Duet sampling generalizes Metropolis-Hastings for efficient DAG sampling
October 23, 2026
The paper introduces duet sampling, a Metropolis-Hastings generalization that uses two proposal distributions per iteration to improve mixing in Bayesian network structure learning. This method addresses high-dimensional challenges in approximating posterior distributions for directed acyclic graphs.
HOW THIS AFFECTS YOU
●
researcherYou can use these new moves to more efficiently approximate posterior distributions in large-scale Bayesian network learning.