"Bayesian Calibration of Dielectric Charging Model for RF-MEMS Devices"
October 12 @ 12:00 PM - 1:00 PM - BRK 2001
Abstract:The Bayesian calibration approach allows for the inclusion of epistemic uncertainty sources (e.g. data and model uncertainty), and it also provides a framework for the aggregation of data of different types that may come from many different sources. Our approach is built upon the Kennedy-O’Hagan calibration framework, which includes a discrepancy term to account for inadequacy in the form of the physics model. The model discrepancy may include both bias and variance corrections, each of which may be either constant or varying with respect to the model inputs. Given that there are many possible choices for this discrepancy, we explore ways of selecting and evaluating among the available options. Our procedure utilizes existing validation metrics (e.g. Bayes factor and the model reliability metric) to inform the decision making process. The overall approach is illustrated using a dielectric charging model described in the previous seminar by Ravi Vedula and Sambit Palit. In particular, this example provides a basis for demonstrating the impact of discrepancy modeling choices.
- Bonnie Kauffman