A fast non-reversible sampler for Bayesian mixture models
This paper introduces a new, simple non-reversible sampling method for Bayesian mixture models that significantly outperforms classical samplers. The method works by extending the target distribution to an augmented space involving pairs of clusters and uses a specific Markov kernel to force movement in one direction. It achieves faster convergence, reducing time complexity from O(n^2) to O(n), while maintaining comparable asymptotic variance.