Dr. Philipp Andelfinger,
Institute for Visual and Analytical Computing
University of Rostock,
Germany
Title: Large-Scale Simulations of Agent-Based Systems: Parallel, Heterogeneous, Differentiable
Abstract
Agent-based modeling and simulation is widely used to study, evaluate, and design systems in domains such as transportation, social sciences, and biology. To keep pace with the computational demands of increasingly detailed and large-scale models, simulators must exploit parallel and distributed hardware architectures while aiming to avoid unnecessary computations. After contrasting common time advancement schemes and their caveats, this tutorial discusses methods for achieving causal correctness, determinism, and efficiency in simulations of tightly coupled agent-based systems using fixed time steps or continuous time. A key focus is on the use of hardware accelerators in order to fully exploit the heterogeneous resources available in today's workstations and compute clusters. As an alternative to the use of additional hardware resources, we cover methods to accelerate simulations by reducing the amount of computation required to accomplish a simulation study's objectives, including the exploitation of locality in agent interactions and the construction of differentiable models to enable the use of efficient gradient-guided optimization methods.
Bio
Philipp Andelfinger is a postdoctoral researcher at the Institute for Visual and Analytical Computing at University of Rostock, Germany. Previously, he was a Research Scientist at Oak Ridge National Laboratory (ORNL), TN, USA, and a research fellow at Nanyang Technological University (NTU), Singapore. He received his PhD in 2016 from Karlsruhe Institute of Technology (KIT), Germany. His research interests include agent-based modeling and simulation, distributed systems, parallel and distributed simulation, and simulation-based optimization. He has co-authored around forty publications in these and adjacent fields. His works received Best Paper Awards at IEEE WSC 2020, ACM SIGSIM PADS 2018, IEEE/ACM DS-RT 2018, IEEE ATC 2016, and IFIP/IEEE DISSECT 2015. Dr. Andelfinger has been a program committee member of ACM SIGSIM PADS since 2019 and serves as referee for journals including ACM TOMACS, IEEE TPDS, Elsevier SIMPAT, SCS SIMULATION, and T&F Journal of Simulation.