On Sampling from the Gibbs Distribution with Random Maximum A-Posteriori Perturbations
This provides a new method for sampling and lower bounds in machine learning, particularly for complex energy landscapes.
The paper tackles efficient sampling from Gibbs distributions by using MAP inference with random perturbations, achieving strong performance in challenging high signal-high coupling regimes.
In this paper we describe how MAP inference can be used to sample efficiently from Gibbs distributions. Specifically, we provide means for drawing either approximate or unbiased samples from Gibbs' distributions by introducing low dimensional perturbations and solving the corresponding MAP assignments. Our approach also leads to new ways to derive lower bounds on partition functions. We demonstrate empirically that our method excels in the typical "high signal - high coupling" regime. The setting results in ragged energy landscapes that are challenging for alternative approaches to sampling and/or lower bounds.