Bagged Filters for Inference on Interacting Stochastic Systems

Bagged filter methodology combines an ensemble of Monte Carlo filters to address the curse of dimensionality in nonlinear, non-Gaussian systems. The research by Edward L. Ionides of the University of Michigan demonstrates its applicability to coupled population dynamics models, such as infectious disease transmission between cities. The method uses spatiotemporally localized weights to select successful filters at each unit and time.

Use Cases

• Evaluating likelihoods for interacting systems based on the described bagged filter algorithm.
• Comparing inference performance against ensemble Kalman filters for coupled dynamic models.
• Modeling disease transmission between spatial units based on the epidemiological example provided.
• Testing filter algorithms that respect smoothness and conservation laws mentioned in the description.

Strengths

• Methodology is presented with theoretical conditions for beating the curse of dimensionality.
• Performance is demonstrated on a concrete coupled population dynamics model for infectious disease.
• The bagged filter is shown to respect smoothness and conservation laws that alternative methods can violate.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Edward L. Ionides, University of Michigan

Collection Method

Theoretical and methodological research paper.