Seminar by Prof. Michael Mascagni

December 20, 2016

Admission Free,No Booking Necessary

Seminar room 2 @ The Institute of Statistical Mathematics
Prof. Michael Mascagni
(Department of Computer Science, Florida State University, USA)
Reproducibility in Stochastic Simulation
In this talk we consider the issues of reproducibility for stochastic simulation. We define what stochastic computations are, and we first focus on random number generation as a key to controlling reproducibility. We look at different types of random numbers and how to make them reproducible. We then show how the Scalable Parallel Random Number Generators (SPRNG) library was designed partly to ensure absolute random number reproducibility. We then consider situations where even with SPRNG, reproducibility is problematic. We then introduce a much more modest level of reproducibility that we call "Forensic Reproducibility." We then consider another approach to reproducibility in stochastic simulations by noting that certain deterministic systems when computed in the presence of round-off error behave more like solutions to stochastic systems.