Professor Raghu Pasupathy
On constructing confidence regions from generic simulation output
Modern simulation methods have dramatically increased our ability to analyse and predict the performance of time-dependent systems. For example, we are able to track the expected path of a hurricane's approach, estimate the 90th percentile travel time in response to congestion pricing measures in a city, and calculate the time-dependent heat flow across new materials as the solution to a stochastic PDE. In all of these environments, in addition to estimating a performance measure of interest, the simulationist routinely seeks some measure of uncertainty, e.g., a confidence region. What is more, the requested confidence region is frequently non-classical in the sense that the "system" being analyzed does not exist yet, and the performance measure being sought is not necessarily a mean.
I will start this talk by asking the question, what is meant by a confidence region on (an infinite-dimensional object such as) the expected path of a hurricane? And, is there a way to think about the previous question in general, so that the answer applies equally in the context of tracking hurricanes as it does when calculating a simulation-based heat flow solution? Then, standing on the steady progress made within the simulation community over the previous three decades, I will formalize a simple method, known to many simulationists through folklore, to construct confidence regions from generic simulation output. The method I present results in provably valid confidence regions and competes favorably with the famous bootstrap and subsampling techniques from classical statistics. It also provides the true analogues of the chi-square and the Student's t distributions --- fundamental distributions resulting from William Gosset's seminal 1908 paper --- for the simulation context.
Biography:
Raghu Pasupathy is a Professor of Statistics at Purdue University. Prior to joining Purdue in 2014, Raghu spent nine years in the Industrial and Systems Engineering Department at Virginia Tech, first as an Assistant Professor and then as an Associate Professor. Raghu's research interests lie in the theoretical and computational aspects of statistical and simulation-based inference, with a focus on stochastic optimisation problems. Raghu has been associated with the simulation and optimization communities in various capacities over the previous two decades, including serving as President of the INFORMS Simulation Society from 2018 -- 2020, and as an editor for ACM TOMACS, Operations Research, INFORMS Journal on Computing, IISE Transactions, and Mathematical Programming. More information, including downloadable papers and computational software can be obtained through his website at
https://web.ics.purdue.edu/~pasupath/ Catch up on our main sessions
Here, you can find a curated collection of videos from our previous Simulation workshop. Explore the sessions to gain valuable insights and knowledge shared by experts in the field.
Watch our SW23 playlist on YouTube here