Skip to content

Machine Learning Methods for Inference

Funded by ONR, NSF, RAE (Swedish NSF)


machinelearning

We exploit and expand upon recent statistical methods such as compressed sensing, sparse approximation and low rank matrix completion to solve long-standing problems in wireless communications and localization.  We have designed new optimization problems based on the unique physics of special communication systems (such as underwater acoustic channels, vehicle-to-vehicle channels, ultrawideband signaling, etc.) as well as proposed new constraints such as unimodality and linear and algebraic structures into optimization.  These new constraints necessitate novel optimization methods and new analysis;  our methods have enabled the design of multi-modal (tensor) sensing and inference solutions.

The solution strategies are semi-parametric and can be employed in a wide variety of applications. Our methods have enabled high performance, moderate complexity strategies that have been validated on experimental data.  New applications include optimal unmanned aerial vehicle placement and system modeling and designs for autonomous underwater vehicle systems.