Agent-based models (ABMs) are powerful methods to describe the spread of epidemics. An ABM treats each susceptible individual as an agent in a simulated world. The simulation algorithm of the model tracks the health status of each agent. ABMs can provide realistic dynamics for the epidemic at the individual level so that users can observe and predict the spread of epidemic over time and space.
ABMs allow users to change public health policies in the simulated world. Healthcare professionals and policymakers can then run what-if tests to evaluate strategies for containment and mitigation. ABMs have been used in modelling various pandemics by epidemiologists [Adam2020].
This project concerns agent-based simulation to evaluate the trustworthiness of epidemic simulation in a
computationally efficient way. It addresses three themes: trustworthiness evaluation, efficient computation, and case studies.
First, trustworthiness evaluation includes simulation and model-checking. The models in this study should allow calibration algorithms to search and tune model parameters to fit existing data in relevant patient records. Due to randomness in disease spreading situations, the simulation algorithm should generate different versions of future scenarios based on the same initial configuration. The distribution of such scenarios provides a measure of the trustworthiness of the simulation. Model-checking provides a way for hypothesis testing under uncertainty. In particular, as epidemic simulation is stochastic, Bayesian model-checking [Jha2009] and public-domain data on COVID-19 will be adopted to check trustworthiness-related hypotheses to complement simulation. Also, trustworthiness reports will be generated to provide a human-understandable interpretation of spreading dynamics and the error distribution of predictions.
Second, computational efficiency issues arise when the agent-based models are stochastic, and the population is large. The models should capture assumptions and rules on the interactions between agents and the environment. Since the actions and their outcomes are not deterministic, it is often necessary to simulate such models for a considerable number of times to model the distribution accurately and allow rare events to emerge. Concerning the computational efficiency issues, this study includes the design of accelerated software and hardware tools for calibration, simulation and model-checking on multi-cores, graphics processing units (GPUs) and field-programmable gate arrays (FPGAs). This can be built on our recent research on acceleration of epidemic simulation [Guo2020].
Third, case studies involve applying the proposed approach to current epidemic simulation, and evaluate their trustworthiness using publicly available data. Our research would help identify potential trustworthiness issues with realistic computational time and resources, and would explore novel ways of developing trusted epidemic simulation.
[Adam2020] D. Adam. Modelling the pandemic. Nature, vol. 580, pp. 316-318, 2020.
[Guo2020] C. Guo, W. Luk and S. Weston, Accelerating simulation for agent-based epidemic models using FPGAs, The ACS/IEEE Int. Conf. on Computer Systems and Applications, 2020.
[Jha2009] Jha, S.K. et al. “A Bayesian approach to model checking biological systems.” Int. Conf. on Computational
Methods in Sys. Bio., 2009.