Heterogeneous computing for epidemiological model fitting and simulation.

Area of Science:

• Computational epidemiology
• High-performance computing

Background:

• Substantial efforts are underway to enhance epidemic preparedness through interdisciplinary collaboration.
• This paper focuses on the computational aspects of infectious disease modeling, specifically utilizing graphics processing units (GPUs).
• Optimizing the use of both central processing units (CPUs) and GPUs requires a balanced heterogeneous computing strategy.

Purpose of the Study:

• To develop an efficient GPU implementation for evaluating small-scale ordinary differential equation (ODE) models, such as the Susceptible-Infected-Recovered (SIR) model.
• To propose an asynchronous particle swarm optimization (PSO) method leveraging GPU acceleration for parameter inference in epidemiological models.
• To infer model parameters for accurate data description through likelihood function optimization.

Main Methods:

• Developed an efficient GPU implementation for evaluating ODE-based epidemiological models.
• Implemented an asynchronous particle swarm optimization (PSO) algorithm utilizing GPU for parallel computation.
• Employed a heterogeneous computing approach, balancing workloads between CPU and GPU.

Main Results:

• Achieved speedups of 10 to 12 times compared to a 32-core CPU system by using a heterogeneous CPU-GPU approach.
• Demonstrated the effectiveness of GPU acceleration for parameter inference in infectious disease models.
• Showcased the performance gains on a standard machine with a high-end consumer GPU.

Conclusions:

• Utilizing GPUs for parameter inference offers substantial performance improvements on typical systems with high-end consumer GPUs.
• Future research should explore newer CPU and GPU architectures for further optimization.
• The developed method holds potential for application to more complex epidemiological scenarios.