The three articles in this special section are extensions of outstanding papers that appeared in the 2021 ACM Conference on Principles of Advanced Discrete Simulation (PADS), the annual flagship conference of the Special Interest Group for Simulation and Modeling SIGSIM. Like 2020, PADS 2021 was conducted virtually, as we were still under the special constraints of the COVID-19 epidemic. Nonetheless, the overall attendance went up, and the conference committee had a good set of candidate papers. This special section is edited by the event's co-chairs, Saikou Y. Diallo from Old Dominion University and Andreas Tolk from the MITRE Corporation. However, they received support in selecting the final papers and organizing this special section from Francesco Quaglia, Editor-in-Chief of ACM TOMACS, and Philippe Giabbanelli from Miami University. Alessandro Pellegrini from the National Research Council and Till Köster from the University of Rostock provided additional advice. The selection process took the original conference paper, its presentation, and the feedback from the judges and session chairs into account. The authors of the identified three papers were invited to provide substantially extended versions of the conference paper, which have been peer-reviewed to ensure the expected academic quality. These three papers were already top ranked by the technical program committee of PADS 2021. However, they were newly evaluated by recognized experts in the field whom the special issue editors had recruited. The full selection and peer review process took several rounds and several months. The resulting three journal articles presented here truly represent the state-of-the-art on advanced distributed simulation.
This collection of articles highlights the diversity and versatility of computer simulation. The special issue grapples with the ability of simulations to run very large entity counts as fast as the hardware allows while maintaining sequentiality. The ability to successfully combine speed, scale, and repeatability opens the door to new worlds only imagined a few decades ago. One such world is the promising combination of data-driven methods with agent-based simulation. This mix of empirical, statistical, and algorithmic modeling into a single agent-based framework allows the exploration of real-world applications that have proven to be challenging to tackle within a single school of thought. Some of those tough problems, such as climate, traffic, and human mobility, are even more challenging to explore at scale without assumptions and compromises, which lead to biases. Each article makes a unique contribution to the field of Modeling and Simulation, yet together, the articles show the indivisible connection between hardware, software, simulation, data, and method. We briefly summarize each article and hope the reader applies the findings and uses the lessons learned for the research.
The first article discusses the Performance Analysis of Speculative Parallel Adaptive Local Timestepping for Conservation Laws and is authored by Maximilian Bremer, John Bachan, Cy Chan, and Clint Dawson. The paper proposes a novel heuristic to address the Courant-Friedrichs-Lewy (CFL) condition. The authors demonstrate that the heuristic is robust and provides significant gains in speed over state-of-art approaches. The paper contributes to the body of knowledge in massively parallel and high-performance simulations community.
In the second contribution Towards Differentiable Agent-Based Simulation, Philipp Andelfinger presents automatic differentiation (AD), a family of techniques to address gradients of general programs. The author discusses AD in the context of time-driven agent-based simulations and studies the differentiable simulations' fidelity and overhead. The author shows that the proposed approach supports gradient-based training of neural network-controlled simulation entities embedded in the model logic and demonstrates that the performance overhead of differentiable agent-based simulations can be reduced substantially by exploiting sparsity in the model logic.
Dynamic Data-driven Microscopic Traffic Simulation using Jointly Trained Physics-guided Long Short-term Memory is authored by Htet Naing, Wentong Cai, Nan Hu, Tiantian Wu, and Liang Yu. This paper was selected as the best paper of the PADS 2021 conference. The paper addresses the co-mingling of data-driven methods with traditional simulation systems and their effectiveness in supporting just-in-time (JIT) operational decision-making. Applying such approaches to Digital Twin City, real-time microscopic traffic simulation is growing but challenging concerning bias which can manifest in dangerously unsafe practices. The paper describes a physics-guided data-driven modeling paradigm that addresses these challenges. The authors conduct several experiments to show the feasibility and reliability of their approach.
PADS’ focus on methodological contributions to the principles of advanced distributed simulation embraces a wide variety of interests in their application to contribute to the solutions of real-world challenges. We hope this special issue encourages the next generation of scientists and researchers to contribute their findings, collaborate with scientists and engineers inside and outside the simulation community, and generate new and useful knowledge in their service to our society.
Index Terms
- Introduction to the Special Section on PADS 2021
Recommendations
PADS: a domain-specific language for processing ad hoc data
PLDI '05: Proceedings of the 2005 ACM SIGPLAN conference on Programming language design and implementationPADS is a declarative data description language that allows data analysts to describe both the physical layout of ad hoc data sources and semantic properties of that data. From such descriptions, the PADS compiler generates libraries and tools for ...
PADS: an end-to-end system for processing ad hoc data
SIGMOD '06: Proceedings of the 2006 ACM SIGMOD international conference on Management of dataEnormous amounts of data exist in "well-behaved" formats such as relational tables and XML, which come equipped with extensive tool support. However, vast amounts of data also exist in non-standard or ad hoc data formats, which often lack standard or ...
Comments