broad class of problems involving synchronization or registration of data across a sensor network in the presence of noise, it is not a panacea; rather important and difficult challenges involving disparate data types, occlusion, model mismatch, and other characteristics of real-world applications remain. Nevertheless, this framework enables an estimation-theoretic approach to the design and characterization of synchronization algorithms that can play a role in larger fusion problems. The Fisher information is expressed in terms of the distribution of the measurement noise and standard algebraic descriptors of the network’s graph structure for several important cases. This leads to maximum-likelihood and approximate maximum- likelihood registration algorithms and also to distributed iterative algorithms that, when they converge, attain statistically optimal solutions. The relationship between ML estimation in this setting and Kirchhoff’s laws is also elucidated.
This is joint work with Steve Howard and Bill Moran.
Biography -- Doug Cochran is an applied mathematician with appointments in the School of Electrical, Computer and Energy Engineering and the School of Mathematical and Statistical Sciences at ASU. He holds S.M. and Ph.D. degrees in applied mathematics from Harvard University and degrees in mathematics from MIT and the University of California, San Diego.
(Refreshments will be served.)