Location
Online
Event Website
https://hicss.hawaii.edu/
Start Date
3-1-2023 12:00 AM
End Date
7-1-2023 12:00 AM
Description
Agent-Based Models (ABMs) have long served to study self-adaptive systems and the emergence of population-wide patterns from simple rules applied to individuals. Recently, the rules for each agent have been expressed using a Fuzzy Cognitive Map (FCM), which is elicited from a subject-matter expert. This provides a transparent and participatory process to externalize the `mental model' of an expert and directly embed it into agents. However, software technology has been lacking to support such hybrid ABM/FCM models at scale, which has drastically limited the scope of applications and the ability of researchers to study emergent phenomena over large populations. In this paper, we designed and implemented the first open-source library that automatically accelerates ABM/FCM models by leveraging the CUDA cores available in a Graphical Processing Unit. We demonstrate the correctness and scaling of our library on a case study as well as across different networks representing agent interactions.
Recommended Citation
Ghumrawi, Kareem; Ha, Kim; Beerman, Jack; Rudie, John-David; and Giabbanelli, Philippe, "Software Technology to Develop Large-Scale Self-Adaptive Systems: Accelerating Agent-Based Models and Fuzzy Cognitive Maps via CUDA" (2023). Hawaii International Conference on System Sciences 2023 (HICSS-56). 3.
https://aisel.aisnet.org/hicss-56/st/self-adaptive_systems/3
Software Technology to Develop Large-Scale Self-Adaptive Systems: Accelerating Agent-Based Models and Fuzzy Cognitive Maps via CUDA
Online
Agent-Based Models (ABMs) have long served to study self-adaptive systems and the emergence of population-wide patterns from simple rules applied to individuals. Recently, the rules for each agent have been expressed using a Fuzzy Cognitive Map (FCM), which is elicited from a subject-matter expert. This provides a transparent and participatory process to externalize the `mental model' of an expert and directly embed it into agents. However, software technology has been lacking to support such hybrid ABM/FCM models at scale, which has drastically limited the scope of applications and the ability of researchers to study emergent phenomena over large populations. In this paper, we designed and implemented the first open-source library that automatically accelerates ABM/FCM models by leveraging the CUDA cores available in a Graphical Processing Unit. We demonstrate the correctness and scaling of our library on a case study as well as across different networks representing agent interactions.
https://aisel.aisnet.org/hicss-56/st/self-adaptive_systems/3