Reducing computational complexity in agent-based epidemiological model calibration: application of deep learning surrogates

Darwish A., Leonenko V.N.
 pdf (995K)

Acute respiratory infections are a major public health concern because they are the leading cause of illness and death in many countries. Therefore, there is great interest in developing models and methods capable of modeling the spread of these infections within communities, with the aim of controlling outbreaks and preventing their spread. Agent-based models (ABM) are one of the most important tools in epidemiological research for modeling epidemic dynamics in realistic populations, but they face significant challenges in terms of computational complexity in their operation and calibration of epidemiological data, as parameter estimation typically requires repeated simulations across large parameter spaces to determine plausible values for key epidemiological parameters. This paper addresses the problem of alleviating computational constraints in the inverse problem of calibrating an ABM model for simulating the spread of respiratory infections in Saint Petersburg. The paper proposes the application of machine learning surrogate to link epidemic trajectories to underlying epidemiological parameters, enabling them to quickly infer parameter estimates from observed epidemic data. This is done by formulating the task of calibrating ABMs against epidemiological data as a supervised learning problem, where sequences extracted from epidemiological trajectories are associated with underlying epidemiological parameters. The research was based on evaluating the performance of attention-based sequence modeling, probabilistic deep learning, and distributional regression for inferring parameter estimates from truncated sequences of epidemic trajectories. Experimental evaluations have demonstrated the effectiveness of this approach and its practical and straightforward application. The results also indicated the superiority of attention-based sequence modeling, as it showed more consistent performance across metrics and horizons in accurate parameter estimation and credible uncertainty quantification. Distributional regression modeling also showed good performance with specific strengths in point accuracy while probabilistic deep learning performed poorly, especially at longer input horizons.

Keywords: epidemiology, ABM simulations, machine learning (ML), inverse problems, сurse of dimensions
Citation in English: Darwish A., Leonenko V.N. Reducing computational complexity in agent-based epidemiological model calibration: application of deep learning surrogates // Computer Research and Modeling, 2026, vol. 18, no. 1, pp. 185-200
Citation in English: Darwish A., Leonenko V.N. Reducing computational complexity in agent-based epidemiological model calibration: application of deep learning surrogates // Computer Research and Modeling, 2026, vol. 18, no. 1, pp. 185-200
DOI: 10.20537/2076-7633-2026-18-1-185-200
URL: http://crm-en.ics.org.ru/journal/article/3695/
Creative Commons License This work is licensed under a Creative Commons Attribution-NoDerivs 3.0 Unported License.

Copyright © 2026 Darwish A., Leonenko V.N.

Indexed in Scopus

Full-text version of the journal is also available on the web site of the scientific electronic library eLIBRARY.RU

The journal is included in the Russian Science Citation Index

The journal is included in the RSCI

International Interdisciplinary Conference "Mathematics. Computing. Education"

Website Copyright © 2009–2026 Institute of Computer Science