From Model-Based Screening to Data-Driven Surrogates: A Multi-Stage Workflow for Exploring Stochastic Agent-Based Models

Paul Saves; Matthieu Mastio; Nicolas Verstaevel; Benoit Gaudou

cs.LGApr 3, 2026

From Model-Based Screening to Data-Driven Surrogates: A Multi-Stage Workflow for Exploring Stochastic Agent-Based Models

Authors: Paul Saves, Matthieu Mastio, Nicolas Verstaevel, Benoit Gaudou

Organizations: IRIT, Université Toulouse Capitole, Toulouse, France

Abstract

Systematic exploration of Agent-Based Models (ABMs) is challenged by the curse of dimensionality and their inherent stochasticity. We present a multi-stage pipeline integrating the systematic design of experiments with machine learning surrogates. Using a predator-prey case study, our methodology proceeds in two steps. First, an automated model-based screening identifies dominant variables, assesses outcome variability, and segments the parameter space. Second, we train Machine Learning models to map the remaining nonlinear interaction effects. This approach automates the discovery of unstable regions where system outcomes are highly dependent on nonlinear interactions between many variables. Thus, this work provides modelers with a rigorous, hands-off framework for sensitivity analysis and policy testing, even when dealing with high-dimensional stochastic simulators.

Figures & tables

Panel of Figure 1 — Figure 1: Multi-stage exploration workflow.

Category	Parameter	Symbol	Range / Units
Environment	Grassland proportion on the map	$Gr$	$[10.0,100.0]\,(\%)$
	Proportion of hunting zone among grassland	$PH$	$[0.0,100.0]\,(\%)$
Plants	Maximum Plant Energy	$PM$	$[50.0,250.0]\,(E)$
	Plant Growth Rate	$PR$	$[5.0,25.0]\,(E/t)$
	Grass energy intake of eating bandicoots	$BF$	$[2.0,10.0]\,(E)$
Bandicoots	Energy Gain when eating	$BG$	$[1.0,20.0]\,(E)$

Table 1: Model parameters, symbols, and experimental ranges.

Parameter	ANOVA (%)	Sobol’ $S_{1}$	Sobol’ $S_{T}$
Band. Energy Gain (BG)	0.09	7.6	45.9
Prop. Hunting (PH)	3.00	4.7	38.7
Fox Energy Gain (FG)	2.86	2.2	28.6
Grassland % (Gr)	0.04	5.2	27.4
Fox Hunting (FH)	4.11	2.1	21.8
Plant Growth Rate (PR)	0.85	5.8	17.8

Table 2: Comparison of Sensitivity Metrics: ANOVA vs. Sobol’ indices (

S_{1},S_{T}

Figure 2: PDP/ICE and Uncertainties variations for the 6 most important features.

Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 3: Simulation results showing the relationship between variables X and Y.

Panel of Figure 4 — Figure 4: Global Regime Dynamics. Comparison of simulation outcome distributions between the initial broad exploration (V1) and the refined sampling strategy (V2). The dominance of the Extinction regime highlights the system’s structural vulnerability.

Figure 5: Model-Based Screening (Phase 1). A Decision Tree (CART) trained on the initial dataset. It identifies the primary anthropogenic tipping point: a deterministic shift toward extinction when the Proportion of Hunting Zones ( $PH$ ) exceeds $\approx 31\%$ .

Panel of Figure 6 — Figure 6: Global Sensitivity Analysis (Phase 2). (a) Comparison of First-Order and Total-Order indices reveals that variance is driven by interactions rather than single parameters. (b) The heatmap highlights the strong coupling between Bandicoot Energy Gain ( $BG$ ) and Grassland Availability ( $Gr$ ).

Figure 7: Nonlinear Response and Mechanisms. Partial Dependence Plots (black lines) overlaid with Individual Conditional Expectation curves (colored lines). The coloring by $BG$ reveals a “Metabolic Trap”: increasing resources ( $Gr$ ) only promotes coexistence if metabolic efficiency ( $BG$ ) is sufficiently high (red lines).

Figure 8: Phase Transition Zones. A map of the Total Euclidean Uncertainty ( $\sigma_{total}=\sqrt{\sigma_{aleatoric}^{2}+\sigma_{epistemic}^{2}}$ ) regarding the probability of coexistence. Peaks in uncertainty identify the precise location of tipping points where the ecosystem is structurally unstable.

Explore similar work

ABM-UDE: Developing Surrogates for Epidemic Agent-Based Models via Scientific Machine Learning
Feb 25, 2026Sharv Murgai, Utkarsh Utkarsh, Kyle C. Nguyen +3Neural Surrogate Modeling Epidemic Forecasting
Monta Vista High School · MIT CSAIL · Sandia National Laboratories
Towards Agentic Agent-based Models: Feasibility, Performance, and Statistical Model Checking
Jul 20, 2026Stefano Blando, Emanuele Guerrazzi, Riccardo Porcedda +3Agent-Based Modeling LLM Agents
Sant’Anna School of Advanced Studies Pisa, Italy · University of Pisa, Italy · CentraleSupélec, Université Paris-Saclay, France +2
Machine learning-assisted calibration of Agent-based Models: surrogate-based optimization with Genetic Algorithm and Particle Swarm Optimization
Sep 4, 2026Duguma Yeshitla Habtemariam, Jihwan LeeSurrogate-Assisted Optimization Agent-Based Modeling
Department of Industrial and Data Engineering Major in Industrial Data Science and Engineering Pukyong National University, Busan, Republic of Korea

Feb 25, 2026J/K move · Enter open · S save

ABM-UDE: Developing Surrogates for Epidemic Agent-Based Models via Scientific Machine Learning

Sharv Murgai, Utkarsh Utkarsh, Kyle C. Nguyen, Alan Edelman+2

Monta Vista High School · MIT CSAIL · Sandia National Laboratories

Neural Surrogate Modeling Epidemic Forecasting Agent-Based Modeling Uncertainty Calibration

Open paper AskPDF arXiv

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

κ_φ(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.

Abstract

Explore similar work

ABM-UDE: Developing Surrogates for Epidemic Agent-Based Models via Scientific Machine Learning

Towards Agentic Agent-based Models: Feasibility, Performance, and Statistical Model Checking

Machine learning-assisted calibration of Agent-based Models: surrogate-based optimization with Genetic Algorithm and Particle Swarm Optimization