ABM-UDE: Developing Surrogates for Epidemic Agent-Based Models via Scientific Machine Learning
arXiv:2602.21588v1 Announce Type: new Abstract: Agent-based epidemic models (ABMs) encode behavioral and policy heterogeneity but are too slow for nightly hospital planning. We develop county-ready surrogates that learn directly from exascale ABM trajectories using Universal Differential Equations (UDEs): mechanistic SEIR-family ODEs with a neural-parameterized contact rate $\kappa_\phi(u,t)$ (no additive residual). Our contributions are threefold: we adapt multiple shooting and an observer-based prediction-error method (PEM) to stabilize identification of neural-augmented epidemiological dynamics across intervention-driven regime shifts; we enforce positivity and mass conservation and show the learned contact-rate parameterization yields a well-posed vector field; and we quantify accuracy, calibration, and compute against ABM ensembles and UDE baselines. On a representative ExaEpi scenario, PEM-UDE reduces mean MSE by 77% relative to single-shooting UDE (3.00 vs. 13.14) and by 20% re
arXiv:2602.21588v1 Announce Type: new Abstract: Agent-based epidemic models (ABMs) encode behavioral and policy heterogeneity but are too slow for nightly hospital planning. We develop county-ready surrogates that learn directly from exascale ABM trajectories using Universal Differential Equations (UDEs): mechanistic SEIR-family ODEs with a neural-parameterized contact rate $\kappa_\phi(u,t)$ (no additive residual). Our contributions are threefold: we adapt multiple shooting and an observer-based prediction-error method (PEM) to stabilize identification of neural-augmented epidemiological dynamics across intervention-driven regime shifts; we enforce positivity and mass conservation and show the learned contact-rate parameterization yields a well-posed vector field; and we quantify accuracy, calibration, and compute against ABM ensembles and UDE baselines. On a representative ExaEpi scenario, PEM-UDE reduces mean MSE by 77% relative to single-shooting UDE (3.00 vs. 13.14) and by 20% relative to MS-UDE (3.75). Reliability improves in parallel: empirical coverage of ABM $10$-$90$% and $25$-$75$% bands rises from 0.68/0.43 (UDE) and 0.79/0.55 (MS-UDE) to 0.86/0.61 with PEM-UDE and 0.94/0.69 with MS+PEM-UDE, indicating calibrated uncertainty rather than overconfident fits. Inference runs in seconds on commodity CPUs (20-35 s per $\sim$90-day forecast), enabling nightly ''what-if'' sweeps on a laptop. Relative to a $\sim$100 CPU-hour ABM reference run, this yields $\sim10^{4}\times$ lower wall-clock per scenario. This closes the realism-cadence gap, supports threshold-aware decision-making (e.g., maintaining ICU occupancy $<75$%), preserves mechanistic interpretability, and enables calibrated, risk-aware scenario planning on standard institutional hardware. Beyond epidemics, the ABM$\to$UDE recipe provides a portable path to distill agent-based simulators into fast, trustworthy surrogates for other scientific domains.
Executive Summary
This article presents a novel approach to developing surrogates for epidemic agent-based models using Universal Differential Equations (UDEs) and scientific machine learning. The authors propose a method to learn directly from exascale ABM trajectories, adapting multiple shooting and an observer-based prediction-error method to stabilize identification of neural-augmented epidemiological dynamics. The results demonstrate improved accuracy, calibration, and compute efficiency, enabling nightly 'what-if' sweeps on commodity CPUs and supporting threshold-aware decision-making.
Key Points
- ▸ Development of county-ready surrogates for epidemic agent-based models
- ▸ Use of Universal Differential Equations (UDEs) with neural-parameterized contact rate
- ▸ Adaptation of multiple shooting and observer-based prediction-error method for stabilization
Merits
Improved Accuracy
The proposed method reduces mean MSE by 77% relative to single-shooting UDE and by 20% relative to MS-UDE.
Enhanced Calibration
The method improves empirical coverage of ABM bands, indicating calibrated uncertainty rather than overconfident fits.
Efficient Computation
Inference runs in seconds on commodity CPUs, enabling nightly 'what-if' sweeps and supporting threshold-aware decision-making.
Demerits
Limited Generalizability
The method is specifically designed for epidemic agent-based models and may not be directly applicable to other domains without modifications.
Dependence on Quality of ABM Trajectories
The accuracy of the surrogate model is dependent on the quality of the exascale ABM trajectories used for training.
Expert Commentary
The article presents a significant advancement in the development of surrogates for epidemic agent-based models. The use of Universal Differential Equations and scientific machine learning enables the creation of efficient and accurate models that can support decision-making in public health policy. The method's ability to adapt to different scenarios and its improved calibration and computational efficiency make it a valuable tool for epidemic modeling. However, further research is needed to explore the generalizability of the method to other domains and to address potential limitations, such as dependence on the quality of ABM trajectories.
Recommendations
- ✓ Further research should be conducted to explore the application of the proposed method to other domains, such as climate modeling or financial systems.
- ✓ The method should be evaluated using a wider range of scenarios and datasets to assess its robustness and generalizability.
