Strona główna » ZA » Seminars » Seminars in 2019 »

# dr Agnieszka Borowska, University of Glasgow, Efficient statistical methods for computationally challenging problems.

Research in physics often starts from a mechanistic model

and then aims to understand how different parameter regimes are related

to certain features in the data. This type of research focuses on the

model and views features in the data as emergent properties. Statistics

follows the opposite direction: given the data, the aim is to find, or

"infer", the parameters of the model that are most consistent with the

observations. In this talk I will focus on three different applications

in which statistical inference is computationally challenging for

various reasons and I will show how state-of-the-arr methods from

computational statistics can come to our rescue. First, I will discuss

optimising physiological parameters of a bio-mechanical model of the

left ventricle for which standard gradient-based optimisation schemes

are prohibitively time consuming. Second, I will talk about a stochastic

system describing cell movement with the outputs of the associated

simulator being very high-dimensional. Third, I will speak about risk

evaluation, or estimating the probability of a rare event, for which

simulating from the model directly typically does not lead to improved

insights on the event in question.