Idea
Surrogate modeling platform accelerating epidemic simulations for real-time hospital planning and risk-aware decision-making.
Research Paper
Core Innovation
This paper introduces a surrogate modeling approach combining mechanistic SEIR-family ODEs with neural-parameterized contact rates learned from exascale agent-based model trajectories. It stabilizes training via multiple shooting and prediction-error methods, enforces epidemiological constraints, and achieves calibrated uncertainty quantification, enabling fast, interpretable epidemic forecasts.
Why It Matters
Hospitals and public health agencies require fast, reliable epidemic forecasts to manage resources and interventions effectively. Traditional agent-based models are too slow for daily planning, limiting responsiveness. This solution enables rapid, calibrated scenario analysis on commodity hardware, improving operational agility and supporting threshold-based decisions like ICU capacity management.
Market Size (TAM)
$2–10B TAM for epidemic modeling and forecasting software; $500M–$1B SAM from hospitals, public health agencies, and health tech firms. Driven by increasing demand for real-time epidemic response and scalable simulation tools.
Potential Customers & Pain Points
- Hospitals – Slow epidemic forecasts hinder resource planning
- Public health agencies – Need calibrated interpretable models for policy decisions
- Health tech companies – Require scalable epidemic simulation tools
- Government emergency planners – Demand rapid scenario analysis under uncertainty.
Business Model
Subscription-based SaaS platform offering epidemic simulation and forecasting tools with tiered pricing for hospitals, public health agencies, and health tech companies. Additional revenue from custom scenario consulting and integration services.
Competitive Landscape
- Covasim
- EpiModel
- GLEAMviz
- HealthMap
Implementation Challenges
- Integration with existing hospital and public health IT systems
- Regulatory acceptance of surrogate model forecasts
- Data privacy and security concerns in epidemic data sharing
- User trust in neural-augmented mechanistic models
Validation Strategy
- Benchmark surrogate forecasts against established agent-based models in real-world epidemic scenarios
- Pilot deployments with hospital planning teams to assess operational impact
- Collaborate with public health agencies for policy decision support trials
- Collect user feedback to refine model calibration and interface usability
Research Paper Overview
ABM-UDE: Developing Surrogates for Epidemic Agent-Based Models via Scientific Machine Learning
Summary
This research develops fast, accurate surrogate models for complex epidemic agent-based models using Universal Differential Equations with neural-parameterized contact rates. The approach enables rapid, calibrated scenario forecasting on standard hardware, drastically reducing computation time while preserving mechanistic interpretability and uncertainty quantification.