Cross-validation-based sequential design for stochastic models.

Area of Science:

• Computational epidemiology
• Healthcare modeling
• Stochastic systems

Background:

• Complex numerical models are vital in healthcare and epidemiology.
• Stochastic behavior in models, while realistic, poses computational challenges.
• Efficiently designing training data is crucial for accurate model emulation.

Purpose of the Study:

• To develop a novel sequential design method for stochastic models.
• To address the challenges of training data design in high-dimensional input spaces.
• To provide a scalable and efficient alternative for emulating stochastic models.

Main Methods:

• Development of a new sequential design algorithm tailored for stochastic models.
• Adaptation of an expected squared leave-one-out error criterion for balancing exploration and replication.
• Application and comparison with existing sequential design methods.

Main Results:

• The proposed method demonstrates scalability in high-dimensional input spaces.
• Effective performance in noisy environments and stochastic settings.
• Successful application to agent-based and COVID-19 models.

Conclusions:

• The new sequential design method offers a scalable and efficient solution for stochastic models.
• This approach enhances the accuracy of emulators in healthcare and epidemiological applications.
• It provides a valuable tool for uncertainty quantification in complex systems.