guest ::
| Home > Publications database > A Comparative Analysis of Classical and Approximate Bayesian Inference Techniques for different Multiparameter Compartmental Epidemiological Models for COVID-19 Pandemic |
| Home > Publications database > A Comparative Analysis of Classical and Approximate Bayesian Inference Techniques for different Multiparameter Compartmental Epidemiological Models for COVID-19 Pandemic |
| Poster (Other) | FZJ-2026-00930 |
; ; ; ; ;
2025
Abstract: Compartmental epidemiological models have been utilized to understand the spread of infectious diseases in populations since [1]. These models divide the population into multiple compartments, such as susceptible, infected, recovered etc., providing insights into disease dynamics. Estimating the model parameters, such as transmission and recovery rates, is crucial for predicting the spread of the disease and informing public health agents for mitigation strategies. Bayesian inference is widely used to tackle this task, as it provides distributional estimates of model parameters and, at the same time, allows to update the model by incorporating new data into it. However, traditional Bayesian approaches often suffer from computationally intensive processes and limited parallelization capabilities. Simulation-based inference (SBI) offers an alternative approach using Artificial Intelligence (AI) for parameter estimation in complex dynamic models, [2]. SBI facilitates a more straightforward workload distribution across high-performance computing clusters, further reducing the runtime. Unlike traditional methods, SBI does not rely on explicit analytical likelihood functions. Instead, it employs computational simulations to generate synthetic datasets that mimic the observed data. This becomes particularly valuable when data availability is limited, as in the case of sparse data during the COVID-19 pandemic when we have only one realisation of the pandemic as a datapoint available.In our work we conduct a comprehensive comparison of the Markov Chain Monte Carlo and SBI methods applied to the same compartmental models, both allowing for chage-points in the parameters and not, and the same COVID-19 data from different German counties. Our evaluation considers key factors such as scalability, sensitivity to prior distributions, running time, inference robustness, and ease of implementation. We present our findings and discuss the scenarios where each inference method is more suitable. computational biology, Probabilistic Methods, Scalable AI[1] William Ogilvy Kermack et al. “A contribution to the mathematical theory of epidemics”. In: Proceedings of the Royal Society of London. Series A, Containing Papers of a Mathematical and Physical Character 115.772 (Aug. 1927), pp. 700–721.[2] Kyle Cranmer et al.. “The frontier of simulation based inference”. In: Proceedings of the National Academy of Sciences 117.48 (2020), pp. 30055–30062.
|
The record appears in these collections: |