Joint Research Conference
June 24-26, 2014
Measurements from common measurement devices can sometimes be biased or “off target. For example, a balance may make measurements that are systematically too heavy relative to the actual weight of the measurand, or the speedometer of a car may consistently measure the true speed as too slow. In these cases, a simple additive term may be estimated to account for this discrepancy. In general, the bias induced by the measurement device may not be so easily characterized as it may have a more complex form or be dynamic. In these cases, a simple parametric form cannot be easily found. In this talk, we present methods for quantifying and adjusting for a “wandering bias” from measurements using Gaussian process discrepancy terms. This talk describes research in progress in collaboration with Richard Warr of AFIT, Christine Anderson-Cook of LANL, David Higdon of LANL.