Transport Monte Carlo: High-Accuracy Posterior Approximation via Random Transport

A methodological paper by Leo L. Duan from the University of Florida proposes a new algorithm for Bayesian posterior estimation. The method uses optimization to solve for a random transport plan between a posterior distribution and a simple uniform distribution. It is described as producing independent random samples with high approximation accuracy and is compared favorably to Markov chain Monte Carlo, variational Bayes, and normalizing flows.

Use Cases

• Sampling from multi-modal posterior distributions based on the described method
• Estimating sparse signals in high-dimensional Bayesian models as mentioned in the description
• Applying soft-thresholding to graphs with a prior on node degrees using the proposed algorithm

Strengths

• Method is compared to state-of-the-art alternatives like Markov chain Monte Carlo and variational Bayes
• The approach is illustrated via several challenging applications mentioned in the description

Limitations

• The description is methodological; the availability and format of any accompanying example data are unknown
• Row count and column-level documentation are absent, limiting suitability assessment

Provenance

Source

University of Florida

Collection Method

Proposed algorithmic research published via paperswithcode