Parameter estimation with Metropolis-Hastings algorithm

Abstract:

Hydrology still lacks physics-based equations at various spatio-temporal scales. Thus, resorting to empirical equations is inevitable.  

These equations have to represent complex processes, which are yet to be understood properly. Thus, we sometimes end up with non-linear statistical equations with parameters to be estimated. Though expert knowledge helps to estimate the range of these parameters, estimating them from a large space (let us say you have 3 parameters with parameter space of 20, 1000, and 50 respectively, that’s 1 million parameter possibilities already!) can be a daunting task. Especially if your statistical model requires lots of computational time and resources.  

In this seminar I shall present a parameter estimation technique, Metropolis-Hastings algorithm, with the particular type being Delayed Rejection Adaptive Metropolis. The algorithm is a type of Markov Chain Monte Carlo method for random sampling. It is widely used in all kinds of fields and applications. The algorithm draws samples intuitively to converge to best possible estimate in the shortest time possible. 

Date/Time
Date(s) - 19/01/2024
4:00 pm - 5:00 pm