Event details from: Known Unknowns and Unknown Knowns: system identification and model calibration of stochastic models

Date
12 March 2024
Time
11:00 – 13:00
Room
Sweelinck Room
Harsha Honnappa

Known Unknowns and Unknown Knowns: system identification and model calibration of stochastic models

The world of natural and engineered systems abound with phenomena exhibiting stochastic dynamics. These phenomena are often modeled using stochastic models, which, for the sake of simplicity and tractability, typically assume Markovianity and time homogeneity. Besides analytical tractability, these assumptions enable easier system identification and model calibration, areas that are well-explored within the research realms of statistics and applied probability.

However, these simplifications often don’t fully capture the reality of these systems. In truth, most natural and engineered systems exhibit a spectrum of complex behaviors, including dynamics that are both time-varying and non-Markovian. For instance, the effects of time-of-day, seasonal changes, and long-range trends are manifest in many such systems, contributing to the complexity of their behavior. I will use simple queueing and compartmentalized epidemic models as vehicles to introduce ‘grey box’ methods that combine stochastic models with deep learning to perform model calibration and system identification. Time-permitting, I will also discuss the challenges involved in calibrating and identifying stochastic models of spatio-temporally varying phenomena (for instance dynamics over networks) and on-going work providing theoretical guarantees for these methods.

Programme

11:00

Meet & Greet (coffee & cookies)

11:30

Welcome & introduction by Huub Dijstelbloem

11:35

Presentation by Harsha Honnappa

12:10

Q&A

12:30

Light Lunch