Architecture-Aware Modeling of Pedestrian Dynamics.

Area of Science:

• Computational epidemiology
• Human behavior modeling
• High-performance computing

Background:

• Predicting infectious disease spread is challenging due to complex interactions between disease dynamics and human behavior.
• Current approaches often focus on relative policy risks rather than precise predictions.
• Simulating human behavior, particularly pedestrian dynamics, for policy analysis incurs significant computational costs.

Purpose of the Study:

• To develop a computationally efficient model for simulating human behavior in the context of infectious disease spread.
• To leverage computer architecture features for performance gains without complex parallelization.
• To enable faster risk assessment of different policy and design choices.

Main Methods:

• Developed a new computational model from scratch, considering computer architectural features.
• Utilized pedestrian dynamics to simulate individual movement in crowded locations.
• Performed parameter sweeps to generate numerous scenarios for human behavior under various policy options.

Main Results:

• Achieved a speedup factor of approximately 60 on two Xeon Platinum 8280 CPUs.
• Demonstrated a speedup factor of 125 on 4 NVIDIA Quadro RTX 5000 GPUs compared to traditional parallel implementations.
• The GPU implementation is optimized for real-time decision-making.

Conclusions:

• The novel model offers substantial computational performance improvements for infectious disease modeling.
• The approach is suitable for real-time analysis and supports informed policy decisions.
• The model's efficiency facilitates the evaluation of social distancing strategies and other public health interventions, as illustrated by a COVID-19 application.