A Rigorous Statistical Framework for the Mathematics of Sensing, Exploitation and Execution
Final rept. 24 Sep 2011-28 Feb 2015
DUKE UNIV DURHAM NC DEPT OF ELECTRICAL AND COMPUTER ENGINEERING
Pagination or Media Count:
A research program has been performed to develop next-generation mathematics for sensing, exploitation and execution MSEE. An important focus of the research has been on a new class of nonparametric Bayesian architectures that constitute a rich modeling framework while still yielding parsimonious representations. Such models are attractive from multiple perspectives i they flexibly adjust model complexity and sophistication to match the observed data, while ii explicitly defining model uncertainty manifested by missing data, and thereby iii linking utility of data to the objectives and associated models additionally, iv these models are ideal for joint modeling of heterogeneous and possibly contradictory data, by sharing an inferred and typically low-dimensional latent space. In the MSEE construct, the utility of data is linked to the sensing objective, which in turn motivates and refines the associated models.
- Statistics and Probability