DEpiABS: Differentiable Epidemic Agent-Based Simulator

2026-02-13 · Source: cs.MA updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Computational Epidemiology · Depth: Expert, extended

Summary

DEpiABS is a novel, scalable, and differentiable agent-based model (DABM) designed to overcome the limitations of existing epidemic simulation tools, particularly in capturing complex dynamics while maintaining computational efficiency and interpretability. It models individual-level heterogeneity, viral mutation, and reinfection dynamics, and is fully differentiable, enabling fast simulation and gradient-based parameter calibration. A key innovation is a z-score-based scaling method that maps small-scale simulations to real-world population sizes with minimal loss in output granularity, significantly reducing computational burden for large populations. Validated through sensitivity analysis and calibration against COVID-19 and flu data from ten regions, DEpiABS reduced average normal deviation in forecasting from 0.97 to 0.92 for COVID-19 mortality and from 0.41 to 0.32 for influenza-like-illness data. This model achieves superior forecasting ability and data efficiency without relying on auxiliary data, making it a reliable and generalizable framework for future epidemic response modeling.

Key takeaway

For research scientists developing epidemic models, DEpiABS demonstrates that a structure-centric, differentiable agent-based model can achieve high fidelity and interpretability without sacrificing computational efficiency or relying on extensive auxiliary data. You should consider adopting differentiable approximations and z-score-based output scaling to enhance model scalability and calibration, especially when working with fine-grained individual-level dynamics for large populations.

Key insights

DEpiABS is a differentiable agent-based model that balances mechanistic detail, computational efficiency, and interpretability for epidemic simulation.

Principles

• Structure-centric design prioritizes feature granularity over data volume.
• End-to-end differentiability enables gradient-based parameter calibration.
• Output scaling can decouple simulation cost from population size.

Method

DEpiABS employs differentiable approximation to transform discrete ABM processes into a continuous computational graph, supporting GPU acceleration. It uses a z-score-based scaling method to align small-scale model outputs with empirical data's statistical moments.

In practice

• Use z-score scaling for efficient large-population epidemic forecasting.
• Implement selective relaxation for non-differentiable model components.
• Tensorize categorical variables and encounters for DABM efficiency.

Topics

• Differentiable Agent-Based Models
• Epidemic Modeling
• Parameter Calibration
• Z-score Scaling
• Simulation Fidelity

Best for: Research Scientist, AI Researcher, AI Scientist, Machine Learning Engineer

Related on AIssential

• The Origins of Stochasticity: Comprehensive Investigations on Uncertainty Quantification for Large Language Models — A new taxonomy and evaluation framework reveal that consensus-based methods excel in quantifying LLM uncertainty, which decreases with model scale.
• The Degeneracy Distillery — The Degeneracy Distillery automatically resolves parameter degeneracies in models, reducing simulation costs and enhancing physical insight.
• Multi-Agent Systems in Emergency Departments: Validation Study on a ED Digital Twin — A hybrid DES-ABM-MAS framework effectively simulates ED dynamics for resource optimization and strategy exploration.
• A Differentiable Bayesian Relaxation for Latent Partial-Order Inference — A differentiable Bayesian relaxation enables scalable inference of latent partial orders from linear data by smoothing discrete constraints.
• ChemGraph-XANES: An Agentic Framework for XANES Simulation and Analysis — ChemGraph-XANES automates XANES simulation and analysis using LLM agents, unifying complex workflows for high-throughput material science.
• Neetyabhas: A Framework for Uncertainty-Aware Public Policy Optimization in Rational Agent-Based Models — Public policy optimization benefits from integrating individual behaviors and real-world uncertainties into agent-based models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.MA updates on arXiv.org.