Sparse Gaussian DAGs Represented as Sparse R-vines for Non-Gaussian Dependence

A methodological paper by Dominik Müller of the Technical University of Munich demonstrates a novel connection between directed acyclic graphs (DAGs) and vine copula models. The work presents a procedure for high-dimensional dependence modeling that leverages fast DAG model selection while accommodating non-Gaussian features like tail dependence and asymmetry. Numerical examples are conducted in hundreds of dimensions.

Use Cases

• Structure learning for high-dimensional systems based on the described connection between DAGs and vine copulas.
• Modeling non-Gaussian dependence features like tail dependence and asymmetry based on the flexibility of vine copulas mentioned in the description.
• Developing scalable statistical models for conditional independencies based on the described DAG framework.
• Benchmarking vine structure selection algorithms based on the numerical examples in hundreds of dimensions.

Strengths

• Demonstrates a novel theoretical connection between DAGs and truncated vine copulas.
• Includes numerical examples in hundreds of dimensions, suggesting scalability.
• Proposes a method that aims to combine the speed of sparse DAG estimation with the flexibility of vine copulas.

Limitations

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

Provenance

Source

Technical University of Munich

Collection Method

Theoretical research and numerical experiments, as described in the associated paper.