Thursday, November 17, 2016 - 3:00pm
209 W. Eighteenth Ave. (EA), Room 170
Statistical Model Relaxations
David Campbell, Simon Fraser University
When scientists give statisticians mechanistic models they usually come with challenges. When the models are differential equations, mechanistic models are simple to interpret and often have few parameters but once integrated into the data space, parameter estimation becomes complicated by attempting to strictly adhere to the model. Other times parameters are only interpretable when inequalities hold, or the data fit only makes sense when the model is monotone. Often, optimization or Bayesian sampling methods are severely impacted by constraints induced by the model. However, when the constraints are relaxed, the algorithms are fast and easy to apply.
This presentation introduces various ways to exploit model relaxations with emphasis on Sequential Monte Carlo methods developed with Shirin Golchi.